Gadgetory

we have some really interesting thermal testing for our GTX 1080i review today that's the spin right here torn to pieces as it so happens we'll be going through those metrics thermals versus frequency mostly alongside the usual gaming benchmarks when we have some synthetics linked in the description below in the article where there's a bit more depth on everything about this card before getting to review this coverage is brought to you by Thermaltake and their tower 900 display case this case has glass tiles all around and is an excellent option for a full custom water loops which if you're buying a car that's $700 you might be in the market for link in the description below for more information going over price first again the 1080i is a $700 MSRP partners can ship above that but the founders Edition still called that the reference card basically is still going to be 700 so they've removed the special founders edition price that costs how much confusion with a 1080 launch that's a good thing this is 700 dead for the EFI version and MSRP with that announcement came the announcement of reduced MSRP for the GTX 1080s there are already today on Newegg Amazon wherever $500 GTX 1080s including $530 EVGA SC models if you're looking a little bit overclocked or gigabyte models for 500 flat so the 10 80s are already dropped 1080i is on the way and AIB partner models of this card were reviewing today should be available within a few weeks pretty shortly actually because this has been kind of in the wings for a while now in the meantime we're looking at the reference version all of these cards ultimately are built on the new GP 102 die if you wanted the full identifier its GP 102 - 350 - k1 and that's for the die underneath the cooler of course and we've already taken the car departs for a few project what you'll see over the next few days there's a block diagram of the gtx 980ti it shows it to layout broken into 6g pcs or graphics processing clusters in a similar fashion in some ways to how the GP 100 is broken up though with some pretty important differences in SM layouts the focus is removed from anything about FP 32 with the gtx 980ti and we have 128 FP 32 CUDA cores 4 SM just like all the other pascal cards in the gaming line when counting a total of 28 SM times 128 we're at 35 84 CUDA cores in total the GTX 1080 GP 104 chip host:20 sm's which totaled at 25 60 cores so are about a thousand cores over the GTX 1080 that seems to align with Nvidia's marketing claims that the 1080 Ti increases performance by 35% over the 1080 that launched in May of last year that will be validating those claims throughout this video the rest of the block diagram is familiar we're faced with 224 T mus and a base clock of 14 80 megahertz or boots 1582 and that results in a textured filter rate of 350 4.4 giga tackles per second at the boost clock we have a video on the channel already if you're not sure what that means a total of 88 Roth's are present on the card compared to 64 on the GTX 1080 a bit of an upgrade there and if you start totaling these things up GT 102 in some ways looks familiar to the GP 100 it's got the same TMU counts the same GPC count but they're still again significantly different in their targets GP 100 of course is built for acceleration not for consumer graphics certainly not for gaming in any way and this means that precision on G to 100 splits itself between 32 16 64 capabilities whereas one of two focuses on F P 32 applications GP 102 is also a bigger die than the GP 104 which was 314 millimeters squared that's the 1080 and GP 102 measures in at 471 millimeters squared which helps accommodate the additional sm8 additional SM so bigger die area since we're getting on a plane in about 6 hours from filming this we're going to jump straight to the benchmarks when the benchmarks today we're using a fully rebuilt GPU test bench for 2017 I believe this is the first I know this is the first GP Review other than the IC X card of 2017 that gives us an opportunity to move to the i7 7700 K which has a higher top end more headroom in frame rates than the 59 30 K we were using previously and it also means that we can refresh our suite of games and other applications with being tested including some new automation done by Patrick Lathan in-house that's really exciting now this also means for all the excitement's with a new testbed we lose some data that's unfortunate so everything on the previous tests can't really compare it anymore it's a different platform different memory different drives different testing methodology for the games of Doom we've changed how we test it a bit and we've added new games so that means that we lose things like the Titan XP comparison since that was on loan to us so we don't have that for you today unfortunately but we do have new numbers for the 1080 1070 980ti reference cards the 1080 TI reference card and then 1r x 480 just to throw it in there as a representative of the $250 market so in the article below we'll have the testing methodology as always we'll also have numbers for future market testing like 3d mark fire so I can tie this by moving forward we're refreshing our GPU bench with the relevant ions contenders first as listed with a splash from the $250 market represented again by the 480 note also that before he competes in a completely different price bracket so we're not really going to pay too much attention to it it's just there in case you want an idea of where stains fall well ad Vega of course once it's ready and we'll slowly add more GPS to this bench with each content piece including aib models if you're curious about those and one final note here because of the limited time between pax Rison and other events we are going to be visiting overclocking and noise testing separately so theirs won't make it into this piece unfortunately but they will be in a piece post pax our thermal tests are up first for these benchmarks for measuring MOSFET temperature in the middle of the board and at the bottom of the board so that will be stats number two and seven on the 1080i on the IC x GT X 1080 or if that's number one and five on the GTX 1080 Fe card-counting bottom up and those are measured with our thermocouples they are K types they're calibrated in boiling and iced water as normally and we have some more information on exactly what those things are in the article let's start with just the 1080p I then add other cards for now we can see GPU temperatures in the 84 to 85 C range with MOSFETs number seven and two closely packed around 65 to 68 C this is actually heavily competitive to a ivy partner set temperatures but the card is also only stock clock loads and the GPU temperature of course looks dismal when compared to IB partners the card requires a significantly higher fan speed in order to sustain a comparable GPU temperature to what will be possible on a IV partners and we can say like confidently because of previous experience with the 1080 and of course you sacrifice noise output I mean 1080 CIF II version if you want to achieve at those temperatures to this end despite using high quality materials externally and internally the cooler is actually not all that great for our audience it's a better fit for system integrators high power cyber power at all and if you have specific needs maybe SFF it could fit there as well we're still recommending axial or liquid coolers for the most part though because you're choking on the clock in a way which is unnecessary with the EFI cooler that's where IB vendors come in of course to fix things here's a comparison to the GTX 1080 Fe for frame of reference to the last gen of coolers from the Pascal lineup we're seeing the GPU temperatures on each device locked to their max permissible limit when under stock settings with the FET temperatures all within the same range that's actually a good thing the GT x 1080 FPS middle MOSFET does run about a Celsius warmer one degree Celsius or so than the GTX 1080 T is middle MOSFET but the card cools effectively identically went under a high-stress workload and of course these aren't the same FETs they're not in the same place so that does impact things as well the measurement is not perfect what matters is the frequency throughput though and that is a perfect measurement since both cards are boosting to a point of hitting thermal limits looking at the clock rate versus temperature we can reveal that the 1080 Ti is able to sustain a higher more stable clock when four Smith's fan to 80% this enters into territory of unbearable noise and does not representative of a sustainable user scenario without a noise damping box of some kind maybe shoved into a basement but it does show how the silicon behaves when temperature is no longer a concern the GPU frequency on the 1080i with Auto fan speeds or around 50% shows that we're bouncing rapidly between about 1341 pretty low to about 15 and 44 megahertz by contrast 10 atti at manual 80% fan speed post frequency almost at a constant 14 80 megahertz with occasional spikes to around 15 44 note that this is with AVR empower virus scenario so we're seeing higher thermals despite a lower clock overall that is expected behavior and this clock rate is not representative of what you would see in a video game just to stick one of our ancient charts back up here that we've now changed methodology to accommodate set temperatures these temperatures that show at a steady state so we're looking at how the GPU performs in terms of thermals once it has been left alone under heavy load for a long period of time the 1080i reference ends up around 50 Hz delta T for steady state temperature when under torture workloads end with an auto fan control the FE version of the 1080 meanwhile switch on 57.5 C as well within our measurement variance and so we must say that the 1080 F II alongside the 980ti and 1080 TI are effectively equal in thermal performance Nvidia claims 5 Celsius improvements with the 1080 TI layout but we can't exactly validate that on the same PCB which is important because one GPU binning to different GPU three different drm we can't validate it exactly because we don't have a means to fit both coolers on the same car that leads within a reasonable amount of time the changes to fetes v rms all that means it's not exactly an apple savile comparison but we are not seeing a 5 Celsius reduction stock you might see it if you match the clock rates or something like that if you're curious how n AI be partner board compares our previous EVGA icx review gives a good look at those numbers in general for thermals you should expect the same high-ish to high temperatures for these reference designs for the 10a DTI as with previous reference designs from both Nvidia and Andy and video is not alone here neither is AMD they both make reference designs that are versatile they're good for sis they're good for other sectors of the market but for our audience for most of you doing system builds DIY style it would be better off with an AIB partner model that cools quieter and better or at least in the very least quieter for the same and that's really not a bad trade-off this cooler is not particularly impressive but neither was the 1080 at these cooler despite that Nvidia is marketing did show in its presentation when Jensen was on stage in San Francisco they showed a twenty thirty eight maximum clock rate with temperatures below eighty five see that may be possible in their test environment we were not able to replicate that in our testing without making the fan speeds so high that they make your ears bleed which may be the case just like with the 1080 in Austin it may be that the card was in a box somewhere in a back room at 100% fan speed because then I could see those temperatures being possible but otherwise I don't know that the 50% more bearable fan speed will make that a reality if you do need a reference board for liquid blocks for example consider waiting for board vendors to ship their stripped-down versions of the same PCB with disposable plastic coolers because those might be a bit cheaper or include rebates at least to drag costs out of it moving on to FBS we're starting with a brand new game to our bench Ghost Recon using the built-in benchmark and set to very high preset at 4k 1440 and 1080p we are looking at GPU performance first with Forte at 4k the 1080 TI pushes an average frame rate of 55 FPS with one-percent lows at 45 and 0.1% lows at 42 comparatively the 1080 F II card runs about 10 FPS lower in all values for a difference in average frame rate of approximately 29% that's not exactly the 35% number that was thrown around everywhere but it's pretty damn close the gtx 1070 reference card lands around 35 FPS average for the same test for a performance improvement of about 58% when moving from the 1070 to the 1080 TI as for the last gen comparison the 980 TI reference card post performance around 34 FPS average at 4k which is really dragging into territory of being stuttery in this title you can get away with 4k lower settings of course but not with this setup shifting to 1440p we see the GTX 1080p i operating at 91 FPS average with lows three times at 74 and 66 FPS that looks good in terms of the frame time consistency the GTX 1080 is just under 20 FPS lower than that at 74 FPS average with lows also fairly close at least to the Ti at 61 and 54 the improvement here from the 1080 reference card is about 22.5% as for the TTX 1070 efi the performance improvement is about 47 FPS from the gtx 1070 to the gtx 980ti FP cars moving us between 62 and 91 average with a 980 TI at 1440p we're seeing a performance improvement of around 60% generationally that's at 57 FPS average certainly playable the RX 480 is on here but again with these settings the unit very high and 1440p it's not really targeted to benchmark these mid-range cards bit unfair if you're looking at 1440p for a mid-range card you'd have to drop your settings and it would still be playable just drop the setting but doom is where we really start seeing things get a bit interesting our doom benchmarking has ditched OpenGL for comparison entirely mostly to make time for other tests and instead it pushes to strictly rely on Vulcan for the test api of choice we're also now using ultra settings for graphics with anti-aliasing disabled and async turned on so you get that extra boost you can find more in the article below at 4k Doom positions the 1080 Ti at around 90 FPS average with lows close behind and in the 70s the GTX 1080 F II card stands around 68 average resulting in an improvement of about 33 percent to the ti the 1070 was never really a 4k gaming device but it's holding on respectively well with doom at these settings were just under 60 FPS average and could make tweaks to go over 60 the difference is about 60 to 64 percent improvement with a 1080p eye but of course that price is also about a two-act to increase the gtx 980ti performs at 49 FPS average at these settings and is mostly playable with the album zooms engine optimizations you can make a few tweaks though looking now at 1440p the GTX 1080p eye is bumping against the limitations of Doom's amorous and that is an engine limitation doom limits itself to 200 FPS due to a physics bug and so frame rates that exceed 200 FPS aren't really seen there cat the excuse numbers in a way that makes the higher end cards capable of exceeding 200 look a little bit lower in the chart and just like a bottleneck would do the only difference here is that rather than a CPU or GPU bottleneck we have one that's not too commonly discussed and engine bottleneck we see these in GTA 5 for quad thread testing at 180 7.5 FPS as well for similar engine reasons we're starting to really illustrate generational differences here with a 980 I've nearly doubled in performance by the 1080p I just to really illustrate this frame cap and doom let's look at 1080p with this our average is a steady 200 FPS with lows no worse than 175 your 1% the GT X 1080 isn't far from this I 190 and throwing effectively equally since they're both bottlenecks by the game engine the 1070 hits 200 less frequently and is met early behind of course you would really for the most part play at these types of settings unless you have a 200 Hertz display but it does show generational scaling better to 980ti since that was even more common when it shipped moving to for honor one of our new titles added we're testing with extreme settings across all up three main resolutions at fork an extreme the 1080i is finally able to crack 60fps my games though just barely read 66 average with 60fps one percent load 57 0.1% loads if you get a 1080 to similar Heights give it an aftermarket card and some tuning but we're talking about a drop of maybe one tick and setting to get there the gtx 980ti is entering unplayable territory with these settings right alongside the RX 480 you would need settings reductions or just really honestly drop the resolution and doing that dropping the 1440p the story changes a bit now something like in our X 480 would be an excellent choice at its price point roughly 250 if you wanted 1440p gaming with pneumatic settings a slight reduction would help frame times but we can have a 1080i today anyway and the gtx 980ti for a generational look at that still holds its ground well against the new Polaris and Pascal cards but posts an improvement of about 50 for FPS by moving the to the 1080i or around 70% the GTX 1080 is plenty playable here with a nearly 100 FPS a virgin type same time using ashes for another dx12 look configured the high settings GP focused on Forte we're seeing the 1080 TI land at 99 FPS average width to 1080 reference at 79 or about a 25% improvement to the 1080p I really not hitting that magic 35 because I'm figure in this test though the jump from the 1070 is naturally a lot higher though so is the jump in price again looking at an improvement of about 62% to the Ti our final benchmark game is Sniper Elite 4 this was recommended by AMD for the bench we spoke to both Nvidia and AMD and gave them both an option to recommend one game for the bench and video like Ghost Recon and D Lite Sniper Elite so we added each although we're not looking in depth at AMD cards today that time will soon come at 4k and with high settings with async enabled and dx12 the 1080 TI operates a frame rate of about 76 FPS average pretty damn high for 4k but sniper isn't the most graphically intensive game on the market the 1080 reference cards has had a hard 60 FPS with it lows dipping into the 40s this is still just fine for gameplay and we're not able to start seeing more visible stutters until jumping to the 1070 considering 4k is more of a 1080 or TI class config the 1070 it holds well on its own at 48 we'll have to see how Andy does on this bench shortly as it should be favorable for the Polaris and Vega GPUs given an async and the x12 focus given that we're sandwiched between PAX East and a few hours and rise in coverage and the risin event and the Nvidia event and other events this review still has some things I'd like to add to it like noise noise is one of the things we're overall we're not recommending an F II reference design anyway if you were to buy a 1080 TI it should probably be a partner model so the noise tests would be nice to further back that up and show things objectively but they'll be have to be added after pax in addition to this you may have noticed that overclocking wasn't a major focus we have one overclocking number with firestrike extreme on the website but I was having some stability issues during overclocking in the last few hours before getting on this next plane so we scrapped it it'll have to come in the next content post probably right after pack so there are still a few things to look at overclocking was the same process as always it just didn't happen to be stable when we were working with this card and that's really a matter of some more fine-tuning so we need time for that stock performance overall though the GTX 980 is still more than enough if you own one of those don't bother upgrading to a 1080p iOS you have a very specific need it's more of a side grade almost you're able to split the difference and trade in with a manufacturer Nvidia sort of is competing with itself by dropping the GTX ndnews in price anyway it's a $500 since that card can also handle 4k with some settings tuning and handles 1440p without much trouble at all the bigger argument you could make would be for high refresh gaming as the 1080i fills that niche where others have fallen short though have gotten quite close 1440p at 144 Hertz is doable in some of these games like for honor if you're okay with dropping too high or very high from extreme in that case assuming no anti-aliasing for owners of a gtx 980ti this is a sizable upgrade and in some cases it's a pretty damn big jump if you want those higher resolutions you're not on high resolutions maybe don't bother there's still mileage left in the 980 TI it's a 780ti that's the most interesting to look at but again unfortunately because of time that wasn't added yet still based on just extrapolating performance that's probably where an upgrade makes the most immediate sense it just comes down to what resolution you're trying to use so many reasons buy a found edition card as always is going to be the reference PCB there will be other reference PCB cards but this is the first one if you want it they'll be out there EK WB already has plans for water blocks compatible with this PCB the 10a TFE water blocks won't work it's a different board different VRM layout but there are plans already for water blogs in the works we've spoken with them about that otherwise wait on partner models for cooling reasons and for potentially custom PCBs potentially better overclocking we've also got build Zoid working on a vrm analysis of this board this is quite a bit different from the 1080 so for this board we're looking at AV RM that is doing some doubling actions we've got to fats per phase which does spread out the heat a little bit so that would help explain why the vrm temperatures on this board are actually really good there can parable to an icx video card the ones EVGA just launched and that's a good thing the place that falls behind is GPU cool and as expected and as seen in the past but the vrm design does spread out heat a bit and helps in that scenario as for whether it's worth buying this is sort of a weird market position right now where Nvidia exists largely uncontested and these last major cards in this type of market where the fury X and fury cards and the rx 480 is the newest thing that they have it's an extremely strong competitor at the 200 250 as price point that's a pretty good winner it holds that market very strongly while Andy is waiting on Vega so they're in a good position for the mid-range they just don't have anything of a high-end yet which makes this a weird thing to review because it's not a lot of competition so what we can say is that the 1080 Ti if you don't need an upgrade now today if you're not building a system this week or something like that it's probably worth looking into if you're looking at a high budget card that said the 1080 does sort of compete with the 1080 Ti if you don't need those 144 120 Hertz refresh hits you could drop down to a 1080 it's still a damn good card and you save a lot of money you can put that money towards something better in the system like better SSDs or something like that if you do need to upgrade today there's no shame there should be no remorse in buying a 1080i 1080 something like that Vega will gone we will review it it's just a matter of one at this point we know first half and that's really all we know in terms of specifics there's not out there so Vega will be compared to eventually the RX 480 so looks good at the 250 dollar price points we would recommend that in that category the 1060 is good but look between the two and figure out what games you're playing because they have advantages the 480 at its price is very competitive 1070 is still good 1080 school good 1080i is great for its gains it doesn't always have that 35% number but overall it's a good card is just a matter of getting better cooling on the things that can perform to its full potential which is something we're doing to can't tell by this setup so as always go to patreon.com/scishow and was next directly gamers Nexus net for the full review like comment subscribe all that stuff quick shout out for anyone who watched 3440 by 1440p benchmarks 1440 I guess if you want there's altra wide benchmarks one of our friends of the site tech gauge comm will have that stuff he does it where I don't just die which I come out if you want to ultra wide stop I trust his numbers we'll see you all next time you