WEBVTT

00:00:00.080 --> 00:00:06.560
we just can't convince you guys sometimes after our video debunking a

00:00:04.160 --> 00:00:12.160
Corsair rep's claim that stacking radiators is ineffective for improving

00:00:09.280 --> 00:00:15.120
cooling you guys told us just how inadequate you felt our test methodology

00:00:14.320 --> 00:00:21.359
was you know what okay fine you guys want science we're going to

00:00:18.800 --> 00:00:25.680
bring you the science colin here spent two bloody weeks making our own data

00:00:24.560 --> 00:00:32.480
logger from scratch and built this behemoth of

00:00:29.920 --> 00:00:38.079
a radiator stack along with this wind tunnel just peppered with

00:00:34.960 --> 00:00:41.440
sensors to find the definitive final

00:00:38.079 --> 00:00:42.960
answer to exactly what effect stacking

00:00:41.440 --> 00:00:47.760
radiators has on water temperature tang wants to help

00:00:45.760 --> 00:00:51.600
you save money by helping you pay for only the mobile data that you use stick

00:00:50.320 --> 00:00:57.079
around until the end of the video to hear about their giveaway or click the link in the video description

00:01:03.760 --> 00:01:11.680
for a quick recap back in the day Corsair called us out for the apparently

00:01:09.360 --> 00:01:17.040
ineffective way that we were recycling the hot air or stacking the radiators in

00:01:15.360 --> 00:01:21.840
our hacc pro now this individual went as far as to

00:01:19.360 --> 00:01:26.400
send us images of their flow simulation data to prove their point our logical

00:01:24.000 --> 00:01:31.759
answer then was to just build a similarly configured system to prove

00:01:28.799 --> 00:01:35.920
that more rads definitely equals more better so we did

00:01:34.159 --> 00:01:40.640
and it did but while some of your complaints can be

00:01:38.560 --> 00:01:44.720
chalked up to frankly not paying enough attention to the video i get it happens

00:01:43.600 --> 00:01:50.479
others seemed to have some merit you guys still

00:01:47.439 --> 00:01:53.200
weren't satisfied and so i greenlit a

00:01:50.479 --> 00:01:57.280
project to go full crazy on a test rig to isolate the variables that we had

00:01:54.960 --> 00:02:01.920
kind of hand waved away last time we started out by reaching out to our

00:01:58.799 --> 00:02:05.200
friends at alphacool for seven of their

00:02:01.920 --> 00:02:07.200
240 millimeter cross flow rads and

00:02:05.200 --> 00:02:11.440
thermistor type temperature probes performance pc also threw some low

00:02:09.440 --> 00:02:16.000
profile fittings our way for stacking them together like this so with a little

00:02:13.200 --> 00:02:20.640
bit of fabrication and some 3d printed end caps here a wind tunnel was slapped

00:02:18.560 --> 00:02:24.000
together from clear acrylic these cross flow rads were specifically chosen

00:02:22.400 --> 00:02:28.239
because they give us some really interesting flow options that we may

00:02:26.160 --> 00:02:32.319
explore down the line so get subscribed if you want to see more from this test

00:02:30.080 --> 00:02:37.680
rig here for this video though we're sticking to serial flow in one out and

00:02:35.440 --> 00:02:41.920
then into the next and so on and so forth because that's how most people

00:02:39.760 --> 00:02:47.519
would configure a water cooling system if they had multiple radiators at every

00:02:44.720 --> 00:02:52.000
single radiator inlet and outlet we've got temperature probes see these puppies

00:02:50.000 --> 00:02:57.440
right here these are going to tell us the difference in temperature or delta t

00:02:55.360 --> 00:03:01.040
from when the water enters the radiator to when it exits but

00:02:59.519 --> 00:03:06.080
we had a problem how do you go about calibrating and

00:03:03.440 --> 00:03:12.080
recording the data from all of these sensors colin never content to just take

00:03:08.640 --> 00:03:13.840
the easy path decided to create this

00:03:12.080 --> 00:03:18.959
this right here with a little bit of code and arduino due chosen for its

00:03:16.560 --> 00:03:23.840
large array of analog inputs gives us full control over our data stream oh

00:03:21.920 --> 00:03:28.080
yeah and don't forget the absolute beauty of a breadboard right

00:03:25.360 --> 00:03:32.480
here that's important the 32-bit atmel processor in here can actually sample

00:03:29.840 --> 00:03:37.840
data much faster than we need but we settled on pulling the sensors every 250

00:03:35.120 --> 00:03:42.400
milliseconds or four times per sensor long story short the arduino does a

00:03:40.400 --> 00:03:46.799
little bit of math and converts the resistance of the thermistor sensor

00:03:44.720 --> 00:03:49.840
embedded in each of these fittings into degrees celsius using the steinhart

00:03:48.480 --> 00:03:54.239
heart equation if you want to give this a shot yourself by the way we're going to have a link in the description to the

00:03:52.319 --> 00:03:59.200
arduino code that we used and a good starting place for some reading now

00:03:57.040 --> 00:04:03.120
due to the varying resistance of the sensors themselves

00:04:00.480 --> 00:04:06.959
as well as the actual wire connecting them even

00:04:04.640 --> 00:04:12.959
the raw output that the arduino sees isn't accurate enough for our purposes

00:04:09.360 --> 00:04:14.720
so we added these trim pots right here

00:04:12.959 --> 00:04:20.079
numbered zero to eight to allow us to calibrate each sensor to

00:04:17.280 --> 00:04:23.520
a known temperature using a water bath once all the sensors were reading within

00:04:21.759 --> 00:04:27.919
half a degree of each other the data is pushed out via a serial connection to

00:04:25.600 --> 00:04:31.600
this surface laptop running putty an open source serial monitor which can

00:04:29.680 --> 00:04:36.320
output all the data to a convenient text file for later processing a relatively

00:04:33.759 --> 00:04:39.120
convenient text file okay so now's a good time to take a break from line of

00:04:37.600 --> 00:04:44.080
science tips here and hydrate lttstore.com before we get into our heat

00:04:41.919 --> 00:04:50.479
source on this test bench right here we've paired a 32 core Threadripper

00:04:46.800 --> 00:04:52.720
3970x furnace of a CPU with an NVIDIA

00:04:50.479 --> 00:04:57.759
titan v which combined should dump about 500 watts of heat into

00:04:55.759 --> 00:05:02.720
our system on the water side of things it's a fairly simple affair with a d5

00:05:00.080 --> 00:05:07.360
pump with a custom top as well as a no-name reservoir

00:05:05.440 --> 00:05:13.280
so that's the setup let's dive into how everything performed

00:05:10.400 --> 00:05:18.880
for our first two tests each radiator was paired with Noctua nff12 fans and we

00:05:16.479 --> 00:05:23.039
used our full assembly of seven radiators stacked one after the other

00:05:21.039 --> 00:05:28.400
down our wind tunnel air flows over the rads from sensor 0 to sensor 7. kind of

00:05:26.240 --> 00:05:33.680
like if our stack was mounted as a front intake on a super weird case that's like

00:05:32.160 --> 00:05:38.880
really deep with water flowing from sensor 0 to sensor 7 the radiator with

00:05:36.639 --> 00:05:42.960
the hottest water inside it is the one at the very front of the case and off

00:05:41.120 --> 00:05:47.759
the hop we can see that most of the cooling happens in this first radiator

00:05:45.680 --> 00:05:53.440
represented by the red line this makes sense because it's getting

00:05:49.680 --> 00:05:55.600
both the coldest air and the hottest

00:05:53.440 --> 00:06:00.720
water coming straight off of our system here the second radiator is orange and then

00:05:58.639 --> 00:06:06.319
yellow and then so on with each successive rad doing less and less of

00:06:03.360 --> 00:06:11.759
the actual work we hit a point of basically negligible returns at about

00:06:08.880 --> 00:06:16.080
radiator six now if we focus in on the initial start of load the first two

00:06:13.759 --> 00:06:21.039
minutes we can see here that each radiator gets progressively hotter in

00:06:18.560 --> 00:06:26.160
sequence which is to be expected however if we look at just the section

00:06:23.919 --> 00:06:30.720
where the load is stopped the results get much more interesting

00:06:28.240 --> 00:06:36.319
check this out when firmark is closed the first radiator stays the hottest

00:06:33.440 --> 00:06:41.280
which we expect but see how these lines cross over subsequent radiators

00:06:38.880 --> 00:06:46.639
this is actually a really cool visual representation of heat soak

00:06:43.919 --> 00:06:51.840
so the preheated radiators farther down the line are actually reheating the

00:06:49.520 --> 00:06:56.800
water coming from that front rad which as you guys remember is doing most of

00:06:54.000 --> 00:07:00.720
the work as it passes through the dark blue line here is the exit water

00:06:58.800 --> 00:07:05.759
temperature and it actually comes out hotter than the other rads in the series

00:07:04.000 --> 00:07:10.720
zooming out a little bit more shows the middle radiators eventually equalize and

00:07:08.400 --> 00:07:13.840
fall into order once again so seeing heat soak right here in the data is

00:07:12.639 --> 00:07:20.080
super cool well more like 39.5 degrees not that cool but

00:07:17.680 --> 00:07:24.639
hey who's counting got him of course one way to counteract

00:07:22.400 --> 00:07:29.280
this reliance on the front radiator is to run the water the other way as many

00:07:27.360 --> 00:07:36.000
of our viewers recommended last time around that puts the freshest air into

00:07:33.360 --> 00:07:39.440
the last radiator for the water to flow through

00:07:36.880 --> 00:07:45.520
so the only variable between test one and test two is that we reversed the

00:07:42.160 --> 00:07:47.599
water flow direction in the loop this is

00:07:45.520 --> 00:07:52.880
generally considered to be the better way to stack and according to our

00:07:50.000 --> 00:07:57.360
testing that checks out by flipping the flow all the rads come up to temperature

00:07:55.440 --> 00:08:02.080
more slowly as the stack gets progressively hotter and if we overlay

00:07:59.520 --> 00:08:06.160
the two water exit temps and offset to account for the ambient slash idle

00:08:04.240 --> 00:08:10.639
temperature of the system the outlet temperature does show minor improvements

00:08:08.479 --> 00:08:16.080
versus test one's flow direction cool quick takeaways then it appears that for

00:08:12.960 --> 00:08:17.919
this combination of CPU and GPU at least

00:08:16.080 --> 00:08:22.000
by the time the water reaches the six radiator we've imparted as much heat

00:08:19.919 --> 00:08:25.840
into the air as we're going to at the highest water temp we were seeing a

00:08:23.360 --> 00:08:30.080
delta t of about four degrees from the first rad to the sixth the last one that

00:08:28.240 --> 00:08:33.839
was actually making a difference give or take a half a degree

00:08:31.520 --> 00:08:39.200
and we also see that the majority of the work is being done by the first radiator

00:08:36.800 --> 00:08:44.640
so our Corsair representatives point about the small gains from stacking

00:08:41.680 --> 00:08:48.399
being offset by the reduction in airflow that comes from adding more restriction

00:08:46.560 --> 00:08:53.120
and more fins does seem to be fair

00:08:50.640 --> 00:08:58.880
but it's also true that there is still performance to be gained so

00:08:56.320 --> 00:09:03.839
for our third scenario we moved on to a more sensible setup right here you're

00:09:01.120 --> 00:09:06.880
looking at one bank of fans and two radiators and

00:09:05.760 --> 00:09:11.519
here there's definitely less volume in the

00:09:08.880 --> 00:09:15.920
loop so we see the temperatures rise much quicker overall we hit a peak temp

00:09:14.080 --> 00:09:19.680
of 38 and a half degrees after 10 minutes of load

00:09:17.279 --> 00:09:26.000
but when we ran the same test with a single radiator we also hit

00:09:23.080 --> 00:09:31.040
38.5 degrees here's where it gets really interesting

00:09:28.000 --> 00:09:33.120
for a sanity check we charted the change

00:09:31.040 --> 00:09:38.399
in temperature from the inlet to outlet side of both the single and the dual

00:09:35.040 --> 00:09:40.880
tests so a higher delta t on one system

00:09:38.399 --> 00:09:44.720
would mean that that particular one is more effective now the data is a bit

00:09:42.720 --> 00:09:48.560
noisy for one of the tests but regardless they look pretty well aligned

00:09:47.839 --> 00:09:54.240
so in my eyes that's pretty definitive when

00:09:51.440 --> 00:09:58.720
it comes to one rad versus two assuming you've only got one set of fans

00:09:56.320 --> 00:10:03.519
a second radiator isn't really going to do anything for you besides add some

00:10:00.880 --> 00:10:08.399
thermal capacity in the form of metal and water mass so i can't believe i'm

00:10:06.560 --> 00:10:12.240
saying it but i was wrong i've been saying that a lot

00:10:11.040 --> 00:10:17.360
lately at least i was wrong for this test now

00:10:15.360 --> 00:10:22.320
as i alluded to earlier a lot of you missed this but the minecraft server did

00:10:20.079 --> 00:10:26.720
perform better with three versus two radiators under full load and for that

00:10:24.399 --> 00:10:31.279
matter so did the hack pro we believe this comes down to either having enough

00:10:29.200 --> 00:10:36.959
airflow to utilize the extra surface area or having a mix of preheated air

00:10:34.800 --> 00:10:41.600
and ambient air mixed in because a typical computer case is not actually a

00:10:39.600 --> 00:10:46.320
sealed system but we still learned a lot here first is that more rads does not

00:10:44.640 --> 00:10:50.560
always mean more better in the case of our one rad versus two in the wind

00:10:48.000 --> 00:10:54.640
tunnel it appears that Corsair is right in that adding a second radiator doesn't

00:10:52.160 --> 00:10:58.160
do anything the data we gathered proves this and this is with a bank of static

00:10:56.560 --> 00:11:02.480
pressure optimized fans these are not like pinner crappy fans second some of

00:11:00.800 --> 00:11:07.040
the perceived gains from radiator stacking could simply come down to

00:11:04.399 --> 00:11:10.959
flawed tests where the system is not allowed to reach equilibrium making it

00:11:09.360 --> 00:11:14.880
seem like it's running cooler but if they had just left it for longer that

00:11:12.800 --> 00:11:18.240
water would have eventually heated up that's an easy mistake to make because

00:11:16.800 --> 00:11:22.000
the increase in thermal mass of the radiators themselves as well as their

00:11:20.000 --> 00:11:25.200
water capacity can make this kind of testing take a really

00:11:24.240 --> 00:11:31.440
long time third and finally you can still get

00:11:28.480 --> 00:11:36.240
a benefit from stacking but only with a correctly configured system that moves

00:11:33.760 --> 00:11:41.040
the air fast enough to maintain a meaningful difference in temperature

00:11:38.240 --> 00:11:45.519
between the air and the water in the last radiator in the stack

00:11:43.839 --> 00:11:51.040
so i really wanted to come out and say boom headshot

00:11:47.360 --> 00:11:53.120
gotcha Corsair but when we isolated

00:11:51.040 --> 00:11:57.920
everything down like this the results didn't quite match up with

00:11:55.440 --> 00:12:03.440
that this is not what i expected but in a way that's kind of great too and i

00:12:00.800 --> 00:12:08.079
learned something today kind of great like our sponsor ting does

00:12:06.240 --> 00:12:12.399
mobile phone service differently there's no contracts overage fees or any other

00:12:10.240 --> 00:12:16.560
carrier tricks you just pay a fair price for the talk text and data that you use

00:12:14.480 --> 00:12:20.800
each month and it's especially great if you're stuck at home using wi-fi instead

00:12:18.639 --> 00:12:24.320
of mobile data with ting you've got complete control over your cell phone

00:12:22.399 --> 00:12:28.560
account so you can set alerts and caps for every device and they've got

00:12:26.399 --> 00:12:33.920
nationwide lte coverage using t-mobile sprint and verizon that means great

00:12:31.440 --> 00:12:36.959
coverage from coast to coast almost any phone will work with ting from an

00:12:35.279 --> 00:12:42.560
ancient motorola razer sitting in your basement to the latest iphone 11 series

00:12:39.760 --> 00:12:45.680
and there's a giveaway over at ting.com if you guys head over there right now

00:12:43.680 --> 00:12:50.839
you've got a chance to win 20 ting swag bags or their ultimate giveaway a galaxy

00:12:48.399 --> 00:12:54.880
note 10 plus just enter at mobile.ting.com Linus giveaway or at the

00:12:53.360 --> 00:12:57.760
link in the video description you'll find the full details and when it ends

00:12:56.639 --> 00:13:02.880
down below so thanks for watching guys i want to throw you guys over to the hack pro

00:13:01.200 --> 00:13:06.880
series which actually inspired this whole experiment you can check out how

00:13:05.120 --> 00:13:11.600
difficult it was to jam water cooling into that case however effective or

00:13:08.480 --> 00:13:11.600
ineffective it ended up being