Ask GN 79: Killing CPUs with Voltage, Fan Reviews Incoming
Ask GN 79: Killing CPUs with Voltage, Fan Reviews Incoming
2018-05-16
hey Ron welcome back to another episode
of ask GN as always leave your questions
in the comment section below for the
next episode quick note here we're going
to play around with the idea of
uploading to sgn videos in a week so
basically we're not removing a video
slot we're going from seven videos to
eight videos per week temporarily it's a
trial if it goes over well we'll keep
doing it but the current plan is to
basically upload - Astraea episodes in
one day so there'll be about twelve
hours apart
and that's just because we're getting so
many good questions lately with the
growth of the channel that we want to
try and accommodate more of those per
week so we'll see how it goes if the
reception is overall positive and if the
performance isn't too heavily impacted
by whatever YouTube's algorithm does
then we'll go with that going forward
but additionally as always we'll have
the bonus fgn on patreon for people who
go to patreon.com/scishow and support us
directly all tears get the extra ass
chin up so that goes up the patreon and
we'll be filming one of those for every
at this point two of the normal ones so
a lot of good questions this week but
before that this video is brought to you
by thermal takes view 37 case the view
37 focuses on highlighting custom PC
builds with its full panoramic window
and tinted front acrylic and our thermal
testing the view 37 performed reasonably
well when considering its looks focused
build which is partly thanks to the
airflow design and the removal of a
bottom power supply shroud for a balance
of looks and performance check the link
in the description below for the view 37
first question is from Justin Parrish
who says when will snowflake host the
news and will there be subtitles and/or
dubs available will will let snowflake
try hosting the next news episode the
next hardware news episode we do have
some good b-roll of her lately though
she's she's discovered a few new found
favorite toys for a little while she
liked the infrared thermometer go figure
it has a laser and now seems to really
really like the basically metal bottle
caps that she just knocks around all
over the floor next go and we'll get her
in some more b-roll shots next one st
reMed says Steve one fan reviews we have
some b-roll we'll put in the I guess
we'll put it up now but basically
recently did a behind-the-scenes video
for patreon where we showed some of our
new fantastic capabilities so it's off
to my right over here we've got some
shots of it but basically we have the
last piece now is coming in the mail so
we thought the last piece was the
costume breadboard that we worked with a
thermal engineer to make we have some
b-roll of making that team a we'll throw
it in but that breadboard is all done so
it's just a it's a power circuit so it
allows us to control via basically the
world's smallest scope the PWM going
into the fan so that will allow us to
eliminate the testbench entirely in
terms of making sure there's no computer
components to keep up for this testing
because a few things here so for fan
reviews we are going to be doing thermal
testing and it will be on a computer let
me just say that first however the most
important thing with a fan review if
done properly just as straight numbers
is going to be something like CFM which
we can do because we have wind tunnel
with a known volume so we can calculate
the cubic feet per minute flow by taking
the linear feet per minute flow and run
it through a spreadsheet with the volume
of the wind tunnel so we think CFM we
think things like static pressure
measurements roughly by impeding the
airflow with known common objects like
radiators and dust filters we have all
that stuff those really the most
important elements those things don't
require a computer to test other than a
laptop to hook up to a scope and send a
signal into the fan and tell it what to
do so we have the last piece it's really
cool it'll allow us to be super accurate
with our testing because we're
eliminating the variable that is a
motherboard which is a gigantic variable
you have no idea how many problems
motherboards have given us the way some
fan headers act versus others it's not
they're not all the same and that's a
problem if you're testing fans with
tight tolerances for differences we are
still doing thermal testing just because
ultimately you could see CFM differences
that look significant but then maybe
when you actually thermally test it they
perform somewhat similar or close yes a
difference that I guess it's all that
perspective
the CFM is kind of an abstract thing
when you're talking about or even linear
feet per minute ultimately at the end of
the day that's not a number that
directly mentally for a lot of us
translates back into performance it's
the most important metric for a fan
airflow is really the only thing that
matters theoretically on paper if you
know the CFM of fan a is better than fan
B that's really all you should need to
know in terms of which one will be
better thermally because it'll follow
and the only difference is for the most
part will be generally error or the lack
of difference might just be because it's
not as relevant as we thought and that's
important to test too so that's why
we're doing some thermal testing I'll
explain it more in a future video but
the short of it is we are basically
we're setting up a semi pseudo synthetic
but not really attached that's for air
flow for noise and then we'll be doing a
realistic test for thermals the whole
point being to tie it all back together
and say like okay so these two devices
LPM might be 100 points apart but what
does it look like thermally does it
really matter if a fan that's $10 more
and 100 LPM better or whatever is
performing roughly the same so that's
that's what we're playing
now when fan reviews found out I need
one more part I said we had one more on
the way that I wasn't expecting to need
that is a benchtop power source so
basically it's just going to allow us to
manually control the voltage and the
amperage going into the fans so we had a
problem with originally I hooked
everything up to an ATX power supply I
just stripped the PCI or the EPS 12-volt
cable and hooked it up to a bench top or
to a an ATX power supply that caused
problems for non pwm fans because it
just didn't work with the way we set it
up so I bought a bench top power supply
and that'll fix that problem I should
have everything we need at this point
they'll be really accurate I'll say that
much but it is taking quite a while
going that said once it's going it
should be pretty quick to knock out fan
reviews as new fans come out so I'm
really really looking forward to it
probably testing I'm most looking
forward to right now outside
our power supply testing which I think
we're gonna talk about that this week -
I just set up the sun/moon test system
in another room so yeah fan reviews
coming soon power supply reviews not
sure exactly when but we're getting
geared up for it we have all the
equipment for the most part and yeah I
know it doesn't let me give you a direct
answer the question I'm hoping to have
some fan tests done and published before
Computex so that should be the direct
answer to your question before economy
tax is my target can't promise it if
something doesn't work the way I want it
to we'll update you but I do think
that's semi-realistic at this point next
question blue in guidance says is it
safe to overclock twenty four hundred
megahertz ram to 3000 plus speeds by
bumping up voltage and kind of loosening
the timings if it's fine to do that does
the memory vendor matter if not would
you recommend more expensive kits and
they'd be like some pee pump it up a
little more so first part of the
question yeah it's there within
limitations it's safe to increase
voltage to push the frequency higher
generally the guidance is so for isin
one a couple guidance pieces on voltage
and these official guidance for risin
one safe long-term voltage is one point
four two core so that'd be where we'd
say stop for rise in V core on rising
one SOC voltage Verizon one one point
two volts is kind of the max safe area
the AP use are a little questionable we
got sony mixed reports on what the
different motherboard vendors and what
AMD recommend for safe SOC for the AP
use that we're just not a hun percent
sure I think our previous AP testing
content we talked about that and kind of
had some conclusions there be dim though
so the memory sticks themselves can take
a hell a lot of voltage like I mean
these a memory stick you could push 1.8
volts into it and it would be fine for
the most part now I obviously make sure
you know what you're doing but there are
people who run memory you have two volts
2.0 2.1 and the difference is why are
you running it that's the voltage so for
our testing we did with the x2 9
platform when we did the ripple TT
stream we
had our memory at something like I think
we ended up settling at 1.65 but we did
some tests at one point eighty five
volts that's fine the sticks were fine
they survived doesn't matter but that's
not a daily system it's not something
that's gonna have thousands of hours of
use on it it's kind of booted run
through some benchmarks and then
disassembled later once you get the
score you want so very different worlds
for daily use 24/7 use 1.5 v tim is the
recommended maximum verizon one risin
two and Intel 14 nanometer including sky
like acts so 1.5 is the recommended
maximum for long-term use even though
the sticks themselves can take more
you're still going through memory
controllers and SOC si IO or voltages to
pay attention to as well so I might as
well cover some of those Rison one I
said 1.4 to 1.2 for SOC 1.5 be dim
horizon 2 under 80 degrees Celsius TDI
and rising to the max recommended the
core is something like one point three
eight but it really depends on your
thermal solution rise and who gets
pretty warm it's not like as hot as X
299 but it's definitely way warmer than
risin 11.25 SOC Verizon 2 is max
recommended for long term and 1.5 v TM
also 440 14 nanometer Intel processors
1.4 to under 90 see if you can manage it
and obviously something like an 18 cores
gonna have a hell of a lot more trouble
with that than something like a 4 core
system agent 1.3 IO 1.3 VM 1.5 so the
only thing to really be careful of would
be increasing your SOC and si voltages
and IO voltages because those are easy
ones to like yes you can push to one
white 4 and it'll probably stabilize a
whole lot of stuff for you but the
trouble is there will be long-term
degradation of IMC so even if it makes
your clock stable and everything looks
ok thermals are ok whatever what you
don't know is that in a couple of months
time the integrated memory controller is
going to take a hell of a lot of a
beating and the frequencies that are
stable today at the voltages you have
today might not be stable anymore at
those voltage
in the future you might have to increase
the voltage and that's just gonna create
just entropy it's just it will continue
to get worse at that point or of course
you might have decreased the frequency
and the voltage to stabilize for for the
new degradation so those are numbers
hopefully that helps with those you have
a question should I just buy better
memory the vendor to some extent matters
like Samsung B died B died and II died
ICS are generally going to be the
easiest to work with and generally will
withstand higher voltages it's not a
hard rule it's just a general rule and
as far as like should you buy coarser
g.skill whatever they all use different
die different kits from different
suppliers so there's high annex in
samsung and to some extent micron so
just buying corsair g.skill won't
necessarily mean you get the b diary die
that you want if you want that you'll
have to check reviews online see if
there's any guaranty that the specific
memory kit you're looking at comes with
B diary die or whatever it is you want
and if it definitely does then obviously
buy that one to apply cost it more for
example G skills tried NZ kits
3600 megahertz as far as I can recall
are supposed to be P die basically
always but sometimes the vendors mix and
match so just make sure you do some
research on it it I would have to do the
same to answer your question at this
point because I don't keep all the
memory kits and their memory solutions
in my head all the time but those would
be the things that look for hopefully
that helps next comment I guess is a
shady Brady actually I'm glad you asked
that I don't think it asked anything but
I was quite annoyed when you and others
published the videos about X 219 I and
how the motherboard VRMs were
overheating etc there's some of the
worst press I've ever seen out of all
the X denying coverage specifically
because everybody forgot to mention the
VR arms were overheating using a single
software test incorrect but I'll get to
that moment that is patently false under
no circumstances would any was anybody
else able to heat up the V arms like a
264 was not what we used even in the
single software testing you're referring
to that's not what anybody was using
in a 64 it's not even that good of a
stress test continues on to say not even
5 gigahertz overclocks playing games and
rendering videos at the same time at the
same time dare borer or whatever his now
that's super disrespectful whatever his
name is is one of those guys who makes
extraordinary claims just to show how
nerdy he is I take everything that dude
says with a grain of salt by the way I'm
not saying you shouldn't run these types
of tests bla bla bla bla I know this
okay alright I've read enough first of
all there bauer is an expert at what he
says and has a professional overclocking
engineer basically that'd be hard to
describe his title he knows what he's
talking about
secondly we ran similar tests there
Barrett so this is probably referring to
this horribly horribly ill-informed
common is probably referring to when
dare Bauer was running tests I think he
published a video called the x29 vrm
disaster we did some follow-up testing
with the Asus Rampage 6 extreme we did a
lot of testing on the EVGA x2 night on
dark and I think that probably was what
we focused on there might have been one
other board but we found mostly the same
things as eated the rampage 6 extreme
was not that impressive for dealing with
vrm he and spoiler alert it doesn't
require one piece of software and
certainly not ADA to create the amount
of heat load on the VRMs you can do it
with real software in fact we ran
basically all of our so d98 exe Thermal
tests with blender
especially when overclocked why is that
because prime95 had some kind of issues
with the frequencies we were pushing the
voltages we were pushing and I think
there was just OCP that was getting
tripped on one of the boards but either
way a couple things to note here
so for VR purposes of erm thermals what
you need to know is how much power is
going or how much current is going to
the board so you can clamp the EPS
12-volt power connectors with a current
clamp and see what kind of current is
going down those lines into the
motherboard
we know the voltage it's probably 12 I
think our Paris light we tested with is
twelve point three volts so twelve point
three volts times whatever 50 amps is
not unreasonable Forex 299 and you got a
hell of a lot of power so I hear over
500 watts in that scenario and in our
power testing for which we have many
many charts we found that in fact
blender with I nine CPUs is capable of
pushing more than 500 watts so you can
push 50 amps down the EPS 12 volt
connectors with I mean you said not even
5 gigahertz which is an insane number to
pull out of your ass because if you're
talking about something like a 79 80 Hz
which is what most the coverage was
talking about that's a really high clock
so that's kind of pushing your your
exaggeration a bit into the realm of
complete not being believable alright so
here's the thing a lot of our testing is
blender blender is not a synthetic
software solution in fact our in-house
our main video editor and also the guy
who does like designs for our merch our
clothes so this logo was done in blender
the mod mat was designed in blender
technically it's all 3d but we just make
it just flattened it orthographic is the
word I'm looking for
this 3d cube blender that's why it's 3d
because blender is a 3d modeling
solution also animation so blender is a
real software solution that real people
use every day and blender also uses
things like a VX which is not just a
prime95 thing and it's not just a stress
test instruction set it's like actually
useful and blender can create the
unreasonable heat and thermal loads we
were seeing on some of the X 299 V RMS
same for de Brouwer dere Bauer did do a
lot of testing with stress software but
at the end of the day all that matters
is that you're pushing the same current
into the vrm same wattage as your
realistic software so it doesn't matter
if prime95 is pushing 50 amps into the
or into the into the board from the EPS
12 volt cable and blenders pushing 50
amps into this and the board via the EPS
12 volt cables doesn't really matter
doesn't matter how that power is getting
created the fact is it's drawing the
same in this scenario and blender uses a
VX by the way so yeah it's actually a
thing that's not just ease for stress
testing so the point is this comet is
horribly horribly informed and we were
able to replicate the vrm thermal issues
with real software this is old news at
this point it's several months old I
don't even remember our own coverage
that well but I remember it enough to
know that Prime gave us issues crashing
with the 79 80 XE a lot of the time and
so we end up using blender for all of
our D lid tests using the D letter back
there all the liquid metal tests the
repel TT stream we use blender to
stabilize that before we went on the
stream so we can figure out what kind of
our baseline clock was going to be so
blender is a great software solution
that does actual things that
professionals use to make movies and
also by the way is a very good stress
tester and that's typically what we use
so this I just wanted to point this out
cuz this comment was just so off base
and the tone of it was absurd to me so I
had to point out how incorrect it was
but yeah I mean there's been a vrm
thermal problem for a while now I'm glad
that you think yours is fine when you're
running your taps at 105 see for the
whole life of the system let me know and
about I don't know a year a year and a
half when it blows up and obviously
better vrm designs have been coming out
for a while now and some large part of
that is because people have picked up on
it's like gigabytes X 470 board with an
actual heatsink that wasn't a mistake
that's because people were really
complaining and the tech marketing
people over there thank you you know who
you are pushed internally for
improvements in that department even
though it wasn't actually needed on that
bored but that's kind of you know it
matters to talk up about things that are
designed in a shitty fashion because
they actually get fixed and at the end
of the day powers power and if you're
pulling a lot of power with any i9 then
it doesn't matter how the power is
created for the most part unless you're
testing CPU thermals in which case it
absolutely matters that's created
because AV experts not a VX will have a
different impact to your CP thermals but
as far as the fear is concerned that's
what we're talking about right now
blender really stressed it so next one
kiss Stowe says how would you choose a
video card for water cooling would you
go for the cheapest variant the founders
edition or a card that has a higher
clock speed already like a board partner
card great question so we previously we
talked about the ten atti armor which is
an MSI card and the 1080i armor has one
of the worst coolers ever outfitted do a
card of its power consumption but it's
got a really good PCB so 1080i armor
used as a gaming X PCB which was one of
the higher end ones originally the armor
was about 700 and the gaming axe was
about 750 poor went out for the GP
prizes I guess but the point was you
could get an armor and basically have a
gaming ax and then just pull the cooler
off of it don't use it with that cooler
put a liquid cooler on it and you have a
really good liquid cooled card that was
actually relatively cheap cheaper then
about the same price as a founders
Edition card at the time that this was
all launched so assuming the price
equivalence it was a crazy good deal
because vrm is better gaming X PCBs
great and and then you put your own
cooler on Sakhalin solutions totally
irrelevant founders edition cards or
reference cards in the case of Vega are
also just fine
the thing with water cooling is the
reference designs have the most blocks
for them so if you want some variety
those are going to be easier to shop for
in terms of your liquid cooling solution
if you have another card that you like
and I'll give you an example that I gave
you one the game index PCB you might
like the VR I'm a lot will it be better
than the founders edition of
overclocking it's hard to say a lot of
overclock
with especially with nvidia gpus comes
down to the silicon quality and comes
down to the throne performance which
you're putting a liquid cooler on it so
that's already taken care of but the
well I guess if it's a full-coverage
block technically if you have a better
vrm that spreads out the heat load more
then and it's a larger PCB like the
gaming x1 so you've got more surface
area of copper covering it for a full
coverage block then technically the the
the fact that you are sharing your GPU
cooling solution with your vrm coin
solution it will be advantaged
advantageous to have a better vrm
because you're spreading the heat out
over a wider area theoretically there
should be less less sort of thermal
density and maybe it'll perform a bit
better but basically what it comes down
to though with NVIDIA GPUs is silicon
quality which is why you see some
founders additions clock and better than
really high-end cards and also the
thermal performance so every 5 degrees
it gets you meaningful gains in
frequency you can gain 12 mega Hertz at
a time all the way down to like 20 C as
long as I've tested it anyway so if you
go down to say 60 degrees from 65 there
will be a meaningful increase and
frequency the point of saying that is
that you put you're putting a liquid
cooler on it regardless so that's going
to take care of most of your frequency
limitations before anything else as far
as power yeah they're better VRS than
others but the gaming axe is one of the
better ones and I just like it because
you can buy an armor and get that so
it's a really cheap way to do it if
there is not a good water cooling
solution for that specific PCB because
it is oversized reference size cards are
fine there are non founders Edition
reference sized cards but you still need
to make sure that the vrm components are
all clear in terms of like clearanced
and tolerance for how thick the block is
mounted so something like the SCT or an
IC X card you might want to buy one of
those if you don't want founders Edition
because you can get a Founders Edition
PCB basically with an
see card and which is an EVGA card and
you get icx sensors in it so there's not
a whole lot of point to this in
performance terms it's just if you want
some thermistors and your cards you can
check the vram performance of erm
thermal performance sort of adjacent for
backside GPU performance you can get
those with an IX card it's really more
of one of those things that you play
with because it's fun to do not
necessarily getting a lot more
performance but I would be completely
fine recommending a reference card for
water cooling because it's just so much
easier to work with and it's cheaper to
get into or if you can do it
something like an armored card which a
bit of a hidden gem
another PCB is great and the only way to
really know that is watched like our PCB
teardown analysis videos or build Zoids
same thing in one of the same and but
when I say that I mean his own channel
actually hardcore overclocking has a lot
more but he does contract them for us
that'd be the best place to start just
make sure you get a block for it next
one soup can man says now that rise ins
main architects lots for Intel how long
do you think Andy will stay competitive
with Intel I think they're fine so Jim
Keller hasn't even been at AMD for quite
a while like at least a year not a
number of years hasn't been at AMD for a
while architectures are in the pipe for
a long time obviously he put in a lot of
his time before he left but there are a
lot of really talented people on that
team I can name a couple of corporate
fellows who are very talented in the
power department I know a couple folks
like Mike Clark talented in the being
the chief architect Department so I
think they're fine
Jim Keller obviously brought a lot of
talent to the table but working with
that team I think AMD CPU division now
is pretty much up to speed on what they
need to do to make a good product I
think they have a really good
architecture and roadmap planned out
they seem to have gone with a multi-chip
approach and it's working out
excellently for them and now that
they've done that I'm I feel relatively
confident knowing the people I know at
AMD in the engineering side that they'll
be fine without Jim Keller so I don't
think I don't think it's like there's
any need for panic
where people are like first of all Jim
cut and just not leave but I don't think
there's need for panic where it's like
he left and suddenly Andy's screwed he's
not the only talented person there
there's CBO divisions done really well
lately so I think they'll be fine
Intel absolutely needs the help with 10
nanometre so it's good that he's going
there if Intel's gonna get anywhere with
with their future designs next also he's
a bit of a mercenary Jim Keller it's
totally cool the guy's a brilliant
architect and he's worked at least
companies on and off for his whole
career so it's not unreasonable or
uncommon for someone with that talent to
shift around between companies as they
take on new challenges and obviously get
pay raises and things like that and the
satisfaction of confident an engineering
challenge is this the last one I think
this last one before the next episode
metal shark says how do you test the
maximum safe amperage of a custom-made
power supply cable preferably without
expensive equipment such as desk bench
power supply configurable load thermal
imaging camera etc the easiest way is to
not test it and look up the wire gauge
so you can check the if like cable mod
or someone if they don't specify the
gauge of the wire you can hopefully
figure it out either by peeling back
some less leaving and hopefully it's
labeled and if not you can measure it
there are you just buy most wires triple
strippers will have something to help
you measure the wire gauge so check the
wire gauge and then look up there's a
table online that will tell you what
kind of current learn to expect for each
gauge like 18 versus 16 gauge for
instance and you should be good to go
that would be the easiest way to do it
without needing to melt anything and and
sacrifice cables so that's it for this
one as always go to store it on
Cameron's nexus net to pick one of our
3d teardown Hughes which is made in
blender which is a real software
solution that's great for thermal
testing things and finding vrm at
thermal limitations got a little carried
away but you can pick it up on stored of
cameras nexus net it's super cool for
our mod mat and also patreon.com such
gamers axis to get access to the bonus
episode of ask the yen and check back
subscribe
more I'll see you all next time
you
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.