Gadgetory


All Cool Mind-blowing Gadgets You Love in One Place

Thermal Testing Challenges (Inspired by Bitwit)

2017-01-29
everyone this video is meant to help out other content creators with their thermal testing and also explain to viewers why temperature testing with cases coolers GPUs is a lot more difficult than it seems on paper this idea came about when I was watching one of fit wits videos Kyle about CPU cooler orientation and it was a great idea we've tested in the past quite a long time ago at this point and Kyle had some interesting findings where he basically said that look here's the data I found but I don't necessarily understand why the numbers are the way they are because I didn't expect it to be that way and that approach just a briefly state here is perfect you can't get a better approach than that to science when you're saying look I've done the testing I've done due diligence to make sure it wasn't something weird as far as I know but I don't fully understand the results here's the data what do you think that's a great way to approach it because the other options are throughout the data what you don't want to do because there might be some really valid thing there that he's discovered that's not being discussed or of course presents it as hard fact which also you don't want to do is you don't naturally understand the data so the reason I bring this up is because I have gone through that so many times the last few years thermal testing up until maybe the last year and a half for us has been a major headache there was a full year where we stopped doing case reviews at all with thermals in them because we were having issues and then brought it all back after I was able to validate our new testing methodology in thermal chambers and really create some numbers that I finally trusted and that was a hard process to go through we've got videos detailing with thermal chambers we've got videos detailing our temperature measurements all that stuff but I want to go through some of the more gritty details today just as briefly as I can because this is a Sunday video for us which means I'm producing it myself the other two guys are off today before getting to that this is brought to you by our patreon backers thank you so those of you who support us you can go to patreon.com/crashcourse maxis to help out we've got a couple goals there we've reached one of the most immediate ones but we're pretty close to the next one so do check out the site if you want to help out Karen's Nexus net or patreon.com slash gamers Nexus and you can support us directly so before getting into the details here any content creators or anyone up-and-coming or established or whatever if you're using any of them at is that we talk about in this video you know give me a shout because this stop has been a long time in the making years and years of work to kind of slowly figure it out and it's expensive to do and sharing with competition obviously we keep some sub secret but I'll put a lot of it out there in this content just you know point folks are ways I would greatly appreciate that let's go over some of the stuff on a table we've got coolers that I've tested we have tools we have thermal paste and we have notes of huge issues I've run into the in the past that have invalidated data or caused major headaches that put us back weeks in production and it's all mostly figured out I won't put everything we do in this video because one obviously competitive reasons but two there's a lot to do so our tests are fully automated which helps the script is internal only but you could write your own script if you wanted to the test environment there's a few things that really matter here one is ambient temperature and we've talked about that in the past another one is the bios settings those are really critical probably more than any other single aspect of thermal testing so let's let's go with ambient temperature first room ambient you want to control the temperature in the room as much as possible if the thermal chamber is not available we don't have some a thermal chamber but we've validated in one so the methodology is known good and that's really all we needed for for this type of testing for room ambient just setting the thermostat is not good enough so if you set your thermostat to say 20 Celsius which is about what ours is they're still going to be fluctuations and some of it could be from vents in the ceiling if a vent is directly over a test bench one day and you move it and it's not the next day that actually does have a big impact and too is ambient fluctuates of course because the things going to toggle on and off throughout the day whether it's heat or forced air or AC or whatever that impact seen and there's really only one good way to control for that which is going to be active logging of the ambient temperature so as we discussed before we use thermocouple readers there's a whole list of caveats with these as well I'm not going to go into these details from training but you use a thermocouple reader you log the ambient temperature ideally you also if do an internal testing like Kyle did with the internal case temperature as well and you do that on a second a second basis so the idea is to create a giant excel file and it's work but it's really accurate you create a big excel file one column is the ambient temperature you've logged other ambient temperatures or probes you blog go somewhere else and then you have your CPU column for either package Recor whatever metric that you trust from the software that's a whole different n2 so once you've created that you can create a delta value and this helps control for the fluctuation and ambience whatever is happening up there from the ceiling vents or floor vents or whatever you can control for that that helps but it doesn't fix all problems and there are plenty of things there that are not quite so straight board either but we did discuss that in one of our previous videos I'll put it in the description or something so a thermocouple readers help we use K types they have about a two point two degrees Celsius range if you're using multiples you will want to calibrate them put a nice water put it in boiling water see what the temperature difference is and then you create an offset in the spreadsheet and just make it build a template around that the next thing is really just bio so CPU the things that matter here when using an Intel CPU especially it's important to disable basically everything you disable see state store power saving states to see table a is T speed step speed shift anything like that disable turbo boost then lock the frequency to a fixed frequency for all cores and then lock the vcore as much as you can and motherboards do have variants here so the motherboard really matters a lot and we showed in our 7700 K review you can have a perks of 20 to 30 degrees Celsius swings if using Auto V core with different motherboards and even fixed v cores that you've configured so fix all the voltage settings you can for the CPU we do one point two eights or one point two seven five depending on what particular cooler and CPU and bench we're working with and that will fix almost all the problems immediately so the reason that is is because when you have a CPU an Intel CPU says I am going to boost as much as possible when the thermals allow and when voltage and power supplies allow so let's say the CP is boosting occasionally it will either have you might have the droop whether that's from your voltage and power supply chain or you could have just a frequency drop and voltage change up or down based on whatever's going on with the test process and of course AVX instructions also can really change things so make sure you understand what those are - that's using prime95 or something like that but with the voltage supply basically with the like with a gigabyte board we tested you can see the voltage go up to 1.4 and then fall to 1.28 and if it's doing that throughout a test process the temperature's going to be all over the place and if you're averaging over say a thousands of cells of data it's not it's just not going to be reliable and I can say that confidently because I've gone through that I know how difficult it is you're sitting there pulling you're in my case many hairs have wondering what the hell is going on why is this configuration that's clearly superior by all measures doing worse than this other one so like an example would be if you tested two same company products maybe a two eighty versus a you know 120 even or something like that if you see temperatures that are really closed or unreliable it's probably because of the accordant cpu setting so all that needs to be fixed or turned off and for anyone who says well as endless cpu is not a real-world scenario anywhere because now it's not out of box why would you test that way well the reason is with thermal testing we're not doing a CPU review so you don't care about how does the CPU do these functions all you care about is how do I make the CPU a steady-state how does it do one thing generate heat and do that one thing with exceptional accuracy and make it reproducible every single test the answer is turn all of that power saving off of the turbo boost and all that stuff off and fix vcore that fixes most problems next thing then so the fans of the bench whether it's open air or in a system they should be connected to power supply and it should be power sight with a voltage that you really trust because if there's fluctuations again if it's a cheapo power supply it might swing from something like 11 to 12 volts and that does the fan speed so that's important to know I don't trust the motherboards generally to control my fans for me some of them do it just fine but we've had too many issues in the past so I go straight to the power supply and what else can we do here software I buy I both note on the sheet software can't be trusted and it's true for a lot of things Andy for example if you're running AMD CPUs for your thermal benchmarks good luck because they use a different temperature scale t TTL and your temperature is with AMD for FX and FM CPUs it will look like it can idle below ambient that is impossible it is not physically possible to have a lower temperature than ambient with any of these coolers on the table weather well yeah any of these click on a table whether it's air or liquid and that's because the the temperature that is ultimately being pulled through the radiator is going to be the ambient temperature so it's impossible so if you see temperatures that when you produce a delta value are negative that means something went horribly wrong it could be a negative 0.5 and that might be a calibration thing with one of these or it could be negative 5 or negative 10 and that's wrong it's not possible so that's that's an AMD thing and that's not just an abuse of software was then this might change but softer pulls data from sensors in the chip and you need to understand what values can you trust for example if kb lake it was it was kind of difficult with all the core temperatures and things the software didn't work quite the way it has in the past for the immediate versions anyway as they updated my change but for kb lake you want to use cpu package through hardware monitor because that's the same as x to you which is obviously trustworthy because in the intel utility so that matters make sure you understand which metric you're using and stick with it whether you're taking all four cores or whatever and averaging them or whether you're taking cpu package I don't use cores with KB Lake I you CP package from hardware monitor or from Ada and then just check it with XT you and make sure your state so that mostly covers software can't be trusted as my bullet points software is also really important though once you've got a base understanding of turn on logging run it for the entire task and generate logs for everything because it doesn't hurt to have more data so we generate logs for CPU frequency at CPU utilization I'm talking about this cooling testing obviously the core all the voltages all the power draw measurements sometimes it's accurate sometimes not depending on the software and the CPU and temperatures and then dump everything into a spreadsheet and the idea is when you do have one of those scenarios really like what the hell is going on here I don't get it you look through the vlogs and you can say oh what's why does the frequency just drop from 4.4 to 4.3 up cause so log files are critically important for that and there's a lot of other stuff I to kind of quickly go through the rest multiple tests in the same location with the same configuration are important so before embarking on any kind of cooler or thermal passing build the test platform change nothing after obviously finally setting changes nothing and run the same thermal benchmark whether using prime95 or whatever else you want to use ADA or any of those tools make sure the test runs are sufficient length to warm up and saturate the cooler and figure out where you're going to start averaging for the load and the idle temperatures and things like that and once you've run that test enough times look at the data and try and understand is it the same what is my variance and if you've got a variance of about a degree and you can't get any more accurate than that which is possible then it's time to add a little note to the charts and say my temperatures I have a variance of plus or minus 1 degrees Celsius and so then you can say if we've got a cooler that's measuring at 51 versus another one at 52 they are effectively identical you cannot do any further validation to prove that one is better than the other and of course functionally the basically the same anyway even if those hard numbers are accurate but if you've got a 1 or - what plus 1 minus 1 degrees Celsius change then just I make a note of that of course hours of 0.5 or so after calibrating for the thermo and all of that stuff a test environment matters too so that would include things like thermal paste open versus closed there I'm not going to go into that today or the there's a couple other themes GPU load things like that so let's let's do kind of talk about go back to Kyle situation Kyle was testing in a system and he is using a utility that loads both the CPU the GPU and some of the memory as unit in heaven and so when you're doing that the the potential thing to look out for is GPU workload because now we could do all these controls I've already listed for the CPU but now if you're involving the GPU it gets a whole lot more complicated we don't involve GPUs for for cooler tests for his Testament but you have to control for one it's got to be the same video card so if you're changing coolers pick a reference board pull the reference cooler off put it on whatever you're testing use the same card and then you've also got to control for of course clock freak like the frequency you have to control for boosting if it's a Pascal cards the whole good luck that's a lot more difficult control for a voltage fan speed power draw and temperature thresholds and things like that I could impact performance ideally you fix the fan speed on the GPU ideally you fix all fan speeds which if I haven't mentioned it yet that's the next most important thing I have to CP voltages every fan you must have a fixed speed for no variants no smart CPU fan header no letting Windows or BIOS or the motherboard or a fan controller or the built-in software for the cooler try and control the fan speed that ruins everything you can do those tests separately but not for the main benchmark but back to Kyle stuff so GPU the back the GPU can radiate heat when we tested the EVGA cards for vrm temperatures we saw that we were able to get the backplate one card up to something like 130 Celsius when we really really really tortured it beyond any reasonable measure by dumping 40 or 50 degrees Celsius air straight into the VRM fan from a liquid cooler so the point is the back of the GPU can radiate a lot of heat in the penny how're your coolers oriented that can impact the CP temperature now that's a valid test environment that's a valid thing to look for but you know it's got to be controlled and understood so the example would be if you are radiating Heat off the back the card and you flip the cooler so that it is now oriented this way it could be sucking hot air in through here and and that would actually increase your CPU temperatures because you're now just yeah I mean you're pushing more heat through the cooler and it's only got so much you can handle before it's kind of saturated so that's something to look out for basically control GPU and everything else with the system everything it's it's really annoying every single thing has to be fixed and you've gotta log it around the test a million times and make sure you understand the the variance for each particular component pretty annoying but great once it works thermal paste so we have their own pace from the same batch that we know same conductivity oh this is getting really really gritty not necessarily that important compared to all the rest because the variance is going to be small here but you know once you've already done all the other work for all this other stuff you might as well so we used with liquid coolers they're all paste applicators this makes sure that we've got a video on it when we apply the thermal compound to any liquid cooler it's exactly the same it's equivalent to coming straight out of a factory and then fixed application method for your air coolers is important as well and of course there's all kinds of oxide variables curing time technically pays can cure we don't really account for that that is one variable I don't care about right now because they're tested and it's not important versus all the other stuff to worry about I think about covers the basics anyone thinking about thermal cameras we've got a whole video on all the problems with using one of these they're great tools but they it's another thing that you got to understand what it's doing so I think that I think that goes through most of it I will add as many images and things as I can to this video as I said it's a Sunday video which means I'm producing it myself so hopefully just the knowledge will help if anyone out there finds this useful please if your content creator just you know let folks know that we we helped out because this has been many years in the making it's been very expensive to develop the methodology and sharing it is a you keep something secret but sharing this much if it's helped you please just let folks know and point them our way so thanks for watching if you have any questions I suppose submit them on the ask Jian video that check those every week tweeted us at gamers Nexus subscribe for more patreon link in the post roll video or at patreon.com slash gamers next to the scammers echoes botnet for the site I'll see you all next time you
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.