00:00:00
- Why would anyone run an app
that is specifically designed
00:00:03
to crash their computer?
(bomb exploding)
00:00:05
That's a good question.
00:00:07
And I have some pretty good answers.
00:00:09
Stress testing applications
can make your computer overheat
00:00:12
or even outright crash, but in doing so,
00:00:15
they can also reveal weak
points in your system
00:00:18
that could have led to bigger problems,
00:00:20
like instability or even data loss.
00:00:23
But of all the options out there
00:00:24
for CPU stress or burn applications,
00:00:27
which one is actually going
to hit your CPU the hardest?
00:00:31
The answer, spoiler alert, is Prime95
00:00:34
at least in terms of heat.
00:00:36
But the whole story is a lot
more complicated than that.
00:00:39
And we just couldn't tell it
without our sponsor, CableMod.
00:00:43
CableMod allows you to
personalize the look of your PC
00:00:46
with custom colored sleeved cables.
00:00:48
Try out their configurator
00:00:49
to build your cables
exactly how you want them
00:00:51
with their realistic cable preview.
00:00:52
We're gonna have the link down below.
00:00:54
Like test driving a vehicle,
00:00:55
stress testing is an important part
00:00:57
of savvy PC shopping and maintenance,
00:01:00
but, let's say you have
never done it before.
00:01:02
You head on over to Google,
smash in some keywords and,
00:01:06
oh, that's a lot of options.
00:01:09
You got your Blender, Cinebench
R20 and Cinebench R23,
00:01:12
OCCT, Prime95, CPU
expert, HeavyLoad, Linux,
00:01:15
Intel XTU, Ryzen Master, IntelBurnTest,
00:01:17
AIDA64, Linpack Xtreme. (gasps)
00:01:19
And that's all just for the CPU.
00:01:22
The unfortunate reality is that
00:01:24
there is no silver bullet best test,
00:01:27
and depending on your hardware, software
00:01:29
and what you're trying to accomplish,
00:01:31
you could end up with
completely different answers.
00:01:34
So yeah, we spoiled the conclusion.
00:01:37
Prime95 runs the hottest, but even then,
00:01:40
which of the 5 tests, 10, if
you include the AVX2 checkbox,
00:01:44
should you run?
00:01:46
To find out, we built up two identical PCs
00:01:49
with just two differences,
motherboard and CPU.
00:01:52
Here's the thing, we
suspected that team blue
00:01:55
and team red might behave differently
00:01:57
under extreme conditions
00:01:58
and we're really glad we tested both
00:02:01
because they absolutely did.
00:02:03
On both platforms though,
00:02:04
we also locked our CPU clock speeds,
00:02:07
core voltages and fan speeds,
00:02:09
leaving just our CPU temperatures
as our stress indicator.
00:02:13
Time to show them the graphs.
00:02:15
- [Colin] Give them the graph, editor.
00:02:17
- Intel first, two tests
didn't work at all.
00:02:20
Ryzen Master, we saw that one coming.
00:02:23
And Intel Extreme Tuning Utility.
00:02:25
We didn't see that one coming.
00:02:26
It ended up ever ring out
the same way in Windows 11
00:02:29
that we saw in our recent
Dragon Canyon NUC video.
00:02:32
Among the tests that did work,
00:02:34
Prime95 small FFT stood absolutely alone,
00:02:38
beating out the next highest
score by 10 freaking degrees.
00:02:43
There just no other way to put this.
00:02:46
We did not believe our results.
00:02:49
So thinking that maybe Intel's
turbo boost power management
00:02:53
was banging up the other
tests or something.
00:02:56
We reran the entire
suite with our voltages
00:02:59
and our clock speeds back to stock.
00:03:02
And to say I'm shocked
would be putting it mildly.
00:03:05
It was like witnessing a brutal
beat down in slow motion.
00:03:09
Again, we have some theories here,
00:03:12
but first, let's get to
the rest of the results.
00:03:15
Our 10 middle performers
00:03:17
had just a 6 degree overall spread.
00:03:20
Then we had our underperformers.
00:03:22
IntelBurnTest, CPU-Z
00:03:24
and OCCT's CPU large
dataset extreme benchmark.
00:03:29
At this point, you can probably see
00:03:30
why we declared Prime95 best in heat
00:03:33
but looking at thermals alone
would be a huge mistake.
00:03:37
Check out the heat profile
and power consumption profile
00:03:40
of Prime95 small FFT.
00:03:42
Huge spikes, then they
just hold indefinitely.
00:03:46
It achieves this by repeatedly performing
00:03:48
demanding mathematical calculations,
00:03:51
but on a data set that's so small,
00:03:53
it fits right in the CPU's on die cache.
00:03:56
No bottlenecks, maximum burke.
00:03:59
But obviously, no real world
application behaves like that.
00:04:02
Even notoriously demanding
professional workloads
00:04:06
like CPU Rendering and Blender
00:04:08
or video encoding will generate
very different profiles.
00:04:12
So if I was validating a thermal design
00:04:14
for a demanding client,
00:04:16
and I wanted to see an
absolute worst case scenario,
00:04:19
Prime95 needs to be part of my toolkit.
00:04:22
But if I care more about stability
00:04:24
in real world applications,
I need to look elsewhere.
00:04:27
Enter Linpack, the most used stability
00:04:29
stress testing tool of all time.
00:04:32
It was originally created
for use on supercomputers
00:04:34
in the 1970s.
00:04:36
And in a nutshell, it's a software library
00:04:38
that performs linear algebra,
00:04:40
this equation specifically, on repeat
00:04:43
and it still behaves the same way today,
00:04:45
except it's been rehashed multiple times
00:04:48
to keep up with modern computers,
00:04:49
which is pretty important considering
00:04:51
that it was written in Fortran,
00:04:52
a code language that is so old,
00:04:54
it was originally input via punch cards.
00:04:57
Fun fact, by the way,
00:04:58
because Linpack is a
library not an application,
00:05:02
it's easy for developers to integrate it
00:05:04
into their own test suites.
00:05:06
And six of the tests we used
are actually Linpack based,
00:05:10
but even then, not all tests are equal
00:05:13
because there are
multiple Linpack versions.
00:05:15
And any of those can be
combined with other methods
00:05:19
to create a complete hodge-podge of stress
00:05:22
for your poor CPU.
00:05:23
This is what a Linpack load looks like
00:05:26
on our thermal graphs.
00:05:28
A spikey wave gets produced
as each equation is solved,
00:05:31
then there's a brief pause
between calculations.
00:05:34
And now that we know
this, it's really easy
00:05:37
to spot this shape in many
of the tests that we ran.
00:05:40
Now, if we switch from
temperature to power consumption,
00:05:42
we see, well, the same thing.
00:05:45
These swings in power are really
tough on not just your CPU
00:05:49
but also your motherboard
VRMs and your power supply,
00:05:52
which is a perfect combination
for evaluating stability.
00:05:55
Which we're definitely gonna check out
00:05:57
on AMD's new 3D V-Cache CPUs.
00:06:00
So get subscribed so you don't miss it.
00:06:02
Also, buy a shirt or
a water bottle. Ha ha.
00:06:04
lttstore.com.
00:06:06
On that note, let's take
a look at our AMD results.
00:06:09
Here, we didn't see anything
that seemed anomalous.
00:06:11
So we left our voltages
and clock speeds locked
00:06:13
per our original test methodology.
00:06:16
To start with, AMD ran much
cooler overall with Delta Ts.
00:06:20
So that's the temperature difference
00:06:22
between ambient and our CPU,
peaking at just 39 degrees
00:06:26
compared to Intel's 56.
00:06:28
Man, 12th gen performs
but, Lord, is it ever hot?
00:06:33
Now, Prime95 still came
on top for thermal stress
00:06:36
but only barely, just half
a degree behind was AIDA64.
00:06:41
Then, the middle of our chart
00:06:42
doesn't have the same
flat middle zone as Intel.
00:06:45
Instead, there's a
fairly consistent decline
00:06:47
with a spread of about 10 degrees.
00:06:49
Note by the way, that
Ryzen Master did run,
00:06:52
but it's right near the bottom
00:06:54
of these thermal stress loads.
00:06:56
We don't know for sure why that would be,
00:06:57
but my tin foil hat says that maybe AMD
00:06:59
doesn't want you to cook your CPU.
00:07:01
I don't know though.
00:07:03
Even ignoring Ryzen Master,
00:07:04
we can draw some interesting
conclusions here.
00:07:06
If you wanna hit it hard,
00:07:08
it's Prime95 for maximum heat output
00:07:10
with AIDA 64 FPU as a close runner-up.
00:07:13
(man laughs)
00:07:14
As for stability, while we already learned
00:07:16
that this is not what we're looking for.
00:07:19
We want a rapidly oscillating power load.
00:07:21
So let's have a look at
one of our Linpack loads,
00:07:23
OCCT Linpack 2021.
00:07:26
And oof, look at that power
draw porcupine right there.
00:07:31
35 watts spikes, roughly
six times per minute.
00:07:34
How about Linux?
00:07:36
Again, very nice, very
stable clock speeds,
00:07:39
which we expect since we locked it,
00:07:42
but we still saw some
intermittent clock drops
00:07:44
and huge swings in power
drop from 82 watts,
00:07:48
all the way to 170 watts.
00:07:51
This is probably the best
example we could have hoped for
00:07:54
to show you why you need to test
00:07:55
with multiple pieces of software.
00:07:58
Every one of these is a valid part
00:08:00
of your stress test tool kit.
00:08:01
Yet, each is different.
00:08:03
Sitting down and inspecting
these results though,
00:08:05
lead to some pretty tidy conclusions
00:08:06
for you to take with you.
00:08:08
If you wanna get in and out quickly,
00:08:09
Prime95 is gonna give you
thermal results on air
00:08:12
in about 15 minutes and
on an AIO water cooler
00:08:15
in about 30 to 40 minutes.
00:08:17
As for stability, a
Linpack load like Linux,
00:08:20
OCCT Linpack or a Linpack
Extreme are the best bang
00:08:23
for your time bucks.
00:08:25
Unfortunately for those ones,
00:08:26
I can't give you a firm guideline
00:08:28
for how long you need to run them though,
00:08:29
because it comes down
to personal preference.
00:08:32
For a gaming test bench,
00:08:33
I might be comfortable with 10 minutes,
00:08:35
but for a video editing workstation
00:08:37
that I want years of service from,
00:08:39
24 to 48 hours isn't out of the question.
00:08:43
Also, none of these synthetic loads
00:08:45
excuse ignoring real world ones.
00:08:48
If you've got the time and
if you're serious about it,
00:08:50
a Blender Render or a
continuous Cinebench run,
00:08:53
followed by a few rounds of PugetBench
00:08:55
would absolutely be a good idea.
00:08:57
Now, it may not be common knowledge,
00:08:59
but the tools that we use to
log and review our data today
00:09:02
are both free for personal use,
00:09:04
HWiNFO and Generic Log Viewer.
00:09:06
And you can check them out at the links
00:09:08
in the video description.
00:09:09
HWiNFO allows you to
log all your sensor data
00:09:11
while Generic Log Viewer
gives you a perfect tool
00:09:14
for quickly and easily charting that data
00:09:16
and comparing against other runs.
00:09:18
What I hope is common knowledge
though, is our sponsor.
00:09:21
Micro Center.
00:09:22
Micro Center is one of
the best places to shop
00:09:25
for desktops, laptops,
computer components, monitor,
00:09:27
TV, networking equipment, and
all your technology needs.
00:09:30
They've got great prices
and great selection
00:09:33
and 25 locations across the U.S.
00:09:35
You can check out the MSI
Sword, 17.3 inch gaming laptop
00:09:39
at Micro Center, featuring
an Intel Core I7 11800H
00:09:42
and Video RTX 3050 Ti graphics,
00:09:44
16 gigs of DDR4 RAM, a 512 gig SSD,
00:09:48
and of course, a 17.3
inch 144Hz IPS display.
00:09:53
Plus, the MSI Center app
helps you easily control
00:09:55
and customize your laptop
for different use cases,
00:09:58
like battery saving or
extreme performance.
00:10:00
New Micro Center customers
can check the link
00:10:02
in the video description for a coupon code
00:10:04
for a free 240 gig SSD,
no purchase necessary.
00:10:08
Offer valid in-store only.
00:10:09
Thanks for watching.
00:10:10
If you're interested
in this sort of thing,
00:10:12
why not check out Whole
Room Water Cooling.
00:10:14
That was a stress test on us.
00:10:16
Ha ha. That's very funny.
00:10:17
I like your joke that you wrote.
00:10:19
(upbeat music)
