GN Live: Hybrid Vega 56 Undervolting & Overclocking (Pt2)
GN Live: Hybrid Vega 56 Undervolting & Overclocking (Pt2)
2017-08-23
are we live I think we might be live we
didn't test this time we're just kind of
just kind of go on live so we are going
to be revisiting undervolt in an
overclocking with vega 56 today two
different things from last time this
time we have the hybrid modding and
we've also got the new drivers which I
think are called 18.1 so 18.1 should
theoretically be better for under both
in an overclocking from what we've seen
so far and I think that might have more
of an impact this time than our previous
stream where the drivers were still
pretty new and and weren't fully
functioning so I think we're good to go
and Chad is tiny yes we are we are alive
so I here's what we've got set up maybe
Andrew can show some of it first of all
we've got the Radeon Settings stuff I
can also use wot tool and stuff like
that but we're gonna focus on Radeon
settings because I mean they got a
driver push so I'd like to see if it
works properly fire strike is right here
we might use some other stuff as well we
have a phone old phone setup that's
being used as a tool for reading
wirelessly from our current clamp which
is today doubling as a volt or a digital
multimeter so right now we're reading
apps coming out of the 12 volt cables
going to the video card this is a
thermocouple reader and it's got two
readings on only one of them is really
relevant the top one so that top number
is going to be the we're getting drop
frame so let me see if I can fix that
let's fix the drop frames
okay we're going down to let's go to
change the stream encoding to very fast
really shouldn't be on the CPU side
let's go down to 8 megabits 8000 apply
okay all right let me know if that fixed
it
CPU 20% ok CPU should be handling it all
right how's it going how's how's it
looking now
so we oh by the way I just got a notice
today that's a Google Fiber will be
available now to this location so if
this is a problem today it's not gonna
be a problem the future we're not using
we're using a 7900 x2 stream actually so
the CPUs really should be fine should
just be the Internet at this point I
think we're good though is it OBS is
Tommy we're good at this point we're not
dropping frames anywhere we drop 27% of
they might start ok
so let's walk through this again still
stuttering so I'm saying still drop
frame can someone I reliable tell me if
we're dropping frames you're not still
getting some chop okay how about now
okay
we're not uploading anything I don't
think so okay what cpu are you streaming
7900 X really is it still dropping cuz
OBS no okay let's see 27 48 2809
alright let's you know what Time Warner
let's get on to 6 megabits per second
apply ok looks good looks good better
good now okay alright okay thank you
alright I'll keep an eye on that we had
to drop down to 6 megabits per second if
things get better I'll try and do better
but that's what we're gonna do for today
so alright let's let's go back over what
I was going over so here we have Radeon
settings we have 3d Markel you said as a
validation tool will use other stuff as
well big note here part of overclock in
an under volt in with the Vega card
right now includes validation with
software so you I don't know I don't
know what these drivers have done but
previously you couldn't really trust the
frequency readout you had to just go
based off of scores and see if it
improved here's the phone that we're
using to readout
currently the current the amperage going
down to the 12 volt PCIe cables to the
video card and then going back over this
now that we're hopefully not stuttering
anymore so going back over this the top
number is the one of the hotspots so
it's a it's a mosfet readout for one of
the hotter MOSFETs on the card
so we want that to be below 100 °c
which should be no problem at all bottom
numbers is just a it's another
thermocouple that's not really placed
properly so we're gonna ignore that okay
so that's most the setup over to the
other side we have the system I don't
know how visible it'll be with the light
we have over there but it is currently
set up with a thermal take CL C so we've
got the flow 360 millimetre CLC and that
is set up to the GPU and then we have
some little heat
sticking off of the MOSFETs which are
just cooled by a fan on top of it so
there's a fan on top of the just sitting
straight on top of the PCB blowing down
at the fats and we've got all these fans
turned down to like now 45% right now
45 to 75 percent because of the stream
so we're gonna be a bit hotter than I
biked but I'd rather not blow out the
audio with fan noise so so I start
working with so that's the setup let's
uh first of all this is brought to you
by Thermaltake they do make the flow 360
millimeter CLC that's on the card which
we used in our hybrid mod and that it
should be on Amazon by now so if you're
interested in it check it out there
we've replaced the fans so I've noticed
some people don't like that I used to
knock to a fan and to Corsair fans which
by the way have different color LEDs
ones a light blue and ones white so I'm
sorry if that bothers you but that's
what we went with ran out of the corsair
ones and through an octa one on there
and then we've got the Thermaltake flow
360 rad on there with the pump which is
a 4.5 gen pump one of the faster pumps
right now which is what i was going for
so we we've we've did a $10 donation and
said take this and buy 10 DS i will do
that i'm not sure i we're gonna be in
Seattle 4 packs in the next two weeks I
think next week so there are good
chicken tender places there we will take
that $10 and use it appropriately thank
you ok so let's start off with just
seeing if this stuff works because last
time it didn't really worry oh you know
what we have something new let me see if
this works this is working let's see if
that works so we're trying to this is
actually a crazy setup too but hopefully
that's outputting the screen and we can
get a direct feed so reset the custom
mode and that's what we need to use I
can hit a power target up to 242 percent
I doesn't really fully apply it we're
gonna stick to 50 right now because
that's what people actually do in in
real life so I can't type in a number
unfortunately let's go to 50 apply
- wah - later if we need to know thing
with memory right now let's expand this
out let's get a some kind of fire strike
thing going now let's just start with a
frozen fire strike output and then I'll
do something more later but for now no
we need to be a windowed but for now
we'll start with froze it froze in the
output window mode yes run okay all
right it's work and sweet okay cool so
while this is starting up the way we
ended up doing the picture and picture
is basically I've got a capture card
hooked up pointed out that machine and
it's supposed to do a direct feed into
this machine we're not able to get OBS
to detect it but I can get a window
viewer that comes with the capture card
drivers to see the output and then we
open that on that machine with streaming
one and point OBS as how window capture
at it so it's a it's kind of a hack but
it's working theoretically okay so get
all this open fire strikes kind of
freaking out and let's see how this
starts how this looks for the new
drivers okay we're just gonna pause at
the same place we did last time right
here and see our current readout is
sixteen point four amps oh yeah I guess
we're I guess you can't even see that
with this setup so we're at sixteen
point four amps right now and this is
just with a 50% power target and with
everything else
oh no it's not stuck actually I had
previously done some under bolting on
here to see if it worked
so let's reset that go to custom mode
and power limit 50% okay so before I
apply this here's what we're at we're at
fifteen point nine amps and that's
before applying anything so let's just
go ahead and apply these changes
apply so we should have much higher now
yeah so now we're at twenty three point
six amps going out of there and that's
uh what is that where's a calculator
twenty three hours twenty twenty three
point four two six let's see that so
that's roughly two hundred eighty nine
watts right now and if we I'm outputting
this screen right now so you don't need
to capture it at this time so we're at
289 watts and if we do some undervolt in
let's see what happens let's well first
of all see if it even works this time
fifteen ninety to apply so twenty three
points six is what we're reading and you
know what let's just take a voltage
reading off the back of the card - so
I'm that's gonna require me moving
around to the other side of the table to
check that so how are we measuring
current with your phone we are using a
current clamp actually we can show you
that in a second because I'm about to
show you the volt readout I don't know
if Andrew it might be better to show
this once I do a reading but we'll see
and maybe I'll try and get it to read it
that might fall over I'll try and get it
to readout on the clamp and make it
visible excuse me
okay so let's show this for a second I
think we're outputting that yeah all
right excuse the mess this is because of
the streaming setup can I switch this
over to volts DC this is the current
clamp oh yeah didn't want to point let's
set that how's that look that's that's
best we can do right now current clamp
is round the 12-volt cables I've got
ground in one of the molex and now I'm
gonna take a volt readout we are getting
one point three one come on one point
three okay so that's the V Corps readout
one point three is what I'm seeing let
me switch this back to current
so what point three volts it is working
the software is relatively accurate in
that regard now we need to do some
actual undervolt in let me check the
chat here all right guys planning on
doing Vega testing with ek block blocks
etc seeing that you got a custom loop on
stream not sure right now now I'm not
sure about Vega will do some stuff with
ek some more anyway what's your one line
review of thread Ripper that is really
difficult to answer could you not bring
the camera round no it's on a tripod
I wrote a quest and a donation to super
chat what's your one hyper-v a threat
Ripper ok wrote an article did a video
on this I don't know if I can compress
it to one line so I'll compress it to a
few seconds thread Ripper was highly
performance we'll use that word for for
production tasks it did well in Premiere
and blender there are a couple others
well I think poverty it ok and also
really good for streaming wild gaming
for the most part like there were some
problems where the 7900 ax was still a
good bit better with the gaming portion
of it but thread Ripper did really well
keeping up with the actual streaming
portion so there's plenty of room there
to decide what you need the review does
a good job of breaking it down I'm not
gonna try and do it today but yeah it's
it's good at some things it's not great
at other things but that's kind of the
idea of hardware's the no piece of
hardware I can do everything really well
thread refer does some stuff
exceptionally well it does some other
stuff not as well as its competition
that's what it boils down to
ok so everything's working we got volt
readouts on the back of card let's drop
this down to eleven hundred millivolts
on each option I might switch to watt
tool because it's easier and what's our
current readout so we just dropped from
twenty four point six to nineteen point
eight and if I just do a quick nineteen
point eight times twelve point three so
out to forty three watts now and we were
at 280 something I think so that's what
we're looking like over here
yes we do have a screen capture today
yeah so there's what we have for current
readout that's at 1,100 millivolts at
one point one I'll do another volt check
in a minute once we get a little bit
lower we were at one point three
originally so this times 1592 let's get
gpu-z open this may still be inaccurate
I'm not sure if the frequency is reading
properly now with the new drivers we're
on 18 point one who it for anyone you
missed that this reads 1590 that says
1592 so the rounding is corrected for
we're not gonna do anything to clock yet
go down to 1080 that's not enough let's
let's do like 1070 1070 is it subliminal
advertising no 10 70 millivolts so this
is putting us at oh we do get a drop
there eighteen point seven eight god
this scale is awful on this current
readout so 18 what is that a drop
eighteen point six times twelve point
three to twenty eight okay it wasn't a
little bit of a job
let's try 1040 I need to like actually
validate this with something running
we're just doing this for now to get a
ballpark as it is working better than
previously this still shows 1590 this is
still running 10 40 millivolts we're
reading seventeen point two amps which
puts us at 211 and we started as a
reminder roughly like twenty four point
five or something now that's a little
bit high I guess we were we're at 289
originally I think so we are dropping
and we're holding clock rate allegedly
it's good at some point that should
either down clock or crash I'm not sure
how the new drivers behave 10:20 we're
not overclocking memory yet okay it's
working this is actually kind of this is
pretty exciting so like last time this
did not really work that well I last
time basically I'd type in numbers and
it try to apply and then just kind of
not do it like it it applied for a
little bit for fire strike or something
the minute I closed it open something
we go back up to full power draw full
voltage at 1.2 or 1.3 through the back
so yeah it looks like we're actually
doing better this time like the drivers
are actually responding frequency is
reportedly holding so we're not down
clocking we'll have to see if it's does
that clock reporting bug though how are
we doing on chat here
so for chat any questions now just the
last one was the one line threader
forgive me a question okay all right
let's see what else we can do with this
so let me show this yeah I'll try to do
that let's see that I just do what if I
put it right there what do you want me
to prop it up on something so we've got
a 16 app reading which is actually about
what it says when we show when we set it
to complete stock without the off the
overpower setup it's going to prop it up
on like this just higher or do you want
it propped okay let's try and do this
without a thermocouple getting sucked
into the fan ideally what's her
temperature we're at 45 C on the hotter
MOSFET
what's ambient ambient is 26 C okay all
right frames drops Ryan are we actually
dropping frames okay so let's look what
can we do further than this
we're currently at fifteen nineteen ten
twenty let's I mean I would expect it
would start not working at some point
all right down clocking yet not reported
down clocking I think we're gonna need
to validate this with a benchmark though
okay so we're apparently still working
we're at fifteen point five amps for
sure I mean that doesn't lie I don't
know that the software does but well
it's easy enough to find out
okay so we're we're at nine eighty
millivolts where allegedly holding
fifteen ninety over here let me do
another readout of the jet stream the
phone screens of the PC we've got enough
problems setting up the stream that's
another thing to break alright
I check over here
okay so we have what are you getting a
voltage readout on the phone right now
okay one point now that is down we were
at one point three so that it looks like
it's working we were at one point three
versus one point two in software and
tripping on lights over here and now
we're at nine eighty millivolts and one
point something lower one point now so
it looks functional let's stop there
let's stop there and run a fire strike
test and just validate with performance
make sure it's not lying to us before we
get too excited a fire strike okay and
I'll look at chat while this runs no do
not include the demo and we only need
graphics run okay all right that's
running let me check chat so chat
options looks good okay so people are
saying that drop frames are done I guess
that's what I want to hear a new
question and super chat Daniel so Daniel
five dollars thank you Daniel says with
V chord down now can you increase the
power to the memory for better HP I'm
overclocked so we've got a couple mods
on this card but basically the the goal
we've got two goals today one goal is to
I'll just leave this set up like this I
guess so one goal today is to just do
under bolting first so we're gonna see
how low and get the voltage while
keeping a stable clock that actually
either performs the same as previously
or ideally a little bit better I'm
shooting for like an extra twenty five
to fifty megahertz over stock because
I'd like to see the performance
improvement that often accompanies a
lower voltage and ok there we go so it
doesn't like it when there's a change so
that's the first goal just straight
under Bolton next goal is we'll look at
some overclock and toward the end of
this and then I'll roll that into our
hybrid part to coverage or part three
coverage at this point with the results
and theoretically to answer your
question
we should be I can't manually control
where the power goes I don't have that
level of control but I can as we reduce
B core we should get some extra power
budget now does that matter not really
because if I start overclocking the
thing like legitimately overclocking it
it's just not gonna be stable on the
lower V core we're still gonna want to
go back up to 1.2 so I think what we're
gonna see is two different types of
results we're gonna see lower power
consumption with the same or better
performance going for better because
otherwise it's not that fun and then the
other option is the switch to current
rates no they want a shot of the current
and then the other option is
overclocking with an actually high clock
and and then who cares about the power
consumption I'm gonna go for as high
power as we can hit with this modded
registry
so here's what we were doing previously
with the old drivers what are we doing
today so we have a 50 nut here's a stock
so stock here's what we had previously
so we had stock at eighteen 892 score
and fire strike stock plus 50% power was
twenty seven six nine we're at twenty
three three eight there's some variance
here so we can just call them pretty
much the same so we're at the same with
a 50% power offset and with the clock
untouched at fifteen ninety we're
getting basically the same output today
with a lower power consumption so that's
good we need to switch back to current
on the readout let me just do that
so let's see if we're actually getting
that lower power draw that we wanted
knocking over boxes now all right so
let's what's our current readout look
like for this setup okay
scores looking good though scores
definitely the same as the 50% offset
with lower power in theory we're about
to find out okay
any other questions here to go over no
signals cuz but yeah yeah yeah so when
you see no signal it's just because the
benchmarks changing resolutions okay so
our current readout says 16 amps so that
is in fact lower we were at 20 let's
just call it 23 to 25 give myself some
room there it's about 20 23 23 24 and
that's with a 50% power offset and stock
clocks now we're at 15 amps which comes
out to some quick math here let's call
it fifteen point five fifteen point
seven is maybe more fair twelve point
three 193 Watts previously let's do a 24
previously we're at 295 so we're like
100 watts lower I need to see if the
scores the same but it looks like it's
about the same so I don't don't run to
the hills with those numbers yet but
it's looking good so far
what's the point of this video
ninety-five percent people don't care to
overclock nor care about it
well I it's fun and it teaches us about
the hardware and we can learn about the
power consumption and we can learn about
how you can fine tune the power
consumption you know people may not care
to overclock but a lot of people sure
screamed and yelled about the power
consumption on the card and so to those
people I would say if you wanted Vega
but you are upset about the power
consumption here's your solution and
we're not sure if it works fully yet but
it's looking pretty good
these results are not guaranteed on all
of your cards but it is looking pretty
good so far
well there'd be a separate video or is
this live stream only so this the main
stuff will be in the live stream but I'm
going to pull so this is just
preliminary like learning I'm gonna pull
all of this data and turn it into a
normal article on video after I'm gonna
need to do some more work on this after
the stream ends because streaming and
actually doing real work is a little
difficult so we are currently at twenty
to thirty seven so we're a little bit
lower stock plus fifty percent
20850 percent stock frequency lower
voltage we are like a percent or less
below that so we're looking pretty good
now we need to see what kind of clocks
we can hit without this thing crashing
or forcing a clock drops so let's just
go back to this setup windowed mode and
and go that route next is that GPU or
total system power that is it on average
yes that is the PCIe cables going to the
video card so it says point nine right
now that's point nine amps and it's
going to go up as this software launches
that's going to the video card through
the PCIe cables it's not measuring
what's going through the PCIe slot but
on these AMD cards that tends to be very
little for Vega that is as opposed to
previously so that's pretty close to the
power you're actually drawing the PCIe
slot is more or less responsible for
things like the the fan power and we
don't even have that fan connected
anymore so okay whereas our where's
gpu-z where'd that go hide
and interfering with the benchmark I'm
sorry okay so we are at we're showing
1590 okay let's just freeze this and
let's pull this over here and then we're
currently on the we're showing the
monitor and that right now so you've got
a small screen for that so let's go up
to sixteen twelve which I'm just doing
to the rounding error anyway so apply
that GPZ thinks at sixteen twelve as
many of you know from our content it's
hard to know if that's actually the
frequency until we validate it let's
just see if we can break it let's just
what was the match last time 1980 let's
just try that first so this should not
work it looks like it failed which would
be fantastic like that would actually
make me happy cuz that means the
frequency wouldn't be misreporting like
it was earlier okay this looks good so
we need to restart now which is what I
wanted restart in okay so quick update
typing in nineteen eighty megahertz does
actually break it which if you missed
the previous content that did not break
it it just told you it was nineteen
eighty megahertz and and that was wrong
but now it appears to actually break
properly so the frequency might be
reporting properly now which if it is
fantastic job by the a.m. the driver
team for fixing that quickly so that was
kind of a it was very annoying just cuz
we didn't know what was actually going
on tell it with fire we're not going to
kill it with fire oh I think we're fully
capable of generating fire but I don't I
don't think we want to kill it with fire
okay
all right so let's let's get this going
again what do we need here gpu-z I need
wat man or wat tool let's stick with wat
man because it seems like it's actually
working now and I'd like to use what AM
the includes for now what tool is pretty
nice but it has bugs of its own as well
although I don't know so that stuff
looks like it's fixed now okay so
clearly the frequency fan is at least
partially resolved
that's reporting the correct frequency
now that makes me happy
windowed mode demo included yes Ron okay
any questions while this is launching
where is the overlaid multimeter we're
pretty close we're getting there so
right now we've got on the phone I set
it up so that it's the current clamp can
also be a digital multimeter so I've got
it kind of set up we're getting there
where can we get GN wallpapers so yeah
that new wallpaper I don't think we've
shown that before publicly it's like the
GN dark logo I'll see what see what I
can do we can probably do like a
wallpaper pack and distribute it it's a
I think it was made in blender right so
it's made in blender we can just render
it out whatever size we want well you
make the same type of stream with Vega
64 I'm not sure we will work with Vega
64 I don't know if we'll do the same
type of stream but we'll we'll work with
it because I just got one in and I have
two coming in from a reader so that'll
be pretty cool okay so let me switch
yeah we'll keep under Bowl things don't
worry about it I just want to before I
go too far with under bolting I want to
make sure it's actually working and make
sure I understand part of the software
so like I said previously this is not me
coming in here with a game plan knowing
how the software works now this is me
coming in here knowing how it worked
previously and hoping that works better
now and I'm learning on stream with you
also this is my process of learning of
things are actually functional ok so
let's go to back down to nine eighty
millivolts that's where we were before
did it work
oh we need to offset the power to 50%
power offset
apply okay is it worth did it go down
let's go back up to 1200 so we're at
fourteen point four I'm applying 1200
now okay 24 what else happened when I
did that okay cool
so here you go this is me showing you
that it seems to be working twenty three
point four amps right we're at 50% power
offset we're at 1200 millivolts one
point two volts let's go down 980 which
appears to be at least partially stable
it might be unstable when we launch a
game drop 10 amps right there which is
like more than 100 watts so that's
pretty cool what do we need to do next
let's see if we get the clock any higher
with we'll find out if this is reliable
or not I'll to do a validation through
software like I said 1650 it applied
does it hold her does it crash and
what's our amperage we're still have the
same current 15.8 well having not the
same it's it's a bit ok there that was a
big jump that's a huge jump so it looks
like it's trying to compensate and stay
alive because we just jumped back up to
22 so we're clearly not let's do it let
me do a multi meter reading on this just
because I'm curious
okay so this says 19 amps we were at 14
something what are we at now is that
Cheyenne volts on the phone 1.2 volts
that is not 0.98 that's not 1 point out
how either so so it's not working
anymore we've we've lost the under whole
thing which I think is because I went
too low and it's trying to keep the
thing alive so lesson learned here if
you think it worked and you don't really
have a direct feed of the current draw I
need to switch back then it might not
actually be staying where it said it
wasn't software so it doesn't look like
there's just a hard crash memory speed
grab is blocked yeah it's not changed
either though I mean this is I haven't
touched the memory speed it's just stock
so okay so that's clearly not working
so let's still have 50 offset let's
let's take it back up let's go to 10 50
does that is not stable since it's over
volting itself back to stock to try and
hold let's get a baseline 1200
I'll check questions in a second if
anyone has any 1200
there's our baseline 24 that's 1.3 if
you check the back the card okay so how
is 1080 that seems fairly conservative
10 80 millivolts with an overclock
applied mind you it's not really a
change I mean there it's kind of hard to
say if that's gonna hold or not so I'm
thinking we might have pushed the clock
too far well I mean yeah it is a bit
lower I guess but
okay so let's bring the clock down let's
let's mmm we were at 1592 and that was
stable and it was trying to get score so
the same score is previously without
Thunderbolts so let's bring it up to how
it's like a modest 16 20 that's a 30
megahertz overclock 16 20 let's bring
this up as well
1,100 1,100 apply ok close this and
watch that
custom graphics only windowed do not
include the demo run okay all right so
let's see how that goes
I'm gonna check some questions while
we're here I'll move this move the
windows around a second three cards
coming in yes we will be trying your
suggestion 19 20 megahertz 10 80
millivolts best combination yeah
don't know that'll work we make the same
stream now tread that one rad tech
gamers Nexus where can we get Gianna
wallpaper so yeah like I was saying I'll
see if we can distribute a pack at least
two patreon and maybe push something
just to Twitter or something and and
then we'll go from there on theirs yeah
it's gonna mess things up a bit but
whatever
okay so 1622 are holding allegedly next
question what are the GP attempts with
your CLC hybrid mod so let's find out 32
celsius reported temperature now here's
the thing the reported temperature seems
to be a little inaccurate with this
software I don't know if it's from the I
don't know if it's from a physics turn
back on oops I don't know if it's from
the sensor just having a normal range so
these sensors have like a plus or minus
two Celsius if it's a K type which is
not it's by a thermistor or something
else it's got some variance or
inaccuracy but basically even though
watt man and afterburner will show a
temperature when it's idle it'll show a
temperature of like 19 C which is below
ambient so it's impossible without
exotic cooling ambient was 22 when I was
seeing 19 C GPU temperature so it's
definitely not really correct but this
is reporting 32 right now we do have a
360 rad on there and it's got three fans
at half-hour p.m. like 50% rpm so I do
believe that it's a good temperature but
I just don't know level of accuracy yes
yeah the 50% is throwing us off
we'll work on it so again a bug with
reported time yes there is there I don't
know if it's a bug it might just be that
the sensor in there is inaccurate so if
that's the case it should be inaccurate
by the same amount just all across the
scale if it's like an NTC or a thermal
couple like a K type normally is but I
don't know I don't I don't really know
what we should be expecting for a
temperature offset at idle after all the
adjustments how much wattage does the
car need we haven't found out yet it's
still drawing more than I want it to
it's what is it pushing like 20 still
yeah it's still put in 21 which is too
high because we're offset 50% you know
let's get rid of this let's see if we
can go back to 1590 I might start using
watt tool just to make sure fifteen
ninety at 1100 let's apply that we're
gonna mess up our score that's fine
it's gonna run physics now now okay
all right well that's not really telling
us anything because our power
consumption still kind of high so back
to the drawing board on this demo what
do we need to do here we've got I don't
want to touch HBM 1592 at 1100 okay
let's see if it just holds a lower value
this time
any more questions in the meantime can
you check temperature with thermal
probes and don't trust the sensors the
problem with that is I'd have to get a
thermocouple in there and those are
going to be tall enough that it's gonna
mess up my cooling we're not gonna get
full coverage everywhere unless I just
dump a truckload of thermal paste on it
and then the transfers just gonna be bad
so I could attach a thermocouple to it
but it's just I mean even with the 1/100
inch thick ones that's it's still gonna
be too much especially when HBM is as
sensitive as it is I don't want a risk
not cooling the HBM because I've stuck a
probe in there on the GPU which would
increase the height elsewhere so I don't
have a good way to get a direct reading
not one that's any more accurate than
this is anyway okay so 19 amps looks
okay
see here's the thing as soon as I do
anything with frequency we're gonna lose
the power power advantage its grand back
ten seventy and just run a test the ten
seventy stock frequency kind of tempted
to drop the offset the power offset but
I really don't want to because that was
the whole point of this originally
you're still in nineteen okay is this
even working at this point and forty
apply I'm not convinced that's working
right now
where is watt all
okay what tool let's see
this is showing what I've typed in just
hit set here okay I'm getting some
movement I don't know that it's really
working like it was earlier no yeah I
mean it's just not changing anymore so I
think we've we hit some trigger where
it's it's not happy let's just do a
restart on this so that's clearly not
working at this point okay so let's
check stuff while that reboots can you
run it to has 10 for the megahertz so I
think you're asking me 10 50 megahertz
Oh way to compare with the fury acts we
kind of did that with Vega fue so yeah I
kind of I already did that a little bit
I don't know I'm gonna redo it right now
but check search the channel or the
website for fury ax they get a clock for
clock or something like that you'll find
we did that previously it was Fe though
okay
why aren't these cards not specified to
be built according to the lowest wattage
voltage amperage possible there is that
was supposed by jean-pierre there is
variance between every GPUs so just like
CP is all GPUs are not created equal
I think they Andy's gone a little bit
heavy on the voltage they seem to do
that in general but from what I've
learned it's there's a few things to it
one the GP is aren't equal and they're
certainly not gonna find you in every
single card that comes off the line to
make sure it's got a volt frequency
curve that his idea is optimal they're
just gonna do the best for all products
and then I also recently learned that
AMD tries well any of the vendors really
try to make sure clocks remain stable
and more extreme ambient conditions so I
don't know why I don't know the whole
details that but if you've got a higher
ambient I guess that's that might
explain why the fan curve is a little
aggressive in some environments because
they're tuning for worse conditions so
there's always room for you to improve
things which is kind of good and kind of
bad depending okay so this time I'm
going to I will just see if it works any
more custom run
does it work
I think we're a little low here let's
bring that up to 1080 again
actually let's start at 1200 just
started our baseline fun I'll switch the
scene in a second okay so we're at
baseline again
what is the asic of the card it's an RX
Vega 56 that's what you're asking if
you're asking the asic quality I don't
know yeah Silicon lottery does cannot
the website but the silicon lottery so
to speak does control a lot of how the
card to perform at different clocks and
voltages and things which is why they
just set a blanket setting for all of
them could still be I would think it
should still be lower but I don't know
maybe they have a really wide range okay
thank you windows 10 that's what I
that's that's exactly I wanted thank you
windows I appreciate it yes thing that's
that's great can we stop now
okay all right so 1592 twelve hundred
twenty three point something apps let's
bring this down again it'll be careful
or Windows will take control 1,100 1,100
I know that should hold okay so we've
seen some change there so it's working
we're it's back to functional we've
dropped 24 watts by going to 1100 let's
go for a little bit more no that's too
much let's try 1080 again 1080 let's do
this incrementally do we drop from 21
not really
10:50
not really a change
oh there's a drop okay but we've changed
scenes too so just to demonstrate 1200
okay 24 amps all right and now 1150
Piko down to 21 so we've dropped three
it's 36 watts 1120 1120 apply 20 20 okay
let's drop another couple watts let's
hold there let's just just stop
I'd like my goal is to get this number
down to like 17 because that would be
the same power consumption as stock and
we'd be able to push higher performance
and that's kind of what we did
previously but also we have the driver
issues so let's just try this and see
where this gets us
okay so fire strike custom run now okay
thank you
okay let's see where that goes how's our
temperature over here MOSFETs are at
like 40 Celsius so they're fine
yeah the newest drivers have definitely
fixed some bugs reduced p6 with while
tool you know II should actually start
using I mean I kind of yeah I don't know
I don't really trust what man yet let's
see how this goes first will mess with
wattle immediately after this because I
keep saying I will okay that's clearly
not doing what I want so we're still
higher power consumption than I wanted
so uh what tool time
okay
yeah and these are like all turned off
but a runner if it don't know if we have
access to these throughout toilet or not
wat man turns them off 1592 that is not
what we want - 20 yeah okay it just
resets for anyone curious so it's I mean
they're just disabled right now so we
have P six and seven 21 okay well it's
22 21 22
we are lower than we were which is like
24 watts lower it's doing something
definitely not as much as I thought it
would immediately but we can do more
here
can't go under one point one volts
because p6 is one point one reduce the
bottle yeah I just did that all right
let me get back to some questions
Anthony many out to Palouse
will there be any comparisons on all
three cards a few 64-56 well we've got
Fe and 56 and we just have 64 comments
so the answer should be yes
when I use a Walkman and set p6 as the
men I mean I kind of did that it's right
there okay so what we want to do at this
point we can there as far as turning
2000 so we got a score improvement but I
don't know that we got a power
consumption improvement let's let's log
the next run so let's we're at one point
one two right now and
I'm just reading some comments while
haven't been update it doesn't have the
attributes yeah okay alright so let's do
fire strike I'm gonna log this wrong
just graphics wanted to be a bit quicker
11:28 59t to 50% offset we might not I
don't know maybe we should just get rid
of that next let's run that is that what
I want yes that is what I want okay
that's from that and then we're gonna
record this run as soon as it starts
anyway and then what I'm gonna do is
reset all these values and record the
next run and then we'll be able to see
cuz the it moves around so much that
it's kind of hard to just eyeball it and
know if we're actually getting value
here that's why under volte that's why I
wanted to livestream it cuz it'll make
me do more work to see if it's actually
applying first of all so we're at 23 so
we'll just let that record and then
compare the two log files at the end
because it'll look really good when it's
frozen when fires Jake is frozen the
minute you start actually testing it
everything kind of you know fluctuates
cuz bouncing around between different
load levels yeah different power yeah
yeah throw in power different teens yes
we know that's why I'm logging this one
because it's too hard to eyeball it
okay please make a water block from the
nock to a full covered block chop up the
heat pipes and put fly water into it I'm
sure you'll figure it out but way too
much faith can you voltage HBM memory
for better overclock I we I think the
answer that question is no if I'm
understanding you correctly we need both
yeah we need both okay what was our what
was their power reading like you can
stop this one check and then we'll
record the next one at 1200 put the same
frequency apply okay
all right run customer not looking for
scores right now just trying to I'm just
making sure that under vaulting is
actually working and we're gonna start
this recording once the loading screen
pops up
okay I started
all right yes so this will tell us if
it's actually functioning and holding oh
yeah good call
well yeah I have to set up another DMM
we should do that though that's a good
call someone suggested measuring voltage
at the core during the test which I
should definitely do that and we can
I'll just have to swap around where my
my volt probes are hooked up because
this this thing can only output two it
can only output either current or
voltage you can't do both simultaneously
but I have another one I can use looks
hotter though right I think that looks
higher which is what we would want to
see that would mean the under voltage is
kind of working all right so what are
these results we were at don't make me
export this
so 1,200 we switch the scene back to us
yeah throat probe turns off when it's
not logging that's why it died 1200 okay
so where is our output
measurements
okay so the previous measurement is
loading right now
all right let's uh let's export this to
the computer CSV send so this is what
are we run this at ten something 1100
give me a second everyone we're trying
to validate the dysfunctions
under-voltage okay so that is being
mailed and then send out the next one
this was 1200 millivolts
send this the graph tool on this
software doesn't let me pinpoint the
current at different points during the
test 1200 send okay keep the keep the
camera therefore it will pull it over
this screen in a minute once I have
these things out of my email
okay you guys probably watching the
watching the current read out closer
than Iowa's so if anyone thinks that it
was actually working versus not please
let me know what you think I'm about to
drop these into a spreadsheet moving
everything around alright so let's just
make a scratch sheet okay and we're
gonna import this first file a play will
move this over to the other screen in a
second download its report records one
okay which one is this
so for anyone who isn't up to speed we
are moving over the records from the
current clamp to make sure it's actually
doing something 1200 today Donald down
first import almost there
and port okay
all right so what's happening 23.6 just
making sure I got the right sheets 23
624 524 422 for okay one more
under voltage alright there we go
okay so file import upload
I'll get back to chat right after this
this last file I need to import okay oh
it looks good so far give me a second to
generate a chart
okay charts
you're gonna bring the you can bring the
camera over if you want to you so let's
do a my output into the camera right now
yes okay so we're just gonna do this
kind of lazily this is which ones which
this is okay this is 1200 millivolts and
we're just gonna pull here down to there
I guess I should set an x-axis era in
x-axis so the x-axis most boring live
stream ever this is like what we do
normally actually I think these
incrementing point two five yeah
zero point two five five zero five
okay alright so there's that one
Josh this chart editors garbage
alright so there's one and we want this
and we're going to want the other this
one was some lower voltage that's call
11 Huntington 80 but okay and then this
we can set properly
all right so what do we get
okay there's our current output for each
test so the under vaulted one is the
redline you get in a can you see that
okay on the screen you're gonna get
infinity alright there we go okay so it
looks like I was working yeah oh yeah
that was working and the frequency was
the same so that appears to be
functional now we need to just get it
lower so we were at 24 versus 22 what's
our average difference here
let's get just an average across the run
I'm just gonna grab where it starts and
then where it stops load
it starts here stops somewhere way up
here twenty two point seven six amps
versus twenty three point seven okay so
I mean it kind of looks like kind of
works okay let's bring that back over
here
I'll check chat so maybe maybe not the
most exciting what a try a hardware luck
eleven twenty thank you I appreciate it
11:20 ok so eleven twenty was that one
so try hardware locks his settings of P
five one point six one three P fight you
said people oh I see one point zero
seven zero millivolts I did not actually
read their article I did see it go out
though I can try that yes
anyone know the app he's using on his
phone so this is the what do they call
it it's just like fluke something or
it's a fluke app the steam do everything
himself no helpers
are the cameras moving super chat John
fig Figueroa said for five dollars thank
you John
left click state six comment and set two
men it is cycling through state zero two
state seven not between state six and
state seven also good luck okay thank
you let's try that state six column and
set two men are we talking what tool or
wat man
I'm not that familiar with what tool yet
what was the suggestion right-click
state six so Ryan if you're a moderating
chat keep an eye on that for me let's
see if see if John sends through another
message to clarify is controlled by AI
camera by answer twenty two point nine
verse twenty three point seven yes okay
so
left click left click state six column
and set two men cycling through zero and
seven not between six and seven
elastically okay you know we're gonna
gotcha I actually did not know that was
a feature in watt man interesting I see
I understand that okay let's see if that
works
thank you John I appreciate it let's
bring let's let's lower this down to a
thousand maybe that did something I
think you're right I think it was
actually cycling through the other
states apply we don't need a wat tool
right now let's run a quick test a
thousand might be a little aggressive
that's that's a little too much let's do
ten eighty and I'll switch the camera
back to the other scene everyone says hi
Andrew Andrews ignoring you he's
operating the camera kind of reared and
you hahaha alright so replying that at
ten eighty millivolts
let's see now that we've set this to man
thank you for that suggestion let's see
if that does anything for us run custom
okay
I'm gonna might as well record this run
I don't I don't think it'll export it
again it's kind of a pain on stream but
we'll keep it closer eye on it though
okay
1592 megahertz ten eighty millivolts
allegedly we'll see if it's stable it
might it may not be stable because of
the lower voltage but we'll find out
okay alright so 21 is what we're reading
that is actually improved that is better
if it I mean it's different for each
scene but it does look lower it looks
like it's actually functioning again so
cool yeah we're at 1080 right now
millivolts 22.9 was lowest before let me
check that for you
the lowest we were at well I don't know
was the lowest we for roughly the same
scene we were at twenty one point nine
up to twenty three point one or
thereabouts
and now with the advice from chat with
the new minimum state set where are we
here we're a bit lower yeah okay let's
look before I export this let's do it
another run I think we're I think we're
cooking with fire now so 10 see if we
can do 10 60 and see if we actually get
a readout that's lower
are we still offset 50 yes
okay
runtastic MOSFETs are looking good
that's just by the way that MOSFET
temperature in the 40s I think it I
don't know what's going to under load I
think it's under 50 though that's what
the fad 50% or something over the
MOSFETs with just those Arctic cooling
aluminum blocks the cheap heat sinks
that are on top of them they seem to
work and pretty well honestly but I you
can get away without those though you
can definitely just need the rector
cooling on those there exposed copper ok
frequency is holding 1592
10 60 millivolts and I'm keeping an eye
on our range here so 21 to 22 something
it looks better
I mean yeah that that's seen this better
go check chat
see
nothing in the new drivers I'm not sure
that's referring to earlier chat but it
is that there is actually like an
improvement the drivers okay so check in
comment alright so how did we do here
when I export this one so you're gonna
have to bear with me for a second this
was at let's taking out this time 1060
1060 and save that note export CSV so
this current clamp was recommended by
builds IDE it's it's crazy good I really
like the login functionality of it 10 60
millivolts 1590 to send ok give me a
second to pull this up over here
at this point you're basically watching
my job people like that they're cool
okay
download that where is Google I don't
have Excel on here
this one is what you guys say 11:20
right was the first one so that's 11:20
and this is gonna be 1060 right yes 1060
okay
file import
alright let's import that
okay you can bring the camera over and
let's switch this machine so let's just
do a quick
quick chart here at a chart 1060
that's ten 60 millivolts okay they're
not going to be aligned perfectly but
all right google you this new chart out
of editors garbage anywhere night
google's listening okay so we want
ten sixty and sixty okay
still 1p7
187 alright
did it drop or is it just not gonna show
up
thank you
it's it on the Rana axis or what left
axis left axis left axis
where's our red line ah
they import the same file or something
let's export this again it's no way that
was identical saved files measurements
now that should've been the right one
now
okay one more time if it works then
we'll keep going cuz I won't have to do
this again after we know it's actually
functional for sure I think it is that I
saw it Hardware luck exposed to that
article 1060 all right send yeah
and what's the ambient in here 26
degrees is it obvious
Paul Hamilton where are you going to are
you are doing a video about the
manufacturer cost and how much it cost
and you make Vega we've modified that
video so many times now the answer is
yes but it's it's going through
modifications again I'm waiting for some
feedback from them on some stuff some of
the numbers I got okay where's my
okay we can work this time that would be
great
file import
I think the next step will just be
overclocking because it's gonna be a
little more straightforward in theory 10
60 millivolts okay all right let's bring
the camera over so edit chart
hence no walls
why is that just magically in 187 again
no cuz I'm on there on one maybe there
we go
alright see what happens ok got this
chart editor so this used to be the
chart editor and you click this they put
what you want see how see how that's
wrong I typed this in like four times
now
come on Google give me a break thank you
yeah okay that chart out of it every
time you click they use old editor
button it's like give us feedback on how
to improve Google just go back to the
old one it worked
this one sucks I hate 22 so it's working
there you go there's your there's your
evidence that under Bolton is working
this time they're not aligned I'm sorry
so we're at something like let's pick
the low points here no point twenty
point four it looks kind of an outlier
low point over here twenty three what's
this down here yes I mean it's obviously
working what's this one so ten eighty
millivolts ten eighty suppose to be
eleven twenty eleven twenty minutes
so eleven twenty millivolts we were
clearly under the twelve hundred and
clearly over ten sixty so it's
functional and our scores should be
about the same I don't think I need to
import these anymore other than maybe
one last one let's uh let's see if we
can draw this a little lower who wants
to in chat send me the this someone
wanted me to check against what hardware
Lux was getting so who if someone could
send me those numbers again what did
hardware locks use and hours may not do
as well be or it may be better Silicon
Lottery and all that but if someone
could send their settings over let me
know and I will put those in over here
and see what we get okay thank you
hardware Lexus settings p5 sixteen
thirteen megahertz and ten 70 millivolts
let me just copy that so I don't lose it
Oh Thank You Ryan okay Ryan moderating
chat so 16:13 1074 p6 so 1613 this may
not work as well for us as it did for
them as I said and what was P what was
the other one 1613 1070 okay so they use
to the same for each one right so 1613
let's see if we can get the same kind of
output that they got were they 50% off
set does anyone know we can we can
either do 50 or we can go down to
whatever they use let me know and let's
set the screen over there again yeah
yeah so fair point our card is a hybrid
is hybrid did have bourbon how our card
is hybrid so performance will be
different there too I don't I don't
think theirs was same settings okay 30
percent I was like okay cool thank you
let's see what it does
get current readouts okay there and I am
going to what are we at for voltage
again ten seventy we're not doing memory
yet so I'm gonna grab another multimeter
I think and check the voltage while it's
running let's do both graphics test this
time loop run let's see if it runs get a
score out of it and it just hit record
I'll truncate that later I don't know if
we'll end up using it okay so that's
that's starting see there's a couple
questions to check but let me get
through this run first
and see if I can switch these over
volt DC okay this is a much worse
digital multimeter than the other one
but it will work where was i measuring I
was measuring right here whoo what's the
what's that amperage reading say Andrew
twenty one point six ish one point one
that's what seems okay let me know when
it updates
is it running again not yet I don't know
yet okay so 20 I have right here in
front of my face 20.8 21 921 so we are
lower definitely we're lower than 1200
millivolts I'm getting 1.15 out of the
back of the card 1.15 volts is what
we're reading what was our setting 10
something 1070
okay well we were at 1.3 when it was
1200 earlier so certainly something's
working
I mean we're clearly lower here on the
current so that's great
the result was 73 watts lower okay good
to know
fire I strike a ultra huh is that right
now 21
205 so let's just take a note of that in
our notes and then someone saying they
ran they run normal fire strike
I'd rather compare against that slide
more data for that so we got a 10 16 13
megahertz rosary score on to a 5/16 13
what is the stock memory 800 800 okay so
how does that compare to our previous
scores on the old drivers stock plus 50%
which was 24 amps remember 24 25
somewhere in there there's calculator so
let's just call twenty four point five
there was 308 actually I recall so this
was 308 when I was checking it at the
time and this one we don't have an exact
number yet but this shows about 21 at
the high end maybe 20 what was the
highest number you saw do you remember
maybe 22 maybe 22 so let's call that
maybe 22 it's 270 so what is that we're
really not that much
the down 38 watts and then what was our
other previous score it's ready 21 205
is really it's sort of seeing better
here than here so the frequency
overclock is working so we are getting
better than we were previously which I
think this was 1592 plus 50 percent with
no under volt and no overclock so we've
improved over that by a bit turn onto a
five versus twenty one or twenty seven
sixty nine so we've seen an improvement
of two percent which is really not that
exciting but isn't an improvement and
then we were at twenty one six six zero
with an HB m and now just with an HB I'm
over clock at nine fifty megahertz so
we're about the same place we were
before some of that's probably a driver
update and then what was the request
here fire strike ultra someone wanted so
let's do a fire strike ultra no demo
window okay so let's run that just to
compare versus their numbers so someone
please remind me what was Hardware Lux
is number maybe change the offset to
1060 millivolts the same frequency OC
okay
same frequency overclock the offsets 10
60 millivolts yeah thank you
so where's our stuff there we go
okay and we're not overclocked in memory
ten sixteen seven seventy I guess that's
your suggestion okay so yeah reducing
temperature will reduce leakage correct
so it's not a perfect comparison
enforcers NL heading out thank you for
watching good luck with work tomorrow
Hardware Luck's got five five one nine
someone was saying five five nine six
okay so somewhere in there
yeah to person that's that's a good good
point two percent performance moves to
the ten percent power reduction is
pretty good
alright see what else did I miss here
Frank Dremel super chat no comment thank
you Frank for the 2.2 dollars much
appreciated
Nathan scantling this is old chat at
this point you guys really need a fluke
125 B scope meter I used one of these as
an industrial electronics tech well I've
got the name of that saved from from
Ryan the moderator so we will look into
that Paul Hamilton I think I answered
that question yes
todrick you what advantage does Vegas
fifty-six have over the fury x and your
opinion is Vega even an upgrade from the
fury X or a side grade I mean it
performs better it's got a way higher
clock so it definitely well yeah he's
performed I don't know I need to look I
need to look at my data Vega Fe clock
for clock wasn't exciting but these
cards you don't run them clock for clock
when you're actually using them they get
56 is the more competitive of the two if
it stays that price target okay so five
five one nine four fire strike Ultra is
that what our target is here so right
five one six six five five one nine okay
well we don't have exactly the same
setup as them either so
all right let's drop this some more
I'm gonna try a 1040 and we'll check the
meter and see if it's actually 1040
apply or at 1613 okay run
Jaso
how's it gonna deal with this I didn't
do windowed how I did okay
1612 1040 okay so we're targeting 1040
here I see zero five eight your live
streams are great I love seeing your
test methodology thank you I wouldn't
call this methodological but I
appreciate it it's really just kind of
haphazard let's see what happens and I'm
on stream so I'm not trying to like stop
a whole bunch and collect data
I mean clearly because I didn't even hit
record on that at that time but thank
you I appreciate it
glad that it's entertaining or
educational or something they oh they
overclocked HBM to a thousand megahertz
okay there you go
there's a difference we'll do that next
this is looking good though we're down
to twenty one twenty two from something
higher than that so down 30 watts 36
watts something like that hasn't crashed
I checked the back of the card before
this is complete volts DC back of the
card is showing
okay if we could actually work crappy
DMM that would be great
so we're showing 20.9 20.8 amps right
now I'm not getting a good reading out
of this
oh there we go okay got it
so what do we get one point one five
five that's like the same we saw a
minute ago isn't it
one point one five five vcore
that's from the back of the card okay
stop that what's our score did it even
work 21 to 95 okay sort of thin variance
let's get up to a thousand on the memory
next so yeah we stopped at like 980 last
time so if we get stuck there that's
okay so I don't actually know that
number right now okay
not 1017 1060 that where we are where do
I go to 1040 1040 okay so let's go back
to their settings I got some more info
from chat so they were at 1070 right
it's all right 1070 1613 more or less
HBM they had it a thousand I know 980
stable I don't know if a thousand is
stable for us so all we're looking for
is a five five one nine I think so let's
run that
so hardware locks five five one nine
with I just got I just went back up so
they were at Wolfgang rein says 1,000
megahertz HBM PP five and six for P six
and seven I think is 1613 and 1070 so
that's what we're running and I think
they were probably offset fifty percent
right on power have you tried using a
sharpie on the beak or you know I
haven't but that would be a good idea to
enraged commenters
and I was scope with the current pro
we've got a fluke meter on the clamp
down there already I do want to go to a
noscope at some point though
what is the results of under Bolton
you're late you say well the fact of
being late means that you missed a bunch
of trying to figure out if it worked and
how it works so we're currently at tense
we're validating on the numbers that
hardware Luck's got per request of a
chat
it's not true apparently apparently news
h3h3 won their lawsuit did you hear
about that I'm just reading chat don't
repeat me look into it later with all
the programs running will it affect the
score of the benchmarks maybe a little
bit but I'm not really running tests for
review right now I'm just running tests
to see if it works
yeah I do think it's ignoring the vcore
settings somewhat I mean somewhat where
we're bouncing around because the scene
is changing we're definitely lower if
you missed it we had a we had a chart
that showed some of the the results and
they were actually lower amperage each
time but I do agree it starts ignoring
your settings at some point here five
for 60 we're retargeting five five one
nine that's pretty close
it's not bad we're 20 megahertz lower on
the HBM what's our improvement over the
last one with what I say we're at 50 100
or something so
HBM helped out a lot just make another
this is getting messy but ultra five for
six to sixteen thirteen nine eighty one
seventy okay alright our name recording
okay so I mean I think I think I see you
kind of it's it's kind of working
I'm not getting the voltages I wanted or
the the wattage out what I wanted
let's yeah I don't know 10 70
you just so I saw some people saying
they got - 70 watts or something
what were what were they getting like
what's the actual number not just the
Delta does anyone know what their actual
wattage their power consumption was and
it was where they tent where they
measuring at the wall or at the rails go
to 1,000 yeah sure why not we can do
that while I'm looking at chat 1000
and again keep in mind I don't know what
system they were using or anything like
that we don't even know if a thousand
stable we're about to find out
even Klein said on Twitter we won the
lawsuit video coming soon a huge victory
for various on YouTube cool good for
Ethan on water what is the higher court
clocks you can push I don't know we
haven't done overclocking yet I mean I
did I did do every clock in but if you
remember that video where I talked 394
watts at stock that's got to be the wall
right
yeah that's at the wall for sure so as
you can see as you can see it didn't
like the thousand megahertz on HP M they
must have had a better bin than we did
were they on 64 they may have been on 64
we're on 56 mind you actually that's a
so were they on 64 could someone check
for me
happy happy sharpie your cat now we
haven't drawn on snowflake with sharpie
because as comments will leave do you
believe that means we respect the AMD
IHS last because we used a dry erase
marker on it to show where the dies
right all right of facts 1000 yes that
is tell me what we got 394 okay so
where's the Vegas 64 is oh is it Vegas
64 Vegas 64 16 31,000 millivolts no wait
that's not right hey 1050 on 64 or yeah
yeah yeah hard relax did both cards okay
56 thank you thank you for the clear
answer all right so that's clearly not
stable 990s kind of stable I tested it
last time is it still are we gonna need
food okay you know I'm getting bored of
this so we're gonna go for overclocking
now I'll do the rest of the undervolt in
as part of the review process I need to
sit here more methodically and like
properly do stuff and write it all down
and everything but we got an idea here
it's not as good as I wanted it to be
but we might be able to improve things
when I can actually sit down and look at
it more closely let's go for
overclocking know so we know 990 is
roughly stable 60 we don't need this at
1070 anymore we go to 1200 you know what
also we don't need is a 50% offset
that's kind of pedestrian so we're gonna
go give it a hundred this kind of buggy
this power play table it seems to do it
doesn't does something but it's kind of
like okay so
firestrike
benchmarks got the most data again on
this once when you go with this home
that's going to rename this window it
okay all right
50 may be unstable we'll find out
so again folks you missed earlier parts
in the stream it appears that frequency
reporting is working better than
previously so good job AMD for fixing
that quickly because when I type in
nineteen eighty megahertz now it crashes
rather than saying we're running at two
gigahertz so that is good news
so 4.2 there we go there you go there's
a number I want to see twenty-nine amps
so we're definitely drawing like three
to four over the fifties hosta head
that's pretty cool
what's our temperature looking like I
need to keep an eye on that now now that
we actually care about such a thin 30 C
reported through watt man set HBM
state32 min max okay I will do that next
time thank you by the way John once
again much appreciated readings from
Greece greetings to Greece from the East
Coast
and then max okay
twenty-eight so we were hitting 28 to 29
which is a 100% offset I worked with
build Zoid and builds what I believe
worked with helm who those BIOS or
power-play mods so what is 29 by twelve
point three that's three hundred and
sixty watts so from for a minute there
we have three hundred sixty watts going
through Vega 56 which I previously were
capped at three hundred eight so that is
a very clear increase so that's really
cool the powerplay table worked and that
put us that gives us more power budget
to play with that means we can get a
higher stable overclock than previously
there that's that's looking pretty
respectable that's what I want to see
notepad
twenty three two one four sixteen fifty
to ninety nine the at twelve-hundred I
wanted a center power was three 56.7 so
three 357 watts which puts us plus 50
and round at the botch yeah that's right
that's right right plus 50 over the
previous 50 percent offset
I didn't see artifacting someone else
might have caught it was looking like
closely but 2324 what's that improvement
high score previously these are all
bugged ignore these 25,000 results those
are bugged invalid previous high score
there we go stock card without the
liquid cooler twenty two two two eight
so we got a four point four percent
improvement that's not bad
I mean that's better than the jumps we
were getting previously cool can it go
higher 16 72
okay here we go
MOSFETs are pretty happy
they're staying under fifty C we just
need them under 100 we'll be good yeah
there's no pad plus plus some times but
it's got a bunch of stuff open in it
okay is that better than 1070 do you
mean then 10 70 millivolts already mean
then a GTX 970 I need to check my
numbers if someone wants to check our
review of the fire strike normal fire
strike 1080p results we'll have a gtx
1070 in there so like I said we've got a
what do we have for power Oh 30 amps
there we go
I think we want to avoid like 500 but
for hundreds okay builds are actually
after he sent me these power tables I
asked Bill join for tables that did 400
watts and he was like after I started
using my muse like hey I think 242
percent actually goes over 500 oops
why not work it it didn't kill anything
yeah can't wait to see how much Headroom
Vega really has someone says I mean so
here's the thing this is like terrible
to do for normal use because you're just
ignoring everything about power the
efficiency goes down really hard once
you start doing these kind of over power
settings like the gains you get per watt
are just so much smaller than stock more
than the 50 percent offset but uh yeah I
mean that's that's what we're looking at
right now where are we
2028 with this scene which is a bit
higher so that's cool
47 48 Celsius on the hot spot MOSFET
which means they're all gonna be under
50 are we out doing our we got to in any
of our previous routes are let me open
up our fire strike results they have 56
reveal
so twenty three five two six what we get
from fire strike previously I'll paste
this in chat once I can find the right
chart fire strike
okay so previously where's our chart
with all its many charts let it fire
strike power consumption okay
there's our scores just go pull up a
chart here sympathetic GPU
okay so previously we were at fire
strike normal 1080p I sort this by
descending data sort by graphic score
okay okay go ahead so I don't I don't
think I have a chart actually generated
for this right now so we're just gonna
kind of look at the numbers on the
screen
okay so is this even to we should I
generate a chart for that it's kinda
hard to see right
so we just want to assign the right
thing
okay so we're getting twenty three five
two six what does that verse of the
previous score versus the previous score
we're at twenty three two one four five
two six and we increased by like 20
megahertz so 22 five to six where does
that put us that puts us Vega frontier
edition stock was that is that with like
a fixed voltage or something let's pick
one I know what it is 1080 SC 222 893
and this is twenty three five two six
now the power consumption is absurd it's
more than a hundred watts more but we're
beatin it so twenty-three what is that
what's that uh what's that look like
twenty three five two six
sorry I don't have a chart for this
right house we're just following the
screen twenty three five to six versus
twenty two eight nine three nine two
eight nine three that's an aftermarket
two point seven two point eight percent
two point eight percent faster than a
stock 1080 do I have an overclocked 1080
in here I don't think so not right now
we do have a 1080 ti stock 1080i twenty
eight
okay that's not what I wanted thank you
our is a 1080i reference is going to be
how much faster 28 - 67 - 23 5 2 6 so
we're still 20% ahead on the 1080 TI
reference card and that's what three
hundred sixty watts going through the
card now this is just fire strike mind
you but we're not we're not really close
to that right now so okay let's bring
that back around and look at the see if
we can push this a little further
hurry I want to try and wrap this up at
some point
1,700 1,700 okay let's see if that's day
way
I'll check chat in a second once this
starts running did I write all these
down already
would I say we're I like 360 something
something like that I can go up in the
history yeah sorry what's right okay
go
so again we're under liquid right now we
have a thermal take flow 360 millimeter
cooler on there technically if you buy
it it comes with three RGB LED fans they
are the advertiser for this stream check
them out we've replaced those fans with
two Maglev and one knocked to a fan
which is driving people crazy because
they don't match but I just wanted to
make sure the thing is where we're
running low our fans beats here for the
stream so okay so what what do I need to
catch up on
discord what's going on in chat yes I
know for yeah fire strike it's not real
game inference we know I already said
that it's the point of saying that
earlier but this is fast and easy to do
on stream and it's easy to validate
performance because I can't trust the
frequency numbers even though they look
better than they used to in terms of
reporting
Vega 56 is the newt okay that's not nice
after you find the highest collage try
undervolt in it I don't know that will
get a lot of room might be able to drop
20 millivolts or something disrespecting
hardware again I think that's a joke but
yeah
20 mark scores aren't translated to
games yes I've said that Garen sexist do
you know why modular PS use have weird
pin outs on the PSU side instead of just
copy and pin out of the connectors makes
custom cables a huge pain I don't see a
reason for it we talked about that a
little bit in our like PSA don't mix and
match power supply cables video it's not
standardized it allows companies to do
what they want so like EVGA he's got
those universal power supplies where you
can plug the cable into any spot and
they wouldn't be able to do that
otherwise so it allows companies to do
stuff like that or to put LEDs and the
connector headers now maybe the function
would be more valuable but that's a
that's the way it is
where do you think the eighth generation
will look at that once the time comes
okay
it looks like some people are having
trouble with we're comparing fire strike
scores looking at gaming right now it's
too hard to do gaming tests on a stream
takes too long I'm just trying to
validate an overclock that works twenty
three seven eight zero a that's an
improvement or 1702 and nine ninety five
hundred one hundred percent juice did
you look at the did anyone look at the
current at all I didn't see this time
it's like 30 amps again probably okay so
I want this to fail so that we can work
on closing the stream and find the
performance limit I'd like to get back
to the other work we have we have to
shoot another video today probably 17 22
that's you know I really wanted to fail
so let's just go a little higher so
that'll give us a performance ceiling
are we sustaining allegedly yes
allegedly we are sustaining the clock
30.4 apps thank you I appreciate that I
can rely on chat keep an eye on that
stuff
okay so 30 point for max thirty point
five okay so we're about where we were
before which what is what's thirty point
five that's it's 325 watts going through
a Vega 56 that's pretty cool
okay found the point failure 1732 how to
reset this
it's nice having the liquid cooling on
here I don't have to worry about
temperature at all just gonna let it go
okay
I switch back to where we've got signal
how is the noise that why it's honestly
not bad it's uh I don't know we can
measure it I don't really feel like it
but it's definitely way lower than my
talking volume probably like 40 DBA or
something but I've got the fans turned
down so when they're when those fans are
at 100% it's like I think we're likes
nearly 70 decibels the other day and
when you walk by you can feel like you
feel the air pressure it's a pretty
pretty sweet setup I'm actually I really
like this hybrid setup it's our best one
we've done I think just in terms of like
its kind of power Plato a lot on it so
that's pretty cool I'm looking forward
to doing that more in the future I don't
know if it works or how it works on
NVIDIA I'd like to try it with their
cards as well but something it's the
furthest we've gone you know you risk
the card at some point but I don't
really care
17:22 eyes it's yeah it's my job so if
it dies on stream that's fine but it
really shouldn't we're well within the
power capabilities of the VR m9d let's
just let's give it a little bit more
room
one-twenty okay
do I have to reset this stuff up here
yes
there we go okay
so let's set this back
I think someone just asked about how we
measure the current so we're measuring
at the PCIe cables using a current clamp
the current clamp is outputting
wirelessly to this phone which was
recommended by builds I'd shout him out
again if you don't know who he is
search for actually hardcore overclock
and go subscribe to him there's a lot
more about this stuff than I do but
we're getting more into it we haven't
done this much with overclocking really
in the past it's it's pretty fun so for
anyone who missed the beginning part the
top number is a hotspot on one of the on
the vrm it's one of the MOSFETs this is
our current there you go 30 amps 3.5
looks like about the peak once again
and we're trying to see where it stops
being stable we lost ability at 1732
megahertz we're at 17 22 right now
did you get a response from board
partners regarding BIOS modification I'm
assuming you're asking about my question
I sent out about will board partners be
able to bypass the locks that prevent
the flashing from applying and the
answer is not really I did not get I
think they're just trying to make a card
at this point a lot of them have been
pushed out to September from what I
understand they give Ryan Ryan link to
build Zoids channel and chat if anyone
wants to see it
so at this point I don't know the answer
that board partners have been pushed
back quite far on their launch with Vega
cards so yeah we'll see once once they
actually have something more to talk
about maybe they'll answer that question
okay try 1613 950 intensity we did that
a little bit ago we're on - we're on to
overclocking now we did under volt in a
bit ago see the valley one two three do
you know why
modular PSU our answer that one huh
Tommy thousand faces does over vaulting
HBM do anything for overclocking memory
we don't have control over the HBM
voltage we have control over the I think
it's like a voltage controller voltage
which talking to builds are there's not
much point in modifying that I don't
know if his opinion has changed but
that's sort of thinking originally so
where are we now twenty four one three
eight hey that's an improvement
twenty four one three eight seventeen
twenty to nine ninety sort validating
with performance and it looks like
performance is saying that this clock is
actually applying underside like 375
Watts here all right so let's start
looking at numbers that's that's roughly
our area of stability so we are at
twenty four one three eight verses or is
our highest stable number before on the
stock card stock cooler stock power we
have now boosted performance six point
five percent that's pretty damn good
we've also boosted power draw it's by
huh
how much have we gone up on power draw
like 22 percent soho six percent more
performance 22 percent more power pretty
good not really
hey super chat thank you it's been a
little while since I've seen one of
those pop-up super chat wishbone five
dollars Thank You wishbone I appreciate
it it helps with doing live streams for
this lon
I think we're gonna spend it on fast
food to have the stream should I buy an
asus strix mega 56 instead of a 64 when
it launches yeah there's the question
because I thought yeah I don't know
actually I don't have numbers on 64 yeah
I just got a 64 n so I'm really curious
to see if without the crazy power stuff
where's our number 22 6 5 9 I'm curious
to see if we outperform 64 with an
overclock on 56 because if you can then
that's a better deal to buy an add-in
board partner 56 overclocked it and
spend less money so I think 56 should
actually be better the BIOS lock really
sucks but I don't know if partners will
be able to work around that we haven't
tested 64 I'll be doing that very sure
we just came in yesterday can you drop
millivolts we already did that
okay so we're any other major questions
I think that's that's my son right now I
don't see anymore super chats right now
okay so I don't know that's I think that
pretty much shows what we're gonna hit
today we broke at 1732 held at 17 22
it's not worth trying to find that one
megahertz barrier this might not be
fully stable in games its stable and
firestrike normal if any you know
justjust because people have been
requesting it I'll run one of these
before we close the stream out so fire
strike ultra you can use this number to
do some comparisons with our previous
ultra numbers and that'll push the HBM
to more so might might cause some
instability at 990 we'll see they're not
too windowed hoops
try to drop voltage okay fine let's try
I'll try to get on to like eleven eighty
or something let me check the check the
voltage on the back of the card
whoa thirty holy crap thirty-two amps
yes 33 yikes
what's that what's that I don't that's
that's quite a bit more our 20 so we're
hitting like 400 watts now turns pi not
gonna lie what's the thermometer say 46
okay
did it crash no it looks like looks like
crashed right that's right oh okay cool
house I have to turn it hot uh what am i
checking over here a temperature of the
GPU I guess should be fine I guess we'll
see it once it closes out okay looks
stable super chat it's okay sir so yeah
I was on the other side of the table
super chat daniel monsanto thank you
where is the security code processor on
the dire elsewhere on the card i'm not
sure i was starting to build to write
about this i think our last
understanding last time i checked with
him was that it's on the die which means
that we're not gonna be able to do
anything about it really some people on
the overclockers forums are comparing
the different BIOS revisions on
different cards and they're trying to do
some reverse engineering is on locket i
don't know if that'll actually work but
that'd be cool you can kind of get get
decently far with power play tables
though which is a registry mod and
that's what we're running super chats
elaine abrahan ten dollars no comment
thank you Elaine I appreciate it it's I
mean again like this is this is the
longest dream we've done but I needed to
do all this work anyway so it's kind of
like doing my job while doing my job
super chat tom's
Bremen Smits says what is it okay that
was that Norway or something kronor
what's your thoughts on liquid metal
instead of Tim well I don't know I mean
like liquid metal between the CPU and
IHS and the cooler called plate that's
pretty good I don't really like the idea
of it on video cards just because well I
don't know it's it's kind of not really
worth it
okay our temperatures are already so low
it just doesn't seem worth it to me so
anyone wondering we were at 59 56 for
the score on that I'll just go ahead and
type this down for my own reference -
956
what'd I say was where we had 33 amps
so what was the 406 watts holy crap
so let's not increase that power target
anymore I think we should stay right
there
a hundred
when I say was 406 watts and that's a
offset of 120 percent roughly okay so I
said we do one more and I need three -
couple more questions so let's run one
more just per request I'm gonna bring
this down to 1180 and see if it is
actually stable people are asking for
undervolt in at the overclock I don't I
don't know with like here's the thing
we're drawing 400 watts
through the PCIe cables once you're at
that point under volt in is not really
kind of like it just doesn't seem that
important but we'll do it see super
chats got one from oh yeah the liquid
metal question yes so CPU IHS in cold
late yes that that seems worthwhile and
in some instances GPU I don't really
like the idea of right now it's kind of
messy Reggie Reggie not oh five dollars
I think Canadian says I just hope this
helps awesome work and no need to chat
me out should have put that up front man
but thank you super chat Zach attack
five dollars thank you for this well
thank you for that
super chat John once again ten dollars
whatever you get for food get a large I
think that's what we'll do I think big
Andrew and I are gonna go to our local
establishment show them the stream and
demand sponsorship super chat David
Knight five dollars go buy some Vegemite
to eat can you even get Vegemite here
all right what's the oh it's the voltage
readout
I didn't see the amperage this time if
anyone saw it let me know what it was
one point two seven volts okay so about
where we were earlier which was one
point three one point two seven volts
with at least 406 watts
I'm back at the chat now
tamps oh yeah good question
why does this not have like a log gpu-z
doesn't see it yet does it no okay it
should be low 5956 okay Sam score
oh that was with the under bolt right
yes okay it did go down
temperature
I have one more
keep an eye on the amperage while I'm at
it okay I forgot to do a windowed
okay benchmarks ultra custom FX window
now
do not include the demo okay see how we
did with this
Frank Dremel $5 thanks for doing this
kind of gritty testing helps a lot
well like I said it helps me a lot too
I'm learning this with you all we're
able to kind of figure out what I need
to actually do once it's time to test
this mod properly put all the numbers
down do the game benchmark stuff like
that
okay
are you going to be in scrapyard wars 7
or 6 as a judge next question GPU
temperature 32 so again I think that's
kind of inaccurate but I mean it should
be kind of close to reality because
we're reading below ambient previously
34c that's with a thermal take flow
three sixty millimeter cooler on there
which is a nays attack 4.5 gen pomp that
I've modded to fit Vega 56 we use the
same mod process we did for Vega
frontier edition and the poem said 100%
the fans are at a I don't know low lower
1575 somewhere in there and we're at 34
Celsius it's pretty damn good
I mean I'm happy with that and that's it
what's our average 29 right now hey
that's better so that's eleven eighty
millivolts so the under Bolton did
actually help at this higher frequency
anymore m80 xb3 with DHC sony boards
coming out I'm not sure I haven't heard
about any lately we are doing some stuff
with AB III of the board right now
Patrick is working in the other room on
that stuff some power testing
okay they're using fluke on the graphics
card we've got a current clamp by fluke
on the PCIe cables yes
32.7 ounce max thank you so I mean not a
big difference about something it's
something okay so an Yin Matt will it
run crisis I don't know if it'll run any
games know what it'll do
Vega 56 okay I'm just looking at chat
yeah sorry three thirty two point seven
max okay so there's your answers we're
at roughly where's our temperature we
were peaking at 37 under that short test
maybe we hit 40 or something that we did
something longer
3950 5956 was our score and that was
with 17 22 990 megahertz on the HBM 1200
millivolts which sustained at 1180 that
we got a slightly lower score the second
time around at 58 26
and 406 Watts here 30 2007 here is that
what it was
this is a mess both to clean it up later
okay so that's what that's there's these
days your overclock there's a 400 watt
Vega 56 pushing 1080 levels of
performance at 400 watts mind you so two
times the power draw or something like
that but it's doing 1080 performance in
fire strike now fire strike is not a
video game
some games will support Vega better and
probably you know do better in terms of
the 1080 comparison I know that you
would likely know that it's not very
each by one of them quake might be one
of them but uh for fire strike it's it's
about tight as a 1080 sc2 all right last
few through here Bing talks what's worse
power let me give this back over here
whoa well run quake to hit pry up 3000
fps why not use excel because I don't
have it on this machine it's just a
streaming machine okay so whose was I
reading Bane talks
what's what's the worst power draw
you've seen for arcs Vega so far well I
mean I 120 percent offset 406 watts at
50 percent offset 308 watts they can't
go past there it is capped by BIOS at
308 watts or whatever the that current
draw is so that's that's the answer that
one
super chat thank you Tom Braz myths once
again 20 I think Norwegian kroner I
think that's what that is what's the
difference between VI D and V core
voltage I've looked that up so many
times and it's not something we're
working with today but on the CPU side
VI D is what you'll use to tell the CPU
what to do and like X 299 is a good
example what were we hittin on that 1.15
or something v ID but the V Corps was
actually like one point I don't know
eight for 700 X there are better there
are actually answers to your question
online I don't have it memorized right
now sorry I don't dig that deep into
overclocking on a daily basis to
memorize everything all the voltages I'm
working on I'm working on it
we've been working with builds all right
he's helping me out my focus mostly on
the tasking stuff though super chat
Sebastian crowd Houser two bucks here's
one gigabyte valley SSD storage thank
you
this camera if I I guess the stream is
benefitting us in that regard the camera
for like a normal shoot is about a
gigabyte per minute with the lower
settings so super chat wishbone five
bucks gaming related question do you
think the Xbox one action project
Scorpio is a great idea there's the
first wide sale water-cooled console is
it water cooled I thought it was a vapor
chamber we've got one coming in though
November 7th it comes out right so i
pre-ordered that we will have one we'll
be doing a teardown if I get it midnight
launch then we'll have a teardown that
morning
online so yeah keep keeping an eye out
for that we're working with it and
result here what is your opinion of the
Vegas series check back with me out to
Vegas 64 I've got a review up for 56 you
can check that out for opinions on 56
with the
the stock card and result here for
overclocking once you start modding it
and stop it
it's doing decently I've drawn a hell of
a lot of power for the performance gate
and we're getting but it's fun to work
with
I wish the bios were easier to mod but a
DOS Steve ever covered games I used to
so we used to do a lot of we did game
reviews we did we do well we still do
gaming benchmarks when new games come
out of the thing is lately there haven't
been a lot of new games so I haven't had
anything to really cover with that but
destiny to that's coming out we're gonna
be benchmarking destiny to a couple
other games are coming out later in the
year John once again 56 plus 64
crossfire 56 Plus Fe is 64 plus Fe I
don't know if I'll do all those I do
want to do some some mix crossfire
testing we've done them in the past with
that we did with a 5/8 in the 480 I was
basically the same as 2 for Hades so
yeah I hear your request and we will
probably work on something like that
live streams hair down for the Xbox I
don't know see like that's the thing we
livestream the teardown I think I think
we'll get more views on it if we film it
and process it and edit it so that we
can like just get to the point faster
and YouTube seems to distribute the
uploaded videos a little bit better
after this the I mean once the stream
ends this video won't get distributed
that much so I think will do I need to
make ro ionics I bought that thing put
550 bucks will probably film it upload
it but we might stream some other stuff
with it ryan length of vega 56 review
and chat ran who wanted that you're
supposed to convince me that I wanted a
Vega 56 Steve our review has data in it
if that convinces you then go for it but
I'm not I'm not a salesman for AMD this
is how my job okay I think I think we've
hit them all so I think we got the
mostly questions can Vega be cross fired
I haven't tested it yet it should be
Ryan from PC perspective tested that
they are focusing less on crossfire
though just like any videos focusing
less on
no one really not enough people do it
anymore do you play games just bench are
you mostly benchmarking
ok so let's wrap that one up
there's your under bolting and
overclocking overclocking towards the
back half wonderful things towards the
front half under both end I'm a little
disappointed with I want to do better
with it I think we'll have to sit down
and spend a lot of time working on that
one because like the saws were kind of
lies to you you hit a threshold where
just the voltage stops applying seems to
do whatever it wants so I'm gonna need
to really sit down and work on that and
then overclock I'm pretty happy with so
overclocking looks like we're stable in
fire strike 17 22 megahertz 990
megahertz on the HBM to 406 Watts
through the PCIe cables quite high and
that again is with a 120% offset on the
power so I'm pretty happy with that
aspect of it not as happy about the
under Bowl thing but we'll get back to
it I might need to under bolt with the
without the power target offset stuff ok
cool
you only under bolted 20 millivolts at
17 22 I know
well I we got to cut it off somewhere
right I'll see what I can do for the
look for the upload look for like the
actual finalized data where everything
is all nice and organized and everything
ok alright thank you everyone for
watching thank you
those of you who donated and we
appreciate it we'll do more of these
they are impromptu if you kind of got
surprised by this one follow us on
Twitter I oh wait right one ounce it
they're like an hour ahead which not
that that much ahead but twitter.com
slash gamers Nexus I'll answer this one
Steve do you listen to heavy metal last
time I answered this Parkway Drive and
trivia mainly right now there's some
other ones in there too though but ok so
we don't know how to end the stream yet
we should look into that we know how to
click the stop streaming button thank
you for watching subscribe to get all
the other stuff we'll see you all next
time
technically still rolling they can still
hear me
all right I wanna hit the stops change
button
We are a participant in the Amazon Services LLC Associates Program, an affiliate advertising program designed to provide a means for us to earn fees by linking to Amazon.com and affiliated sites.