Gadgetory

everyone welcome back to another episode of ask GN I sort of did these in real life recently I helped our contributor Patrick stone who's a high school teacher teach a computer engineering high school class about video cards and GPUs so some of the questions for this one will actually come from that class but we've got a bunch from you as well from discord and we might have one or two from YouTube leave your questions in the comment section below if you have any for next week the big ones though what are your hopes present - how do you tell the difference between video cards in terms of what actually makes them good and then we've got a couple on thermals that are really interesting like does the temperature scale linearly on the CPU versus the ambient temperature before that this video is brought to you by thermal Grizzly makers of the conductor not liquid metal that we recently used to drop 20 degrees off of our coffee lake temperatures thermal grizzly also makes traditional thermal compounds we use on top of the IHS like cryo not and hydro not pastes learn more at the link below so the first question comes from discord I don't have the user's name written down but they asked what's gns collective hope for the tick Zen cores next year which I would assume is N+ and looking to 2019 what about Zen - also could you do a newer video or feature on memory performance and the differences between dirt-cheap 2133 2666 and more mid-range or high-end 3,000 plus megahertz memory so addressing the latter part of that first the memory stay tuned we just spent a lot of time testing memory all the way up to 4500 megahertz we're just sifting through the data and we're rerunning every single one of those tests on another motherboard now because they work differently on different boards depending on the memory support of the boards so we're working through that it's a really cool feature getting to forty five hundred megahertz was really hard actually it took a lot of manual tuning because you run into an issue where not only are you now bidding for better memory which you can just buy off-the-shelf that if you want to spend a ton of money on it but you're also betting your CP you for a hike or clock and a high IMC performance so to get both of those things is pretty difficult if you want both the memory controller to be good enough to handle something like forty six hundred and if you want the memory itself to the handle forty six hundred forty five hundred and you want the cord o'clock high but as for Zen to and these n plus stuff so the the answer here what is the hope for it first of all I try not to have hopes about any product ever it honestly I don't really speculate a whole lot or or kind of like create a fantasy world of performance for different products because we're so invested in the day-to-day operations and testing that yeah I don't really think that far ahead when I'm trying to predict performance for the future that said if I were to say what I wanted to see in Zen I think one of the most valuable things that Andy could could achieve with the next Zen implementations would be higher clocks and they've got the core account down pretty well they're competing with Intel right now Intel won't be able to truly respond in a serious way probably until late next year and in terms of core count that is and they responded with coffee like pretty well but it's clear that there's more there to do in terms of catching up with just raw core count and the price to per thread performance when you look at applications like blender but AM D has a lot of catching up to deal with pure gaming and there's room for them to do it so Zen was a major change from FX especially in terms of IPC fortunately it's also the first step of this architecture so the refinements you would expect to see increased frequency this is typically what we see it historically with other CPUs from Intel and AMD alike what I would like to see specifically is I would love to see a 4.0 base or very close to it and I'd love to see a four point three to four point five gigahertz boost on the the core series competitors so the r3 r5 our seven four four five I don't need a four three that would be great anything in that range a couple of hundred extra megahertz up to four point three boost would be amazing and maybe even they just get to call it 4.2 or 4.3 then they have some XFR on top for single threat for a single core performance that would be great so that's what I'd like to see that's on my wish list what I don't think we'll see is one of the slides I saw recently which I think may have been faked or was I don't know a very misunderstood leak if it was an actual leak but I saw a slide somewhere that's had people in the comments at least suggesting a possibility of five gigahertz and 12 cores for I suppose what would effectively be an r7 replacement I don't think that's gonna happen I could be wrong and I would love to be proven wrong but five gigahertz is like a 1 gigahertz increase and you're talking about increasing the cores that's that would be an amazing engineering feat again it's possible but I definitely wouldn't put my hopes on that 5 you're talking about going up to the speeds that Intel can achieve with heavy overclocking so even if it happens in the very least you're throwing all power efficiency out the window hundreds of watts to get up to that kind of speed in that core account and Rison is pretty good at being power efficient it also just like a lot of other architectures on the market does start losing that power efficiency as you push the overclock to its limit that's kind of what you expect with overclocking but yeah I don't think 5 gigahertz is happening probably at all I don't think 5 gigahertz with 12 cores is happening and when I say at all or not happening I don't mean ever I mean I don't think it's happening with Zen Plus and I'm not sure about Zen to know one is but I don't I don't think I would anticipate that but we'll see I have to see more information on Zen tea that's that's a while out at this point so it's possible but Zen plus definitely wouldn't expect it so yeah my my wishlist is a higher base and boost and then beyond that it's more of logistical things than actual hardware so better communication with motherboard makers would be fantastic and he needs to kind of you know you saw this with X 399 and risin where there were two issues with boards the one rise in first launched you could get as many CPUs as you wanted for the most part but you couldn't get any motherboards Intel's having the opposite problem so it was an issue of the board manufacturers didn't have enough time from the time that the final micro code was submitted to the time that the press had samples motherboard makers had less than a month to get their BIOS finalized and you have to factor in the time to ship the samples and organize everything - so two weeks to get everything ready so that was a big problem and that impacted launch it also impacted things like memory support memory sub timings all this really small stuff that added up and hurt and these initial launch and they improved a whole lot of it really cleaned up but that could be improved with better communication and that is both on Andy and on the motherboard makers I think the motherboard makers at this point know that AMD is serious now so I think there's a better chance of Zen PLAs and Zen to having more out the gate support in terms of actual supply and quantity of boards and probably a more serious ear lended when it comes to planning the product and making sure it works with everything so I think it's an issue of board manufacturers maybe being in a position of fool me once and not wanting to invest a whole lot in it and of AMD not necessarily knowing how to communicate what their needs were to the board manufacturers or convinced them of a good product hopefully a lot of thats fixed just by being tested with time another thing here better education for media so we kind of went through this ourselves where every time we asked AMD for use cases for things like thread representing you know you'd run into scenarios where it's like okay so I get it it fast but if I can kuta accelerate this task where GPU OpenCL accelerated then what's your use case give me an example of something that I can test to show that this processor makes more sense in this application than a GPU does so they're pretty bad at answering that and we found a lot of answers on our own for example with blender if you do an orthographic render with an image that effectively looks 2d at the end where you're rendering tiles very quickly it matters more from what we've seen for the most part to have a higher tile count in flight actively simultaneously than just to have a GPU pushing it because with thread Ripper you get 32 tiles in a flight simultaneously on the 1950 act and that's really valuable but no one told us that we had to find it ourselves unfortunately we can actually find that stuff because we use blender in-house but if you don't use blender in-house as a reviewer and you just use it as a benchmarking tool you might not know so that's something where Andy could improve on his education for media there's a lot of other examples of that too where ffmpeg h.264 encoding was something we found at thread Ripper really good at and we don't do GPU a transcoding for our compression needs so thread Ripper was a great use case for that and so is our 7 but had to find it ourselves so those are the things and you can improve on its communication and education for their partners and then hopefully some frequency boost as well next question this one is from supa who says when discussing delta T over ambient results for cases coolers etc is it relatively accurate to assume that if I ran an identical test in my home with a different ambient temperature I get the same a delta T basically if a CPU heats up 50 C in a room at 25 C to a final temperature of 75 degrees is it safe to assume it also heat up 50 degrees in a 35 degree room to a final temperature of 85 degrees this is a great question so we're assuming a few things here if you've ever taken a physics class they probably made some kind of joke like assume a spherical cow or something like that Mythbusters did that one is well the reason you do that is because obviously to discuss this we have to assume that when you say you're running an identical test that you also have our identical hardware because the variants even in silicon can change those numbers so let's assume that let's assume that we shipped you our set up the the question just to recap is if my room is 25 and your room is 35 will the temperature increase exactly by 10 degrees in your room in their final readout because your room is 10 degrees warmer and the answer is kind of but not really so this is something we've spoken with VSG from thermal bench about this in the past months ago and I think included part of his answer previously but what it comes down to is the more recent launches from Intel from Haswell onward behaved a bit differently than some of the others so before has well it was more or less a one-to-one just like you're asking with Haswell onward so once you're at steady-state the core temperature will increase almost the same as ambient so that's that kind of follows the pattern but the IHS temperature is not the same increase with ambient at steady-state that's where the difference emerges so when we spoke with vs geothermal bench a while ago he observed that a CPU verse IHS temperature increased variance was about 1.4 to 2.2 or two point four degrees Celsius on a Haswell sample and that was tested with a hot box that I believe the SG configured between I don't know what the low end was but let's just call it 30 C and I know the high end was ADC so that was his ambient change this is something 280 pretty big so speaking with him it sounds like this is a result of thermal resistivity of the heat spreaders alloy having a hyperbolic function with temperature so it starts to run away a bit in other words as the ambient temperature increases which means no it's not perfectly linear like we would like but the CPU core temperature somewhat follows that curve at least for a while and as you do increase ambien it does get more extreme and exaggerated so basically the ambient temperature it to an extent we can basically write it off as an assumption of these will be pretty close they will be at steady-state pretty much a one-to-one gain in each but once your ambien starts getting let's say I I'll just give what our tolerance is if it starts going above like 3 degrees different that it's time to probably consider alternatives the testing you might be able to get away with 5 but as you increase ambient more and more it will run away more and more three to five degrees isn't really gonna hurt anything realistically within the other resolution of the test like the measuring equipment and the software measuring but once you start getting beyond that it does get a bit questionable next question is from I'm a Jedi bra who says vrm on motherboards if you use the same CPU to test multiple boards will the vrm of each board help determine the volts needed to run the chips table at overclock better vrm equals a better OC question mark or just help with temperatures to spread out the power and such so for this one there is an extent to the usefulness of how high end you go with V RMS and motherboards there are definitely diminishing returns at some point once you start looking at a $200 board versus a $250 board a lot of people would have trouble finding a difference in terms of the max table overclock the differences are elsewhere often there is definitely a difference to some extent with the boards though so an example of this would be one of the other z 370 boards we've been working with with the 8700 K so on actually X to 99 boards as well on some boards the better ones I'm often able to get the same clocks as the lower end but with a little bit less voltage some of this comes down to the vrm itself some of it comes down to the vendor and their BIOS implementation for example the load line calibration setting has a slightly different curve with each vendor and when you need to start using that whether it's a flat line or a slightly curved up line or slightly curved down line that impacts the voltage going to the CPU ultimately and the B droop so that's part of it vrm quality is part of it spreading the heat definitely if you're using for example be 350 board with one of the worst heat sinks and comparatively one of the worst heat sinks and you're using you have limited airflow you can definitely make it throttle we've done it pretty easily actually but that impacts your overclock through an indirect means thermal throttling so as for ignoring thermal throttling things like that the answer is yes kind of you might be able to get a slightly higher clock or more likely you get about the same clock and you have better voltage tuning options if it's a better BIOS but uh yeah I spoke with builds worried about this too and he agree agreed that there's a limit to the extent of how much you actually gain in terms of noticeable clock and voltage differences when overclocking on the high-end stuff but there is a difference the other thing is how much power can i push so if you're if you're someone like their Bower and you know you're gonna push nine hundred to a thousand watts into the CPU you better have a good idea and understanding of what the motherboards of erm can tolerate and what kind of cooling you need to tolerate it whether that's liquid nitrogen or something else so there are definitely places where it matters a lot more but for the the non Allan to user there's a limit next question is from Darth this one's really quick Darth says I've been given conflicting advice when recording gameplay on the same hard drive as the game you're playing assuming 70 to 200 rpm over SATA 3 is there an issue of affecting the video quality when using something like shadow play the answer is basically no not with shadow play not with I can't remember how to pronounce it now relive a real live relive right yeah and these relive that name was an awful choice yeah I think so I don't remember whatever it's called I know I know when we did the first video on it there's no difference maybe with frat well yeah with fraps so with perhaps you will with hard drives especially it'll bog down the drive and so if the game is on that drive you will definitely hurt the framerate we've seen it and fraps basically records losslessly it's hundreds of megabits per second so that would be an instance yes but shadowplay don't worry about it RJ fleming on ask GN number seven that's going back you said you use the same bench to control for tested devices ie CPU GPU heatsink etc does that mean you also do not take driver BIOS or OS updates to ensure data is not contaminated for example 1070 TI on 388 0 9 but the comparison with the 1080 was surely on an older driver just curious not much can be done without rerunning the entire nvidia lineup on a newer driver unless you did that in which case you guys live in the lab also do you use system images to keep though eyes clean we do live in the lab I especially live in the lab but if you so the last question do you use iOS M system images to keep those clean we do so when I worked at Dell we learned we had a really good system of pixi booting where you would do boot over Ethernet and so we'd pixie boot into a server and then you stored all of the system images at the time we use the Norton Ghost and Norton Ghost would allow us to clone the drives often including the drivers sometimes you'd do a bear clone and then not not a clone of a bear but a bear as in Baron clone and then pull that down so that allowed us a lot of it afforded us a lot of convenience and flexibility where during testing something goes wrong it's the same thing here where if your OS gets contaminated and it will over time the drivers get dirty or updates break things or whatever you can just pull down an image no big deal it's like 10 minutes to do to pull down the image and you just start over with a completely fresh OS all the test tools installed the entire environment is ready to go so it's like a 10-minute clean os+ set up you can't get much better than that so yeah we keep system images specifically to make sure that there are no issues with dirty drivers because that happens a lot as well and a bad driver ruins the data it's not even just the driver it's even if you use ddu which we do it's possible that something gets left behind and messes up the results or maybe the chipset drivers get messed up or when you're swapping motherboards and CPUs and GPUs and RAM windows has a tendency to tie itself into knots so this is what we call driver gremlins and testing where you don't necessarily know what's wrong but something's wrong one of the easiest ways to create this problem is to move a drive from an AMD platform to an Intel platform or vice versa and observe the performance disparity and that's just because Windows 10 has gotten better at it but absolutely I would not do that for testing we have isolated in tellen a.m. the images for that reason the rest of the questions do you control for BIOS yeah that one's pretty easy to control for we only update BIOS if it has something we actually need or if we're specifically reviewing BIOS and I have another folder on the server with all of the BIOS is backed up so we can pull them if we need to go back for some reason if we need to rollback drivers drivers we control for as best we can the thing with drivers is you kind of you have to look at what change so we keep close contact with AMD and NVIDIA so that we can ask what changed in this driver and then we do some tests on our own so it's normally 1 or 2 ad hoc tests just a dry run through everything make sure nothing's changed that hasn't changed then we can keep that data just a little bit longer and if it has changed clearly something needs to be rerun so the difference with something like a 1070 Ti on 380 809 is that often the newest video card launch with its press drivers that we use internally the they sometimes get iterated on before they go public the press drivers we use internally are a repackaged version of the most recent previous driver just with included support for the card almost never there's a difference for performance it's just basically they add support for the card so fortunately we're able to use the previous driver which we probably used for another test and then use 388 oh nine or whatever it is for the 1070 TI that said every now and then a driver release completely wipes the results clean destiny to had such a big driver update from Nvidia recently that we can't use that data anymore for new GPU benchmarks until we rerun the things that are relevant so what I do obviously part of what we do is try to retain a large amount of data for comparative purposes and then as stuff ships I will rerun tests with a more limited suite of card so destiny to again as an example because I know that's updated I will next time I need to attach the game I'll rerun it with probably the card I'm testing plus its competitors plus and minus one price category and then we'll leave everything else out and it sucks to lose some of that data but that's that what things like firestrike are there for things like Metro last light that's why you keep some of the old games so that you always have scaling data that drivers will never impact and then then you can update the most recent games with the most recent drivers so there's a good spread of here's the recent real use case and as an academic study here's some scaling information if you're trying to figure out how it works with whatever and that answer is the other question people ask a lot which is why do you still test some of these old games that's why it's there's nothing wrong with having an old game or two on there it's not taking anything away from you and it adds data for people who need it for older cards because just testing the newest stuff doesn't help people people are trying to figure out do I need to upgrade from my three year old device yes or no not well people who are building a brand new system needs to know how the new stuff performs but a big part of the market isn't hopefully that answer some there's so much more information to that but I'm just going to stop there if anyone has more questions about it let me know below and I'll take them next time and I think there's a let's take this one a question from so this is from the class that I helped teach down GPUs so the big takeaway I wanted them to have at the end of the day was how to identify what makes cards good in general because these are our high school students they kept asking me which card is best for GTA 5 what's best for destiny - it was always what's best what's the best video card and if you take that question at the face of it the answer is a 1080i or a Titan XP you should probably shouldn't buy those though especially now if you're a high school student who's either sponsored by their parents or work in a low wage job probably so that's the hard answer but you know the teach this is all stuff all of you know with teaching how do you address your needs my needs are this price this game do I care about 3d rendering whatever but the next thing I try to teach them was how do you look for and identify what makes a card good because I'm talking about between AIB partners here not between GPUs because everything I tell them about the 10 Series today will be useless in a year when they have the money to build a computer because the 10 series and rx5 80s will probably be sort of slowly vanishing from the market as new things roll out whenever that may be so instead of focusing on the rx series on the 10 series what we did was to talk about cooler differences and things like that where you identify a good video card by hopefully looking up some reviews where they've done actual thermal testing we do a lot of this and power and noise testing so it's important to identify what am I using the product for how much money do I have I try to figure that out before I start looking and how much am I willing to flex on the price and then you're looking at ok with my needs maybe I want silence maybe I want overclocked ability then you go and look for the card and and this is stuff most of you know however looking for a good cooler isn't that simple as pick the biggest one because as we learned with the zotac amp extreme card the biggest cooler isn't the best cooler so it does take looking for reviews and if they don't exist for what you want you can kind of look for photos of like a side view of the card and try and look down to see where this where's the contact between the cold plates and the vrm components are there thermal pads where there should be is there an unnecessary amount of plastic does it look like the cooler itself is mounted with hardware to metal plates like the back plate and the base plate in multiple locations or does it look like it'll flex a bit how does it look for sag things like that that's kind of what you look for and then on component level you look for videos from people like build Zoid to figure out if those will survive long temperatures and long up times for overclocks so really basic stuff but the kind of fun thing with that experience was how the wide spread of the knowledge level in the class there were people who had never built a computer and didn't know what a gtx 1070 TI was which I named because it just came out and there were people who had built their first computer already so it was actually really difficult to kind of like step back enough where everyone is following you but still provide depth which I guess is sort of what we do here just added more with it with a baseline assumption of knowledge that's a bit higher so that's it for this one if you have questions for next time leave them in the comment section below or on discord which you can join by joining our patreon group patreon.com slash gamers Nexus for that you can pick up a shirt like this one at store that gamers Nexus Totten that we presently have this anniversary edition model on sale and of course subscribe for more I'll see you all next time