WEBVTT 00:00:00.160 --> 00:00:07.720 i'm really excited about this video guys if you believe what Intel tells you 00:00:04.480 --> 00:00:12.719 you'd think that this core i9 00:00:07.720 --> 00:00:15.280 9900k 8-core processor is a 95-watt CPU 00:00:12.719 --> 00:00:21.560 while if you believe what AMD tells you you'd believe that this ryzen 7 3800x 00:00:19.039 --> 00:00:27.199 also eight core processor is a 105 watt CPU 00:00:24.080 --> 00:00:30.720 so then clearly if AMD's number is 00:00:27.199 --> 00:00:33.200 higher and Intel's number is lower 00:00:30.720 --> 00:00:36.079 that Ryzen CPU it's gonna run hotter and kick more heat out into your room 00:00:34.960 --> 00:00:41.920 right actually not necessarily so as it turns 00:00:38.879 --> 00:00:44.640 out there is no industry standard way 00:00:41.920 --> 00:00:50.160 that we can all agree on of reporting the power consumption or the TDP of a 00:00:48.000 --> 00:00:53.920 computer processor so those numbers that i was talking about 00:00:51.760 --> 00:00:57.600 before well it turns out it's up to third parties in the media to 00:00:55.600 --> 00:01:05.680 investigate who is representing their product realistically and who is 00:01:00.480 --> 00:01:05.680 painting a rosier picture than reality 00:01:09.680 --> 00:01:15.439 well get on with it then so on our table here 00:01:12.960 --> 00:01:19.759 we've got two benches one for team blue one for team red with our eight core 00:01:17.200 --> 00:01:23.119 cpus installed to keep everything equal we're running our memory at the same 00:01:21.280 --> 00:01:28.880 speed between them and they're each running a fresh install right yes fresh 00:01:25.360 --> 00:01:30.799 os install on a pci express gen 3 drive 00:01:28.880 --> 00:01:34.799 remember guys Intel doesn't have a consumer PCIe gen 4 chipset yet now 00:01:33.360 --> 00:01:41.280 here's the tricky part measuring the total thermal energy 00:01:37.360 --> 00:01:44.000 output of a CPU is not as simple as just 00:01:41.280 --> 00:01:48.240 strapping the same cooler onto them running prime 95 and then recording the 00:01:47.040 --> 00:01:53.439 temperature because of differences in die size ihs 00:01:51.200 --> 00:02:00.560 solder quality software reporting accuracy etc it's actually possible for 00:01:56.960 --> 00:02:03.680 a processor to be outputting more heat 00:02:00.560 --> 00:02:05.520 while registering a lower temperature 00:02:03.680 --> 00:02:10.959 or vice versa so what Anthony's been working on 00:02:08.399 --> 00:02:17.040 is a way to instead capture as much of the heat from our 00:02:14.000 --> 00:02:18.400 processors as possible into a known 00:02:17.040 --> 00:02:23.599 thermal mass then measure the Ryzen temperature of 00:02:22.080 --> 00:02:29.200 our thermal mass rather than the reported temperature of 00:02:26.560 --> 00:02:32.879 our CPU now before we can do this we have to prepare a couple of things 00:02:31.120 --> 00:02:37.680 normally you'd never insulate the back of your motherboard and it's power delivery components that's a great 00:02:35.519 --> 00:02:41.440 recipe for premature failure but we want as much of the heat going into our 00:02:38.959 --> 00:02:45.519 thermal mass to be from the CPU as possible so 00:02:43.280 --> 00:02:49.120 sorry motherboard times two 00:02:46.640 --> 00:02:54.319 we also need to insulate our mint julep cup here that way we can plot the change 00:02:52.319 --> 00:02:59.040 in the water's temperature over time without worrying so much about the heat 00:02:57.120 --> 00:03:02.800 that we are losing to the surrounding environment through the metal walls of 00:03:01.040 --> 00:03:06.000 our cup not all containers have insulated walls like the LTT water 00:03:04.560 --> 00:03:09.760 bottle which keeps your drink cold on a hot day lttstore.com 00:03:08.640 --> 00:03:14.720 now let's take a look at how we're mounting our thermal capture device to the CPU honestly even compared to zip 00:03:13.760 --> 00:03:20.239 ties this is going to be pretty sketchy but we need to make sure that we could see 00:03:18.000 --> 00:03:25.040 the coolant inside while maintaining the same mounting pressure on both systems 00:03:22.879 --> 00:03:29.599 and a weighted piece of glass ticked both of those boxes 00:03:27.120 --> 00:03:33.440 now you guys might be wondering if all we need is a thermal mass 00:03:31.680 --> 00:03:38.400 why use water why not just put a big hunk of metal 00:03:35.760 --> 00:03:44.959 with an embedded thermal probe on top of the CPU and the answer of course is that 00:03:42.239 --> 00:03:49.760 that's not nearly as much fun to look at enter thermochromic ink this stuff comes 00:03:48.159 --> 00:03:53.680 from lcr hall crest and it's called chromogen and what it does is once the 00:03:51.760 --> 00:03:57.680 water reaches 80 degrees celsius the ink will begin to change color from white to 00:03:55.360 --> 00:04:00.799 black basically whoever hits that point first is outputting the most heat since 00:03:59.360 --> 00:04:04.239 this is permanent we'll have to be sure to add the same amount for each run uh 00:04:02.720 --> 00:04:09.120 we didn't bring a syringe so do you want to just pour a cap full uh yeah sure i 00:04:06.159 --> 00:04:09.120 can try okay sure 00:04:10.480 --> 00:04:15.439 and it goes whoa 00:04:13.680 --> 00:04:19.040 give a little mix with our k-type probe that's what those are for right yeah and 00:04:17.280 --> 00:04:22.959 we'll go ahead and don't get cut by the edges of this glass please 00:04:24.160 --> 00:04:31.520 so we've got just about 300 milliliters of water in 00:04:28.880 --> 00:04:38.320 there and then our ink probably puts us somewhere in the neighborhood of about 00:04:35.160 --> 00:04:42.960 325s about a tablespoon of ink 00:04:38.320 --> 00:04:45.040 so ah let's go ahead and turn on the 00:04:42.960 --> 00:04:49.759 machine that sounds like an idea 00:04:47.600 --> 00:04:52.720 not sure if it's a good idea all right let's fire up our thermal 00:04:51.360 --> 00:04:56.000 probe here 10 degrees so you guys might have 00:04:54.560 --> 00:04:59.199 noticed that our coolant is pretty chilly that's because we put it in the 00:04:57.680 --> 00:05:02.639 fridge before we started to make sure we had a little bit of time to boot up the 00:05:00.800 --> 00:05:07.759 system before we needed to start the test so the plan is to hit go on blender here 00:05:05.440 --> 00:05:10.960 at exactly 22 degrees celsius so i'm gonna go ahead and hit it with uh with a 00:05:09.440 --> 00:05:15.440 load here to try and heat up our water a bit okay 00:05:12.880 --> 00:05:20.000 so we ran into a slight problem the instant we fire up any kind of CPU 00:05:17.680 --> 00:05:24.639 intensive load our processor temperatures go to 00:05:21.840 --> 00:05:28.400 thermal throttling territory boom which means that we are not getting the full 00:05:26.880 --> 00:05:33.840 power out of the chip because it's holding itself back we unfortunately 00:05:30.720 --> 00:05:36.960 selected a stainless steel vessel for 00:05:33.840 --> 00:05:38.400 our water and dye solution 00:05:36.960 --> 00:05:44.160 but no worries the good folks at madrina's have us covered they shipped us these weird 00:05:41.520 --> 00:05:51.160 camping cups and alex found one so we are going to uh convert this 00:05:47.120 --> 00:05:51.160 to be our cooler now 00:05:58.160 --> 00:06:05.880 do you think it's steel do we have a magnet that won't tell us if it's stainless steel wish us luck 00:06:06.479 --> 00:06:13.199 so good news and bad news alex good news is i removed the bottom of the cup the 00:06:11.039 --> 00:06:17.120 bad news is there is a whole lot of space in between the one 00:06:16.000 --> 00:06:21.800 bottom and the other more different bottom cup number two 00:06:24.160 --> 00:06:29.360 there we go okay so 00:06:27.199 --> 00:06:33.680 after a bit of milling alex has gotten us this little disc here and as you can 00:06:31.440 --> 00:06:39.440 see it's got a little bit of a cut out here on the inside and what we can do is 00:06:36.080 --> 00:06:40.639 just slot our cup right into that mix 00:06:39.440 --> 00:06:45.039 this up put it around the edge and it should be 00:06:42.479 --> 00:06:50.240 watertight within about 20 minutes i never said i was good at arts and 00:06:46.840 --> 00:06:51.120 crafts i am measuring out 00:06:50.240 --> 00:06:58.080 wait why is this milky 00:06:54.000 --> 00:06:58.080 oh wait what it was in the cap 00:06:59.759 --> 00:07:05.759 so we're back 300 mils of water oh wow we kind of 00:07:04.479 --> 00:07:10.479 need more now so we're back 00:07:07.639 --> 00:07:15.599 400 milliliters of water later we have a slightly larger vessel this time let's 00:07:12.960 --> 00:07:18.639 use this cap this time let's put on some fresh thermal compound i see you just 00:07:17.199 --> 00:07:21.680 cleaned it off okay it's a little on the heavy side but 00:07:20.319 --> 00:07:25.919 should be okay and 00:07:23.360 --> 00:07:29.039 schwippity schwampty does it fit 00:07:27.680 --> 00:07:34.160 please tell me it doesn't interfere with anything well there's only one way to 00:07:31.520 --> 00:07:37.199 find out let's fire this mess up are we missing anything 00:07:35.440 --> 00:07:42.080 um aside from like the probe the glass 00:07:40.479 --> 00:07:46.160 yeah aside from that no i just want to know if it thermal throttles 00:07:45.360 --> 00:07:53.840 well we're not going to have the mounting pressure ah that's fine i can just push on it 00:07:50.319 --> 00:07:55.280 show me the blender 00:07:53.840 --> 00:07:59.520 that looks like good news CPU temperature is actually reasonable 00:07:57.360 --> 00:08:02.720 35 is way too high yep thermal throttling i don't think 00:08:00.960 --> 00:08:06.240 it's making proper contact uh that's fine i can just push on it 00:08:04.479 --> 00:08:10.160 all right i sanded off the side let's see if it's a 00:08:08.000 --> 00:08:15.120 contact error or if we just have a different problem 00:08:12.400 --> 00:08:20.800 so it looks like it actually is working 00:08:18.160 --> 00:08:25.919 but we just need a little bit of coolant flow within the container 00:08:23.759 --> 00:08:29.120 and that dramatically affects our CPU temps 00:08:26.879 --> 00:08:35.120 and then once it's moving the heat from the CPU causes more 00:08:31.680 --> 00:08:37.839 convection yeah we are turboing to 4.0 00:08:35.120 --> 00:08:42.800 4.1 gigahertz across all cores right now so we may be in business although 00:08:40.399 --> 00:08:47.519 i have a new problem um that i identified if we actually wait until the 00:08:45.360 --> 00:08:52.720 coolant reaches 80 degrees no matter which CPU it is it will have 00:08:49.360 --> 00:08:54.880 long thermal throttle before then 00:08:52.720 --> 00:08:58.160 right right so 00:08:56.160 --> 00:09:02.480 i was trying to avoid this but we are just going to do this 00:08:59.839 --> 00:09:07.600 the simple way with a very short loop we've got a reservoir a pump and a CPU 00:09:05.200 --> 00:09:11.519 block no radiator no fans because remember we want our water to absorb the 00:09:09.680 --> 00:09:16.399 heat and we want to track the changes in temperature over time 00:09:13.760 --> 00:09:20.560 now we do introduce the variable of our pump actually giving off some heat into 00:09:18.560 --> 00:09:25.360 the loop but given how significant our CPU is as a heat source 00:09:23.600 --> 00:09:28.880 i don't expect it to cause too many problems with that said i predicted lots 00:09:27.839 --> 00:09:32.080 of things today that weren't going to cause too many problems and 00:09:31.279 --> 00:09:35.480 yet here we are 00:09:36.000 --> 00:09:43.760 so are we going to run the reservoir without the cap on yes 00:09:39.600 --> 00:09:46.000 why don't we start with 500 mils 00:09:43.760 --> 00:09:48.399 that should be a pretty good amount for this loop actually that might be too 00:09:47.519 --> 00:09:51.839 much 400 mils here we go 00:09:55.200 --> 00:09:59.920 ah crap is it leaking 00:09:58.720 --> 00:10:05.279 no it's just not a lot of water so that was what 250 00:10:02.560 --> 00:10:08.480 mils it's a cup yes 00:10:06.720 --> 00:10:14.640 something went right so let's do a quick cinebench r20 run 00:10:11.200 --> 00:10:16.959 our coolant temperature is 16.5 degrees 00:10:14.640 --> 00:10:21.680 which means we've still got time we are turboing over 4 gigahertz we're 00:10:18.800 --> 00:10:26.640 at about 4.2 and our hottest core is about 51 degrees right now so this is 00:10:24.480 --> 00:10:31.200 exactly the behavior that we are looking for so 00:10:29.120 --> 00:10:35.519 we've decided we're gonna formally start our graphs 00:10:32.720 --> 00:10:39.200 once the CPU hits 100 load and it begins actually 00:10:36.720 --> 00:10:42.800 completing the render so that put us at it somewhere in the neighborhood of 00:10:40.480 --> 00:10:47.440 around 28 degrees so now all we got to do is strap in and 00:10:44.880 --> 00:10:51.279 see how long this takes before it gets hot 00:10:49.120 --> 00:10:56.640 so it's been about uh five minutes of actual render now and 00:10:53.279 --> 00:10:58.959 we're up to about 43 degrees we need uh 00:10:56.640 --> 00:11:01.920 we need another 35 degrees and then theoretically this will 00:11:00.399 --> 00:11:05.839 turn black it'll be pretty sick 00:11:04.000 --> 00:11:11.680 unfortunately we only have about another 25 degrees to go before we thermal 00:11:07.839 --> 00:11:15.200 throttle so it might go a little slower 00:11:11.680 --> 00:11:17.360 for that last 10 degrees yeah 00:11:15.200 --> 00:11:21.680 what's it currently sitting at 75-ish what's the clock 00:11:20.399 --> 00:11:26.079 4.2 4.1 4.2 00:11:24.399 --> 00:11:30.160 that's a little high i think it's 00:11:27.760 --> 00:11:32.720 multi-core enhancement enabled oh 00:11:31.279 --> 00:11:37.600 well we were planning to run it twice anyway weren't we yeah 00:11:35.360 --> 00:11:41.279 oh wait well hold on a second we were chit-chatting there and well 00:11:39.200 --> 00:11:46.000 this color change die sucks because it starts changing colors at like 00:11:43.200 --> 00:11:50.320 60 degrees well unless it got hotter in the blog so why don't we just ride this 00:11:47.680 --> 00:11:53.519 out to 100 degrees when the CPU starts really thermal throttling and then we'll 00:11:52.399 --> 00:11:59.440 just plot it yeah so we are on the brink of thermal 00:11:56.640 --> 00:12:01.279 throttling now our coolant is at 68 degrees 00:12:00.480 --> 00:12:06.800 uh core number two is at 99 degrees we're actually still running at 3.8 to 3.9 00:12:05.360 --> 00:12:10.560 gigahertz so uh 00:12:08.560 --> 00:12:14.639 pad respect for that i guess but we're gonna have to shut this experiment down 00:12:12.399 --> 00:12:19.600 pretty quickly of course though we would be remiss if we didn't wait until 00:12:17.360 --> 00:12:26.880 uh we got another 0.3 degrees on here and then we can let it 00:12:22.839 --> 00:12:29.279 end so we made yet another observation 00:12:26.880 --> 00:12:31.839 this time it was that the settings that our CPU was running at 00:12:31.040 --> 00:12:38.720 weren't really stock and they weren't really multi-core 00:12:36.000 --> 00:12:43.040 enhancement either what the why is this a greater volume of water this time 00:12:41.120 --> 00:12:46.959 i could have sworn it was oh wait did i say 350 mils last time i might have 00:12:45.360 --> 00:12:50.880 missed a hundred mils there let me put in exactly 100 mils and see if that 00:12:49.040 --> 00:12:55.200 makes up the difference nope that weren't it 00:12:53.240 --> 00:12:59.120 absorb absorb my pretty 00:12:57.120 --> 00:13:02.800 this is not going to be in the video okay so a couple rough things happened 00:13:00.959 --> 00:13:08.079 yesterday did this system just hard reset what 00:13:06.480 --> 00:13:12.720 all right so we learned a lot yesterday we got this cool graph of our water 00:13:10.320 --> 00:13:17.120 temperature over time and all that good stuff but even just on Intel alone 00:13:15.120 --> 00:13:22.560 that's not the complete story because yesterday we were running the default 00:13:20.240 --> 00:13:26.959 behavior so Intel's stock with multi-core enhancement disabled but a 00:13:24.560 --> 00:13:33.600 multi-core enhancement allows your CPU to boost for a longer period of time so 00:13:31.040 --> 00:13:37.519 for this run here we have enabled multi-core enhancement and we're going 00:13:35.120 --> 00:13:41.279 to go ahead and hit it again once we hit 22 degrees 00:13:39.279 --> 00:13:47.519 round number two here we go we're turboing to 4.7 4.8 ish 00:13:45.279 --> 00:13:51.040 seems good we're not expecting a difference here though we're just 00:13:48.639 --> 00:13:55.839 expecting it to stay at its maximum turbo frequencies for 00:13:53.200 --> 00:14:00.800 longer come on baby five start blending me 00:13:58.560 --> 00:14:06.079 six seven eight nine there it goes 27.5 okay good enough 00:14:04.800 --> 00:14:10.480 all right let's see how she heats up 00:14:08.639 --> 00:14:16.639 now this is weird we're only about a minute into our render and our slowest 00:14:13.440 --> 00:14:18.240 cores are already sitting at 4.1 00:14:16.639 --> 00:14:22.240 gigahertz that's not very enhancing is it no 00:14:20.720 --> 00:14:26.880 that's like slightly behanced still i don't remember 00:14:24.720 --> 00:14:31.120 seeing 4.1 at this stage of the game last time around no 4.1 didn't happen 00:14:29.680 --> 00:14:35.680 until like i want to say 10 minutes in 00:14:32.880 --> 00:14:38.880 okay so as you can see our liquid turned dark and that means that this test is 00:14:38.079 --> 00:14:42.880 done uh it hit 69 degrees like our last one 00:14:41.600 --> 00:14:47.519 at about 16 16 and a half minutes for now it 00:14:45.600 --> 00:14:51.440 looks like multi-core enhancement did nothing more than increase the thermals 00:14:51.440 --> 00:14:55.600 weird now i just need to take my ice water 00:14:54.480 --> 00:15:01.920 fill it up and fire it up we are using xmp i figure 00:14:58.480 --> 00:15:03.680 that with AMD cpus anyway 00:15:01.920 --> 00:15:08.240 the performance of the CPU and therefore how hot it gets it's going to be broadly 00:15:05.839 --> 00:15:13.360 determined by the speed of the RAM so if we use 2133 memory or something 00:15:10.800 --> 00:15:17.680 like that it might not be a fair test it's actually a little funny it's not 00:15:14.639 --> 00:15:21.279 warming up as fast as the Intel CPU did 00:15:17.680 --> 00:15:23.519 so 21.5 i'm going to stop prime95 switch 00:15:21.279 --> 00:15:27.000 back over to blender and then when it hits 22 i'm gonna start 00:15:25.600 --> 00:15:32.959 it off 4.1 gigahertz and only 00:15:30.079 --> 00:15:35.920 42 degrees on our coolant that seems 00:15:34.399 --> 00:15:40.639 pretty similar actually 00:15:37.760 --> 00:15:44.240 similar now that i look at it yeah but it seems to be tracking a little bit 00:15:42.399 --> 00:15:49.199 lower a little lower so we're going to need the full length 00:15:46.160 --> 00:15:51.519 of our run in order to find out 00:15:49.199 --> 00:15:56.000 whether or not AMD is actually 00:15:53.440 --> 00:15:59.360 outputting less heat here okay so we're almost 00:15:56.959 --> 00:16:05.600 11 minutes into our test at this point our water is currently at about 56 00:16:02.399 --> 00:16:07.040 degrees celsius now if we check and see 00:16:05.600 --> 00:16:13.519 where we were at 11 minutes on our 9700k 00:16:10.720 --> 00:16:17.680 it was actually 58 so we're still a couple degrees colder 00:16:15.600 --> 00:16:23.360 it's starting to look like the uh the processors are are separating 00:16:20.959 --> 00:16:28.639 we're at 63.5 degrees and 00:16:24.240 --> 00:16:30.240 15 minutes in Intel was at 66 degrees 00:16:28.639 --> 00:16:36.560 so it looks like even given AMD's 00:16:34.160 --> 00:16:40.959 higher rated TDP their CPU is actually outputting less 00:16:39.839 --> 00:16:44.800 heat one thing that we probably won't graph 00:16:43.279 --> 00:16:51.279 just because that wasn't really the point of the video but that we thought was interesting at the end of the test 00:16:48.399 --> 00:16:55.839 Intel had rendered around 300 tiles and change whereas even now 00:16:54.240 --> 00:17:01.839 almost complete on the AMD side of things we've done nearly 1900 so 00:17:00.399 --> 00:17:06.000 that's uh well that's that's rough that's rough 00:17:04.160 --> 00:17:11.280 what's our current time at our current time is at 20 minutes and 23 seconds 00:17:08.480 --> 00:17:16.720 okay we're at 67.2 degrees so yes our AMD processor did output less 00:17:14.160 --> 00:17:21.679 heat so then why is it that our results seem to differ from what both AMD and 00:17:19.600 --> 00:17:25.839 Intel state on the packaging the answer is that it comes down to how 00:17:23.600 --> 00:17:30.480 the two companies measure TDP Intel measures TDP based on 00:17:27.679 --> 00:17:34.880 recommendations for cooling solutions assuming the processor will run at its 00:17:32.400 --> 00:17:39.360 base frequency in other words their number is intended as a guideline that 00:17:37.039 --> 00:17:45.360 system integrators can use to avoid thermal throttling not necessarily to 00:17:42.720 --> 00:17:50.880 allow the CPU to turbo up to its maximum all the time by contrast AMD measures 00:17:48.480 --> 00:17:54.960 their TDP as the maximum power a processor can draw for a thermally 00:17:52.960 --> 00:17:59.280 significant period while running a typical load and that might seem like 00:17:57.200 --> 00:18:04.080 the same thing but thanks to precision boost it's not AMD's ryzen cpus will 00:18:02.080 --> 00:18:10.240 attempt to run at the fastest possible speed at all times similar to a modern 00:18:06.960 --> 00:18:12.400 GPU and this is where AMD measures their 00:18:10.240 --> 00:18:16.480 TDP so if Intel were to measure their TDP this way they would actually have to 00:18:14.559 --> 00:18:21.120 measure it with their turbo boost limiters disabled which would make it 00:18:19.280 --> 00:18:24.880 measurably higher now the thing is 00:18:23.120 --> 00:18:28.160 i don't think either of us is in a position to propose like an 00:18:26.400 --> 00:18:32.880 industry-wide uh way of 00:18:29.679 --> 00:18:35.200 standardly measuring TDP clearly we're 00:18:32.880 --> 00:18:40.000 not in that position it'd be nice though what we do know is that having everyone 00:18:37.679 --> 00:18:44.960 go their own way on this is harmful to consumers who use this spec as a way to 00:18:42.720 --> 00:18:48.080 choose between one product and another so in the long term hopefully the major 00:18:46.799 --> 00:18:52.320 players in the industry can come together and standardize on a way of 00:18:50.160 --> 00:18:56.320 measuring but in the meantime the best advice we can give you is to just ignore 00:18:54.640 --> 00:19:01.039 it unless you're comparing apples to apples so like within Intel's own 00:18:59.440 --> 00:19:05.840 product lines and even hopefully the same product families and rely on 00:19:03.840 --> 00:19:09.360 independent reviews like our recent video card buyer's guide which actually 00:19:07.360 --> 00:19:13.919 Anthony hosted they should go check that out right yeah sure heck yeah so thanks 00:19:11.440 --> 00:19:16.960 for watching guys hope you enjoyed it see ya