WEBVTT

00:00:00.060 --> 00:00:05.759
why would anyone run an app that is specifically designed to crash their

00:00:04.319 --> 00:00:10.019
computer that's a good question and I have some

00:00:07.980 --> 00:00:15.120
pretty good answers stress testing applications can make your computer

00:00:11.940 --> 00:00:17.460
overheat or even outright crash but in

00:00:15.120 --> 00:00:21.300
doing so they can also reveal weak points in your system that could have

00:00:18.600 --> 00:00:26.400
led to bigger problems like instability or even data loss but of all the options

00:00:24.420 --> 00:00:30.900
out there for CPU stress or burn applications which one is actually going

00:00:28.980 --> 00:00:37.200
to hit your CPU the hardest the answer spoiler alert is prime95

00:00:34.680 --> 00:00:41.160
at least in terms of heat but the whole story is a lot more complicated than

00:00:39.420 --> 00:00:45.540
that and we just couldn't tell it without our sponsor cable mod cable mod

00:00:43.980 --> 00:00:49.440
allows you to personalize the look of your PC with custom colored sleeve

00:00:47.460 --> 00:00:52.680
cables try out their configurator to build your cables exactly how you want

00:00:51.120 --> 00:00:58.620
them with their realistic cable preview we're going to have them linked down below like test driving a vehicle stress

00:00:56.160 --> 00:01:03.420
testing is an important part of Savvy PC shopping and maintenance but let's say

00:01:01.500 --> 00:01:09.479
you've never done it before you head on over to Google Smash in some keywords

00:01:05.400 --> 00:01:11.760
and oh that's a lot of options you got

00:01:09.479 --> 00:01:16.979
your blender cinebench R20 and cinebench R23 occt prime95 CPU expert heavy load

00:01:14.820 --> 00:01:23.100
Linux Intel XT ryzen Master Intel princess 64 Lin pack extreme

00:01:19.740 --> 00:01:25.320
and that's all just for the CPU the

00:01:23.100 --> 00:01:30.479
unfortunate reality is that there is no Silver Bullet best test and depending on

00:01:28.439 --> 00:01:35.100
your Hardware software and what you're trying to accomplish you could end up

00:01:32.220 --> 00:01:40.920
with completely different answers so yeah we spoiled the conclusion prime95

00:01:37.979 --> 00:01:45.960
runs the hottest but even then which of the five tests 10 if you include the

00:01:43.320 --> 00:01:50.759
avx2 checkbox should you run to find out we built up two identical

00:01:48.840 --> 00:01:55.920
PCS with just two differences motherboard and CPU here's the thing we

00:01:54.000 --> 00:02:00.360
suspected the team blue and team red might behave differently under extreme

00:01:58.079 --> 00:02:05.700
conditions and we're really glad we tested both because they absolutely did

00:02:02.939 --> 00:02:10.860
on both platforms though we also locked our CPU clock speeds core voltages and

00:02:08.459 --> 00:02:15.959
fan speeds leaving just our CPU temperatures as our stress indicator

00:02:13.800 --> 00:02:20.160
time to show the graphs give them a graph editor

00:02:17.040 --> 00:02:22.620
Intel first two tests didn't work at all

00:02:20.160 --> 00:02:27.360
rise and master we saw that one coming and Intel extreme tuning utility

00:02:25.560 --> 00:02:30.900
we didn't see that one coming it ended up erroring out the same way in Windows

00:02:29.220 --> 00:02:36.660
11 that we saw in our recent Dragon Canyon Nook video among the tests that

00:02:33.540 --> 00:02:39.480
did work prime95 small fft stood

00:02:36.660 --> 00:02:46.980
absolutely alone beating out the next highest score by 10 freaking degrees

00:02:43.560 --> 00:02:49.140
there's just no other way to put this we

00:02:46.980 --> 00:02:54.720
did not believe our results so thinking that maybe Intel's turbo

00:02:52.260 --> 00:02:59.879
boost power management was bunging up the other tests or something we re-ran

00:02:57.239 --> 00:03:05.099
the entire Suite with our voltages and our clock speeds back to stock and to

00:03:03.120 --> 00:03:11.099
say I'm shocked would be putting it mildly it was like witnessing a brutal

00:03:07.260 --> 00:03:13.680
beat down in slow motion again

00:03:11.099 --> 00:03:18.480
we have some theories here but first let's get to the rest of the results our

00:03:15.900 --> 00:03:23.879
10 middle performers had just a six degree overall spread then we had our

00:03:20.819 --> 00:03:27.780
underperformers Intel burn test CPU-z

00:03:23.879 --> 00:03:28.860
and occt's CPU large data set extreme

00:03:27.780 --> 00:03:36.060
benchmark at this point you can probably see why we declared Prime 95 best in heat but

00:03:34.080 --> 00:03:41.040
looking at Thermals alone would be a huge mistake check out the heat profile

00:03:38.459 --> 00:03:47.640
and power consumption profile of prime95 small fft huge spikes then they just

00:03:44.640 --> 00:03:49.440
hold indefinitely it achieves this by

00:03:47.640 --> 00:03:54.540
repeatedly performing demanding mathematical calculations but on a data

00:03:51.959 --> 00:04:01.440
set that's so small it fits right in the CPUs on die Cache no bottlenecks maximum

00:03:58.680 --> 00:04:05.340
burn but obviously no real world application behaves like that even

00:04:03.299 --> 00:04:11.159
notoriously demanding professional workloads like CPU rendering and blender

00:04:08.040 --> 00:04:13.260
or video encoding will generate very

00:04:11.159 --> 00:04:17.340
different profiles so if I was validating a thermal design for a

00:04:15.180 --> 00:04:23.280
demanding client and I wanted to see an absolute worst case scenario Prime 95

00:04:20.220 --> 00:04:25.020
needs to be part of my toolkit but if I

00:04:23.280 --> 00:04:29.880
care more about stability in real world applications I need to look elsewhere

00:04:27.240 --> 00:04:34.139
enter linpak the mo most used stability stress testing tool of all time it was

00:04:32.580 --> 00:04:39.300
originally created for use on super computers in the 1970s and in a nutshell

00:04:37.020 --> 00:04:44.280
it's a software library that performs linear algebra this equation

00:04:41.400 --> 00:04:48.540
specifically on repeat and it still behaves the same way today except it's

00:04:46.560 --> 00:04:51.840
been rehashed multiple times to keep up with modern computers which is pretty

00:04:50.580 --> 00:04:56.100
important considering that it was written in Fortran a code language that

00:04:53.940 --> 00:05:01.800
is so old it was originally input via punch carts fun fact by the way because

00:04:58.919 --> 00:05:06.960
linpak is a library not an application it's easy for developers to integrate it

00:05:04.320 --> 00:05:12.960
into their own test Suites and six of the tests we used are actually Lin pack

00:05:09.419 --> 00:05:14.759
based but even then not all tests are

00:05:12.960 --> 00:05:20.100
equal because there are multiple Lin pack versions and any of those can be

00:05:17.699 --> 00:05:23.820
combined with other methods to create a complete hodgepodge of stress for your

00:05:22.500 --> 00:05:29.639
poor CPU this is what a limpac load looks like on

00:05:26.940 --> 00:05:33.479
our thermal graphs a spiky wave gets produced as each equation is solved then

00:05:32.220 --> 00:05:38.580
there's a brief pause between calculations and now that we know this

00:05:35.759 --> 00:05:42.180
it's really easy to spot this shape in many of the tests that we ran now if we

00:05:40.800 --> 00:05:47.759
switch from temperature to power consumption we see well the same thing

00:05:45.060 --> 00:05:52.199
these swings in power are really tough on not just your CPU but also your

00:05:50.100 --> 00:05:56.580
motherboard vrms and your power supply which is a perfect combination for

00:05:54.419 --> 00:06:01.680
evaluating stability which we're definitely going to check out on AMD's

00:05:58.440 --> 00:06:04.080
New 3D zcash CPUs so get subscribed so

00:06:01.680 --> 00:06:08.400
you don't miss it also buy a shirt or a water bottle haha lttstore.com on that

00:06:06.780 --> 00:06:12.600
note let's take a look at our AMD results here we didn't see anything that

00:06:10.620 --> 00:06:18.180
seemed anomalous so we left our voltages and clock speeds locked per our original

00:06:14.820 --> 00:06:20.880
test methodology to start with AMD ran

00:06:18.180 --> 00:06:25.199
much cooler overall with Delta T's so that's the temperature difference

00:06:22.139 --> 00:06:28.139
between ambient and our CPU peaking at

00:06:25.199 --> 00:06:33.360
just 39 degrees compared to Intel's 56. man 12th gen performs but lordy is it

00:06:32.280 --> 00:06:38.699
ever hot now prime95 still came on top for

00:06:35.580 --> 00:06:41.819
thermal stress but only barely just half

00:06:38.699 --> 00:06:43.560
a degree behind was Ida 64. then the

00:06:41.819 --> 00:06:47.880
middle of our chart doesn't have the same flat middle Zone as Intel instead

00:06:45.900 --> 00:06:53.100
there's a fairly consistent decline with a spread of about 10 degrees note by the

00:06:50.520 --> 00:06:56.639
way that rise and master did run but it's right near the bottom of these

00:06:54.780 --> 00:07:00.479
thermal stress loads we don't know for sure why that would be but my tinfoil

00:06:58.740 --> 00:07:04.979
hat says that maybe AMD doesn't want you to cook your CPU I don't know though

00:07:02.940 --> 00:07:09.720
even ignoring ryzen Master we can draw some interesting conclusions here in if

00:07:07.020 --> 00:07:14.460
you want to hit it hard it's prime95 for maximum heat output with Ida 64 FPU as a

00:07:12.419 --> 00:07:18.900
close runner up as for stability well we already learned

00:07:16.800 --> 00:07:23.099
that this is not what we're looking for we want a rapidly oscillating power load

00:07:21.660 --> 00:07:31.220
so let's have a look at one of our Lin pack loads occt linpak 2021 and oof look

00:07:28.800 --> 00:07:37.979
at that power draw porcupine right there 35 Watts bikes roughly 6 times per

00:07:34.440 --> 00:07:40.680
minute how about Linux again very nice

00:07:37.979 --> 00:07:46.680
very stable clock speeds which we expect since we locked it but we still saw some

00:07:43.560 --> 00:07:49.259
intermittent clock drops and huge swings

00:07:46.680 --> 00:07:54.539
in power draw from 82 Watts all the way to 170 Watts this is probably the best

00:07:52.860 --> 00:07:59.460
example we could have hoped for to show you why you need to test with multiple

00:07:56.699 --> 00:08:04.020
pieces of software every one of these is a valid part of your stress test toolkit

00:08:01.500 --> 00:08:07.259
yet each is different sitting down and inspecting these results though lead to

00:08:05.520 --> 00:08:11.039
some pretty tidy conclusions for you to take with you if you want to get in and

00:08:09.000 --> 00:08:15.660
out quickly Prime 95 is going to give you thermal results on air in about 15

00:08:13.080 --> 00:08:21.419
minutes and on an a a water cooler in about 30 to 40 minutes as for stability

00:08:18.060 --> 00:08:24.000
a Lin pack load like Linux occt linpak

00:08:21.419 --> 00:08:27.180
or linpak Xtreme are the best bang for your time bucks

00:08:25.319 --> 00:08:30.240
unfortunately for those ones I can't give you a firm guideline for how long

00:08:28.919 --> 00:08:34.320
you need to run them though because it comes down to personal preference for a

00:08:32.459 --> 00:08:39.240
gaming test bench I might be comfortable with 10 minutes but for a video editing

00:08:37.260 --> 00:08:45.600
workstation that I want years of service from 24 to 48 hours isn't out of the

00:08:42.599 --> 00:08:48.720
question also none of these synthetic

00:08:45.600 --> 00:08:49.860
loads excuse ignoring real world ones if

00:08:48.720 --> 00:08:54.360
you've got the time and if you're serious about it a blender render or

00:08:52.320 --> 00:08:58.440
continuous cinebench run followed by a few rounds of Puget bench would

00:08:55.980 --> 00:09:02.040
absolutely be a good idea now it may not be common knowledge but the tools that

00:09:00.120 --> 00:09:07.680
we use to log and review our data today are both free for personal use HW info

00:09:04.980 --> 00:09:10.800
and generic log viewer and you can check them out at the links in the video

00:09:08.880 --> 00:09:14.580
description HW info allows you to log all your sensor data while generic log

00:09:12.660 --> 00:09:19.260
viewer gives you a perfect tool for quickly and easily charting that data

00:09:16.620 --> 00:09:23.760
and comparing against other runs what I hope is common knowledge though is our

00:09:21.000 --> 00:09:27.420
sponsor Microcenter Microcenter is one of the best places to shop for desk

00:09:25.920 --> 00:09:31.440
top tops computer components monitor Steven networking equipment and all your

00:09:29.700 --> 00:09:36.360
technology needs they've got great prices and great selection and 25

00:09:33.839 --> 00:09:40.740
locations across the US you can check out the MSI sword 17.3 inch gaming

00:09:39.120 --> 00:09:48.180
laptop at Micro Center featuring an Intel Core i7 11800h NVIDIA RTX 3050 TI

00:09:44.519 --> 00:09:52.440
Graphics 16 gigs of ddr4 RAM a 512 gig

00:09:48.180 --> 00:09:55.019
SSD and of course a 17.3 inch 144hz IPS

00:09:52.440 --> 00:09:58.500
display plus the MSI Center app helps you easily control and customize your

00:09:56.760 --> 00:10:02.459
laptop for different use cases like battery saving or Extreme Performance

00:10:00.360 --> 00:10:07.080
new Micro Center customers can check the link in the video description for a

00:10:03.839 --> 00:10:09.180
coupon code for a free 240 gig SSD no

00:10:07.080 --> 00:10:12.360
purchase necessary offer valid in-store only thanks for watching if you're

00:10:11.040 --> 00:10:17.100
interested in this sort of thing why not check out whole room water cooling that

00:10:14.459 --> 00:10:22.339
was a stress test on us haha that's very funny I like your joke that you wrote

00:10:20.100 --> 00:10:22.339
foreign