Ask GN 60: More Bones to Pick About CPU Thermals & Boards
Ask GN 60: More Bones to Pick About CPU Thermals & Boards
2017-10-12
hey Ron welcome back to another episode
of ask Jan I think we're on our 60th
episode now I think which in the world
of what we do is a special number
because 60 stands for 60 fps of course
so because of that we are not doing
anything special to celebrate but thank
you for joining if you have a question
for next episode leave it in the comment
section below
got a couple follow-up ones again on
coffee like have some content on gpu-z
which I'm trying to get answers on a few
popular questions there and some other
stuff on just cooling and electricity
and things like that so lots to go
through today this video is brought to
you by the be quiet dark based Pro 900
white edition the DBP 900 marks a return
to full tower cases equipped with ample
harddrive support effective at noise
damping foam high performance fans and
the option to be inverted into an
alternative layout the tinted tempered
glass window and Qi charger at a high
untouched to an already well-built case
learn more at the link in the
description below so let's start with
this episode with going through a list
of beefs and grievances user crimson
posted like three comments now on
different videos saying including the
review of the 8700 K saying did you why
or actually it wasn't even did you test
before deleting it was like why did you
test after delayed in which we need to
have a conversation about how to phrase
a question I think but ignoring that as
I answered the first time to the
question no we did all the testing on
the 87 or K stock before deleting except
for the deleted thermals obviously so I
wanted to address that in a video
because it keeps getting posted and the
user is obviously not reading my account
replies which is fine YouTube is filled
with comments but putting it here
hopefully will answer that question for
other people who may have had it we
always test those things stock out of
box before any mods and then the mods
are tested within the realm of the mods
so dewetting was done and the thermals
and dilating section after everything
else was done obviously now part of that
is because I mean one it's out of box
but two we don't want to destroy the
sample before we
through the review so yeah hopefully
that question doesn't come up again but
this next one is there's a two more here
so this was on the content where I was
talking about it was the news video
where I was talking about how thermal
testing works and how you can't just say
like so-and-so got 92 and this person
got 68 and it's like it's not a score it
depends on everything so yeah back to
that stuff this this one
Jordan malechy Oh said always prefer
out-of-the-box animals in all caps to
reflect what the general public will
experience in setting up their systems I
agree with the concept however a couple
things here
what does out-of-the-box things mean so
with regard specifically to the CPU I
would agree that out of the box means
you're not deleting it and putting
liquid metal on so we could establish
that if the CPU comes with a stock
cooler then you could also test with the
stock cooler and that would be pretty
valuable information the 8700 K doesn't
come with the stock cooler so the next
thing you're looking at is what cooler
do you use now for us we use one cooler
for all of the thermal testing so that
we have comparative data if we wanted to
present it as comparative data CP to CPU
this gets really difficult to do though
just total aside here but comparing CPU
thermal data between architectures is
just it's it's basically impossible to
do it accurately because you don't know
if the sensors on CPU a are different
from the sensors on CPU B they could be
in different places they can measure
different things like the average
different ways we don't know that until
name they don't share that information
even comparing to AMD versus Intel
thermals is basically a fallacy can't
really do that you can however compare
power consumption which is why power
consumption is so important because for
lack of a trustworthy thermal metric
because no software outputs a thermal
metric and says this is identical on AMD
and Intel because that's not how it
works for lack of that we can't a power
consumption at the EPS 12-volt rails and
you can more or less figure out the
now put because power and temperature go
hand-in-hand I mean he is energy so
that's why the power consumption charge
is so important you there are ways to
compare the two I'm not going to talk
about them today we've been looking at
some of them but none of them involve
software because it's just not gonna
work so that was a complete aside as far
as like comparing CP temperatures across
architecture things like that but the
next part so this guy's comment of I
prefer out of box thermals to reflect
what the public will experience the next
part of that is let's say we all agree
on a CPU cooler maybe maybe for low end
CPUs we all agree that a hyper 212 or an
analogue to the hyper 212 would
represent a low end user build in terms
of cooling options so we just kind of
assume we all agree on that
the next part though is what software do
you use to burn the CPU in and this is
why we use Prime and blender because
prime95 there's a lot of different
versions of it like 30 of them but they
all kind of do slightly different things
and all of them in one way or another
will torture the CPU in ways that either
will or won't ever be experienced by
user for gamers all of the time Prime is
kind of absurd kind of like fur mark is
however blender also serves as a test
that's somewhat real-world and
represents how the CPU can be used in a
non synthetic application now prime of
course represents compute workloads to
some extent so it is kind of a realistic
application but it's not realistic
enough for most of our audience hence
the edition of blender you get into
things like games those don't really
work for thermal tasks for a number of
reasons but blender is a pretty good one
because blenders it's really consistent
it maintains 100% load non-stop for the
most part and it's super easy to
replicate it does the same thing every
time so it's what you want without being
prime which is also crazy good at what
it does but it's it's got a lot of
difficulties in there in terms of
benchmarking and it's not always
representative so for out-of-the-box
thermals reflecting what the public
experience is
the question is again what does the
public the public expect to see there
and so that's why we do blender and
prime prime gives you a complete torture
scenario where you know this is one of
the worst possible cases for the CPU it
should not get any worse than this and
you want to know that because you want
to know under a worst-case scenario does
it throttle or experience some kind of
bad behavior because if it does that's
clearly a flaw by the company's
manufacturing or engineering even if
it's not common to be experienced you
need to control for those worst case
scenarios and then you have blender for
a real world because the next question
of course is does the CPU function as
expected in a real intensive scenario
like blender again so that's why we do
those and the point of saying all of
this is to highlight that once again
like it did in the news video
temperatures and thermal testing it's
it's a lot more involved than just
turning on hardware info and telling
it's a log because you really need to
understand what's going on another good
example is prime so you can use version
twenty six point six twenty nine point
two twenty eight point five any of those
twenty six point six is non AVX
and that's important so if you use
twenty nine point two or twenty-eight
point five with AVX workloads and you
start spinning off let's say typer
thread it to a VX threads per core
that's going to generate a hell of a lot
more heat then twenty six point six with
no AVX so again it's important to
understand what applications being used
to test and that's why you can't compare
thermal data across site without both of
them doing the same thing with the same
cooler with same frequency with the same
voltage with the same ambient
temperature and so on so yeah thermal is
for our testing our numbers for
temperatures are meant to be compared to
our other numbers for temperature unless
otherwise noted it's not always
conducted the same way we kind of leave
it up to you to read the chart title and
make that that decision of is this the
same as the other one
because a lot of time they're not but we
have all that information on the chart
title so you can tell so hopefully that
helps some of it now
next thing so the this was a similar
comment that I want I'm addressing these
now because thermals recently have
become out of nowhere more interesting
to a lot of you we've been covering
thermals for like years now and no one's
cared but suddenly people care so time
to talk a bit more about how it all
works but this college from Allen
Tinsley who said thermal should be
consistent across the board 30 degrees
Celsius is huge this is referencing an
hour
I don't know 70 vs. lioness is 92 or
whatever so 30 C is huge thermal should
be consistent across the board heat sink
is irrelevant when pro review is
performed the same benchmark since all
respectful reviewers use the best they
have not sure what that means
gamers Nexus lost a lot of street cred
with this video okay so street cred
aside which is something that I've
always strived for I've never wanted
anything more than than to be a street
so um that's a GN hat by the way
by one hahaha so yeah this comment no
just no heatsink the heatsink is not
irrelevant why do all of these heatsink
companies exist if it's irrelevant so
okay if I put a hyper 212 on one GPU and
then I put the other one under like
let's not even go crazy let's just say
that knock to a ninety dollar air cooler
that's gonna be a huge difference like
when you look at a CPU cooler review
right we published CPU cooler reviews
you look at one of those and you'll
notice the temperatures different for
basically all the CPU coolers why would
the temperature be different because it
matters which one you use so that's a
huge point in what kind of temperature
output you get and if you use a
lower-end cooler with a thermal test you
start running into throttling scenarios
which is very important data to have
it's important to tell users
hey the CPU cooler is inadequate to cool
the CPU under these conditions however
it also means that you don't actually
know how bad the thermals are because
you're throttling so early in the
process that you don't get that
comparative day
because it's it stops itself from
heating up further so yeah again both
sets of data can be and probably are
accurate within their parameters but
know it thermal performance is not
consistent across the board of reviewers
because everyone uses a different piece
of software and hardware to cool the CPU
even just putting the thin in a case
versus not can I have a big difference
and as far as like the one argument you
could make is thermal performance for
the most part should be fairly
consistent out of box with a stock
cooler if the CPU includes own with
which the 8700 K does not with a stock
cooler with the same motherboard with
the same efi with the same voltage and
frequency tables with the same piece of
software and with the same piece of
monitoring software all those things
bundled together their whole performance
should be consistent I agree
but that's an awful lot of things to be
consistent between tests and we don't
have a big cabal of reviewers that
gathers annually to determine who uses
what so it really like I I know it's
more work but you really just have to
look at all the different testing
approaches and find the one that fits
your scenario does it to pull the data
from it that matters most to you that's
all there is to it so yeah hopefully
that kind of addresses that one I'm
gonna go a bit further here and refer to
our previous motherboard differences
content 4x 299 we have some charts we
can throw them on screen briefly I'm not
going to talk through them I don't have
the data in my head but you can check
the video basically just switching
motherboards can impact temperatures
massively I'm talking like 20-plus
degrees Celsius we saw this with the
7700 K originally when gigabyte had
their gaming 7 board that we now use in
bike actually Izzie 270 board but at a
time it had outrageous Auto voltage
settings where it was blasting the CPU I
don't remember I think it was one point
three five to one point four volts like
all the time
and they fixed it but doing that means
that the temperature of that CPU on that
motherboard is going to be way higher
than on let's say a gigabyte or an asus
board that has an auto voltage of maybe
1.25 and actually to that point I do
remember these numbers I think gigabytes
before the fix was something like one
point three five to one point four volts
at the same frequency mind you and the
other board we tested we were able to
get stable on an MSI board at like one
point one eight
drop one point three five one point one
eight volts is massive that means your
power consumption drops a lot and that
means that your thermals drop like
20-plus degrees Celsius in some tests so
just again to reiterate the motherboard
used matters and this bleeds into one of
the actually really good suggestions or
questions from the from the last ask G
and let me see if I can find it's out of
order here this one was from Nelson
Fernandez de Lima who said Steve it is
starting to make more sense instead of
only doing CPU benches start to do also
motherboard vendor benches this thing of
some then this so this topic of some
vendors having differences in the
thermals because of the voltage you can
figure are not the same as the users
obviously not not english-speaking for a
sangwich but I understand what you mean
I think most people do so if I
understand this question correctly its
Steve instead of us just CPU review
insert CPU name review we need to better
take a better note to the motherboard
used because the motherboard impacts
thermals because the voltage configured
is is not the same across boards that's
more or less what I'm getting from it
and I agree so yeah this is where we've
been doing actually a lot more of these
so we had the X 299 motherboard
differences impact on thermals video and
previously we had one for Z 270 with the
game and seven so we're doing it more in
trying to raise some awareness to this
but yeah basically Auto voltage and
frequency across all these boards can
vary a bit frequency not so much of they
comply with Intel spec but voltage for
sure
so for example I don't know if I can
publicly say who it was but one of the
board vendors I spoke with recently was
saying that the reason some of their
voltage readings that we saw were higher
than their competition by about I think
it was like a hundred millivolts or
something it was it was a decent amount
higher so they were saying the reason
that there actually was it was less than
that it was like yeah anyway smell is 60
millivolts so the reason that their
voltage readings in our test were 60
millivolts higher than a competitor's
voltage readings was because they found
in a specific application Adobe Premiere
actually the CPU with a multi-core
enhancement feature enabled would become
unstable and so what would happen is if
you enable MCE
or whatever it's called in that
particular board the CPU would boost to
its maximum single core turbo except
across all cores and it was stable for
the most part but every now and then you
hit an application that's a little weird
blender is one of them and apparently
Adobe Premiere is too and so the
motherboard vendor has no control to
target Adobe Premiere and say hey when
this application launches increase
voltage by 60 millivolts all they can do
is say increase voltage by 60 millivolts
period all the time or it can figure a
table for it but that's what you get and
so leading back to Nelson's question or
suggestion I yes I mean the differences
between motherboards can be tremendous
especially early in a launch cycle now
they they eventually kind of even out
over a few months sometimes it takes a
few of them a year to get everything
correct but they kind of give an out at
launch though please always keep them
and we'll do our best to kind of mention
this in a very short version every time
but keep in mind that when you're
looking at temperature and power numbers
specifically those two they're gonna be
pretty big differences board to board
and in our Exton iodine test recently we
saw that one of the Giga by EF I is
pulling I think about 70 watts more than
one of the competing boards I want to
say in a Susie EFI
that's huge 70 watt increase you know
you go from like 190 or whatever it was
to 260 that's a pretty big jump that's
enough to surpass some competing chips
so it comes down to like who looks bad
there is it the CPU maker is it AMD or
Intel or is the motherboard maker it
should probably be the motherboard maker
but it's hard to always make that
distinction
especially when it's early in a launch
cycle and because there's so there's so
little understanding I guess in general
of how much the motherboard impacts
these things it pretty much always goes
back to either AMD or Intel being at
fault which can certainly be true but a
lot of the time it is the board vendor
and it gets bit muddy because you could
say well the board vendor was waiting
for specs from the CPU vendor so it's
really the CPU vendors fault ultimately
like Intel and the x-29 instance of
pushing launch forward a couple months
from August to whatever it was that's
partly on Intel but and AMD the issue is
that with rise ins launch which were
just about as many as xc-99 a lot of the
issues those board vendors had or
because AMD gave them the final
microcode two weeks before launch so
it's kind of messy everyone's kind of at
fault to some extent but the point is as
viewers and readers you have a
responsibility to yourself to understand
that there are differences between
boards and when you look at review
numbers between review sites just look
at the number figure out what board they
use go look at the other number to
figure out what board they use and then
rather than saying these two sites are
in conflict with their data they're
either both wrong or one of them is
wrong instead look for other data that
corroborates that by using the same
boards and figure out which board is is
less desirable so a pretty long section
there but let's go through some of the
others this one is from well this one I
actually came with with photos this from
RJ Flemming 82 on the patreon backer
discord if you want to join that you can
go to patreon.com/scishow and axis and
join us there so RJ fleming asked
question on air cooling sli as you can
see i
have the 120-millimeter Noctua on top of
the GPU I've seen GN and other channels
put a fan over an SLI config for
thermals most times pushing air down
with two inlets in the rear
I've made this nock to a fan pull air
from between the card so it's pulling
out my thermal seem better this way
is there a reason GN prefers pushing air
down on the card or a fan of this config
yes so we are pretty much always doing
those kinds of tests either an open air
or in an open like big case and
generally speaking for our setups
particularly with open air all that
really matters is that you feed as much
cold air to it as you can and then
you're trusting if it's a blower card
you're trusting that it just exhausted
all out the back and if it's not then
you're basically just going for pushes
much cold air and as possible and then
the rest of it will just by nature of
all this airflow going one direction all
the linear feet per minute or whatever
airflow it'll just kind of find its way
out the back of the system and this gets
to a point where when we do some more
extreme really like hot thermal burn
ends I think we did with oh when I had
Vega set up to draw like 4 or 500 watts
of power when we do that if I'm not
doing a thermal measurement test and I
just want to get the hot air out of the
room so that I can keep benching and
testing games and stuff I'll actually
set it up so that the the bench is kind
of shifted so it's facing the exhaust
out of the window open the window and
then set up some fans case fans near the
window that just push air out so that's
that's the part you don't see if I'm
pushing air down onto the card I
probably have something else behind it
off the camera off the out of the photo
that's getting it out of the room
because that's when we do live streams
like in this setup right here in which
case all that air is normally going on
to my feet or face so that's that's
where the heat goes in those instances
one more note that's related to both
this and the previous set of questions I
mention all the stuff about motherboards
ambient also matters so keep that in
mind we log ambient second a second with
thermocouples and that allows us to make
sure it's consistent or if it's not make
adjustments or modifiers as necessary
but that's just another point of
consideration is ambient ambient room
temperature
next question liquid paper from discord
asked Steve can you talk a bit about the
temperature effects of different
mounting locations for a radiator in an
AIO setup I think Jay did a small piece
on this a while ago and found that a
front-mounted radiator is best for CV
attempts I understand that a various
case to case what temperature's would
someone generally see you mounting on
the top exhaust versus intake we've done
some of these tests in the past there is
no one-size-fits-all answer cases are
the least scientific tests you can do
because every case will behave
differently with every every
configuration some cases for example if
they have the clothes off front panel
like the s340 elite they're gonna have
worse their own performance of the front
mounted radiator then obviously case
with a mesh or in the same case like
that's through 40 elite we actually did
some testing on that one with a 140
radiator mounted in the the front the
top the rear as part of that NZXT blog I
was talking about last episode and I
don't exactly recall all the data
offhand but you kind of have to start
measuring a lot of things like if you're
gonna get really serious about it you
have to look at things like the GPU
temperature as well because
front-mounted might be better in some
cases for the CPU like it with a mesh
front especially but you have two
considerations with a CLC you might be
forced to have your tubes up which is
suboptimal you want them down and the
second consideration is now you're
taking all that heat from the CPU and
you're dumping it into the graphics card
so do you have a means for the GPU to
deal with that heat and does it require
increasing GPU fan speed to a point
where your case is now louder and it'll
be louder anyway from having the cooler
in the front where there's a mesh anyway
because you have more noise escaping
from the front so there is no
one-size-fits-all answer I will say what
I have found generally speaking is that
I prefer to
Mountain the top if I can for that
because if it's in the front the the
tube orientation things are concerned
but also in the front mounted position I
just I don't like that heat going into
the case so my preference is because GPS
are way more sensitive to thermals than
CPUs these days my preference is keep
the CPU reasonably cool and if you have
a CLC that's not going to be a problem
and then get the GPU as cool as as
possible in that case it's cool is
mechanically possible so I like to do
top mounted but it's not always better
depends on the case and sometimes top
mounted intake is something that you
want we tested this in the coop like
Hales ero for years and years ago we're
a top mounted intake both helped CPU
thermals with a tower cooler and
significantly lowered the vrm
temperatures around the CPU which is a
concern with some be 350 boards and
other lower end boards that have hot be
RMS so keep all that in mind but yeah
sorry there's no simple answer we have
generally found that with the right mesh
front it is cooler in the front but
again Vega and Pascal are both so hyper
sensitive to temperature that you're
going to notice more from a 5c increase
there then you will from a 5c increase
in the CPU and the CPU just it really
doesn't care it won't throttle until it
hits t.j.maxx let's do a let's see let's
do one more with a note the note first
it's Raymond Kwok asked I think on
YouTube you have two questions asked
this on YouTube do you know what the
hotspot temperature measurement in gpu-z
actually measures there's speculation on
forums no I don't I sent a message with
that question to the guy who makes DBZ
from tech power up and hopefully he'll
respond but if he if he knows and
responds I'll answer it next week next
the last question Dennis Nedry asks how
much electricity our users wasting
leaving their gaming rigs running idle
games going 24/7 could you maybe sure
how much kilowatt hour usage is in a
common gaming bill in the last two to
three years we show power consumption
tests and most of our
use these days so you can get those
numbers there but let's say idle without
games depending on on what power
settings you have in Windows and SSD and
what you allow to sleep and stuff like
that it can be let's just assume kind of
a higher performance profile where
you're not really letting it properly go
to sleep you're still looking at at the
low end like 50 watts and at the high
end my system idles at 200 watts it's an
older FX series system it's not
configured very well and it gets hot so
50 to 200 watt range pretty big but
let's just say you're about 70 watts cuz
that's mostly what we see in our high
performance profile idle test with most
GPUs and CPUs if you're there it's not a
ton of power but you can take that
number and post and go find a kilowatt
hour electricity cost calculator put in
the cost for your region I don't know
what it is and calculate for the number
of watts it's gonna be pretty damn low
for idle idling in games isn't really
idle I know what you mean you mean the
users walked away but if the the system
is running a game that can be that I
mean that's a big range that's like 100
watts to a thousand watts or more just
depends on their configuration but let's
take kind of a safe configuration that
maybe is reasonable from other people to
have let's call 300 watts 300 watts to
burn 24/7 adds up fast and I mean we can
actually I'll just do a quick
calculation of this assuming a I'm going
to assume a 10 to 12 cent per kilowatt
hour cost because that's about what it
costs us for electricity and let's put
that in okay so I'm using rapid tables
here I just came out first so let's say
power consumption is 300 watts for some
game with maybe 10 seventy or vaga 56
well no not that ten seventy or five
seventy type of hardware or something
like that that's not being stressed too
hard and let's say it's being used 24
hours a day and let's say your kilowatt
hour cost is 12 cents which i think is
the US average that comes out to $26 per
month 316 dollars per year in terms of
cost now of course there's a question of
should you be needlessly burning
electricity if it is truly needless then
probably not but there's your answer
it's it's pretty hard to make
electricity cost a lot and 300 watts is
a middle-of-the-road estimate so you can
put the number to get more accurate but
still 316 bucks a year when it's not
needed that's pretty good savings so
hopefully that gives you an idea last
question was JL Rockefeller gamers Nexus
why does your logo look like a pokemon
ball I believe they're called pokeballs
thank you very much come on okay all
right thank you for watching leave
questions below if you have them
subscribe for more go to
patreon.com/scishow Gary's X's tell us
that directly or if you want to be one
of the questions from here that was from
discord chat and finally if you can't do
patreon but you want to help elsewhere
store a gamer's access net we have
shirts like whatever I'm wearing right
now I can't look at it because then
it'll blow out the mic audio so
subscribe more I'll see you all next
time
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.