Gadgetory

the numbers are in for the gtx 1070 founders edition video card despite being in taipei taiwan this week for Computex which is coming up in a few days we managed to pull off a cold benchmark and review of the card with all the usual FPS thermals and architecture analysis before getting to that content all of this coverage for the first few days a Computex is brought to you by msi and the new x99 a gaming motherboards including the gaming pro carbon here is a spec stable first off check the article for full detail on the gtx 1070 we've compressed it heavily to fit into a video format but you'll get additional game tests and detail in the article the gtx 1070 operates on the GPU 104 GPU from nvidia starting the pascal architecture the reference GT x 1080 also operates on a GP 104 but they're differentiated by subversion 8 and the 1070 is AGP 104 - 200 that's the important bit the 1080 is the GP 104 - 400 the silicon is the same the architecture is mostly the same where the diet has been somewhat simplified on the gtx 1070 to reduce cost the heart of the chip is still the 16 nanometer FinFET design which operates at a slightly lower voltage than the planar process previously and exhibits less power leakage than planar data path optimizations are also in place for performance improvements something we've spent a few thousand words in our GT X 1080 review covering if you're curious about that the first major changes in 1070 is SM count the GP 104 200 GPU has 15 SMS rather than the 20 SMS GP 104 - 400 because of the SM reduction TM use are naturally also reduced the 1070 operates on 120 TM use and those can pipe 202 Giga texels per second core counts is affirmed 1920 cuda cores operating at a boosted clock of 1683 megahertz stock the 1070 also uses gddr5 not 5x and that's at 8 gigabits per second with a capacity of 8 gigabytes time to dive into the benchmarks first I know that's what everyone wants to see right away as always all the methodologies to find on the website links in the description below along with additional game benchmarks that aren't shown here if you're curious about how we test those things here's a look at the temperatures for the GTX 1080 and gtx 1070 founders Edition cards we've also got the 1080 hybrid on here which is a card that we built custom from the scraps of a 980ti hybrid and a desecrated 1080 reference card or founders edition card the gtx 1070 operates about equal to the GTX 980 reference with hitting fifty three point eight six Celsius the delta temperature over ambient which is a few degrees cooler than the gtx 10a TFE at fifty 7.51 celsius that extra few degrees will actually help give us some extra Headroom in overclocking we'll discuss later on as you'll see in our three-part hybrid build series where we explore the temperature limitations The Fury exits at thirty six point three nine Sal's heaters a bit warmer than the other liquid cooled cards but that's because puri cools its vrm and memory with liquid as well not just the GPU the 1070 is about six point five five percent cooler than the GTX 1080 entirely resultant of the die simplification and here's a look at the thermals over time charge this data is pulled from the same test as the peak load temperature average and shows the test data over time the Nvidia founders edition cooler isn't particularly impressive but it is reasonable in the very least the FE cooler performs about where previous reference models have performed but as always AIB partners will make significantly more effective coolers on their cards that be cooler operates with a maximum fan noise level of about 47 to 50 decibels which is reasonable when compared to other devices just a reminder these are Delta values this means we subtract ambient so add your own ambient temperature back in when you're looking at these numbers of your house is about seventy F 20 Celsius you add in 20 Celsius to these numbers that gives you the operating temperature of the GPU in your environment and we removed Delta we do the Delta value for a few reasons start subtracting ambient helps control for the natural variance with the HVAC system between tests and it also obviously helps you compare on your own ambient temperatures and so ours is not the same as yours so here's a brief endurance chart this shows the frequency over time versus temperatures as you can see the 1070 sustains a pretty stable frequency throughout the endurance of the test you can read more about this in the article link in the description below time to talk about game benchmarking now we've got a mix of the usual OpenGL dx11 apps but we've also got some of the new DX 12 and Vulcan applications and we're gonna start off with doom which is brand new from its author and runs on OpenGL with Vulcan support coming in the immediate future doom is one of the few games where the G 1070 performed slower than the gtx 980ti and raw frame throughput the two cards trade blows depending on game and the 1070 often matches evenly or out matches the gtx 980ti the previous flagship the gtx 980ti runs about 11.5 percent faster than the gtx 1070 and raw frame throughput at 1080p the gtx 1080 ties with the 980ti only because we've hit a CPU bottleneck and the 1070 sits below the CPU throttle points hitting nearly 115 FPS average the tightly times 1% and 0.1% low frame metrics 88.3 and 72.7 respectively against andy's current high end at single GPU card the gtx 1070 outperforms r9p reacts by about eight point eight percent or one hundred fourteen point seven FPS on the 1070 versus 105 average on the pure ax against the 970 nvidia successor the 1070 pushes an additional 12 point seven percent of frames per second 1440p plants the gtx ten saudi firmly within playable range running above 70 fps and 4k begins to stress the card a little too much for a perfect fit with regard to frame times and fluidity moving on to the division it appears that 1080p testing is too easy on the gtx 1070 it doesn't choke the card for framerate until he moved to higher resolutions at 1440p the gtx 1070 runs at about eighteen point four percent slower than the GTX 1080 and about even with the 980ti the tent the 970 sits at fifty five point seven FPS which is twenty three point seven percent slower in FPS than the ten seventy at 1440p 4k is a little too intense for the gtx 1070 in the division though the 1080 has a lot less issue with it and remains the only single GPU card that can regularly handle 4k gaming at very high ish settings and the 1070 could play in 4k but you have to drop to medium and lowest settings looking at shadow of mordor the 1070 holds a strong second place following the 1080 by 17% and leading the 980ti by eight point three percent at 1080p that gets switched around as the resolution increases with 1440p establishing a massive lead on the $700 1080 though at these frame rates it really becomes somewhat irrelevant until stepping into 4k anyway and pushes the 980 InP reacts beyond the 1070 the 1070 still sits at 77 FPS fully playable and marks itself and easily find improvement over the 970 at forty eight point seven FPS or forty-five percent different and it's still barely affordable as well let's move on to dx12 benchmarks so ashes of singularity is the first of those benchmarks for simplicity of presentation we've broken into a few charts with ashes of singularity the first one shows dx12 is dx11 performance all this data is from the satellite shot to a part of the benchmark and videos gtx 1070 founders edition lands in second place on the 1080p chart with 10 FPS disparity between the brand-new $700 1080 and the $450 1070 that's a fourteen point two seven percent delta for a couple hundred dollars the GTX 970 runs approximately thirty five point one seven percent slower an average FPS than the ten seventy and DirectX 12 mode quote unquote and these are 9p reacts push significantly lower frame rates with dx11 something we analyzed and the 1080 review you'll find out again below and the fury ax eliminates the GTX 10 seventies lead once resolution has increased this chart next shows the edges of singularity frame x and that's the latency between frames this is comparing dx12 to dx11 performance gains between architectures so you're looking at four percent delta versus the previous generation versus the previous api and as shown and videos moved to asynchronous compute algorithms and compute preemption has assisted tremendously in its performance gains over the maxwell generation will explain compute preemption and NVIDIA async computes in our 1080 review so if you want to learn about that again hit that content the 1070 sees benefits and architectural changes to data path organization and compute preemption ultimately yielding an additional 28.3% frame time improvements over DX 11 this isn't quite as good as the GTX 1080 is near 50% gains but it's still pretty damn good Andy of course it looks about 120 percent gains the X 11 to 12 with the pure e^x but a lot about is just because of Andy's tremendously low performance in DX 11 next up is Vulcan with the Talos principle at 1440p the gtx 1070 performance still stands above the 60fps mark well above it in fact with 110 FPS average on Vulcan and 119 average on dx11 and even manages to hold marginally about 60fps performance at 4k the eighty performance for GTX 1070 runs 12.2% slower than the GTX ten 82.6% faster than the 980 I reference and that's a difference that pre overclocked 980 eyes would overcome the last gen Maxwell GTX 970 runs about 37% slower than the 1070 we've made a big deal out of overclocking on the new Pascal architecture with our GC X 1080 review and part of that is because we toured the thing apart and through a liquid cooler on there to see just how far we can actually push it once the thermals were accounted for so the things that changed here first of all it's on boost 3.0 now which is a new version of Nvidia's boost technology and that does change things a little bit for overclocking including the introduction of scan OC on EVGA precision which means that you'll be able to sort of semi automatically scan for and set your overclock including a frequency versus voltage curve that's brand new we don't use that though we do the tried and trued true sort of basic overclocking method which is really just slider tuning so you tune the core clock memory clock and then the voltage as necessary along with fan RPMs or just throw a liquid cool around there and that's how we do our overclocking for these tests but let's look at the table with our stepping results for overclocking this table shows our progress stepping as we tested each modification to the GPU our final tested results burned in with an endurance test and stayed strong in 1987 megahertz core 4608 megahertz in memory and that's a 215 megahertz overclock on the core and 600 megahertz overclock on the memory which only seemed limited by V BIOS and voltage regulation at this point not necessarily an uncommon thing while you're looking at now are the performance charts for that to overclock some games saw more than a 25 percent performance increase but the range was a more reasonable 9 to 30% GTA 5 at 4k benefited from a 12 point 6 percent gain on the 1070 with a 1080 benefiting 14 point 8 percent on the EFI cooler mordor at 4k a twenty nine point nine percent hike at 1080p and at 1440p that was still 22.3% with the 1070 overclocked doom at 1080p moves into the 120 Hertz playable range with a nine point two percent jaunt to 125 point 7 FPS average the GP 104 - 200 it shipped runs cooler by nature of its simplified die and it's lower overall clock rates but there are other things to look at too so the 1070 trades blows with a 980ti and that difference could be made up for in some cases by a pre overclocked AIB version of the 980 I so do keep that in mind overall the 1070 is a good improvement over the 970 from the previous Maxwell generation and that improvement measures about maximally maybe 30 percent but we do see a pretty good range and they're depending on a lot of things like the game but generally you see about a 30% Delta between the 970 and the 1070 until pushing it to high resolution so 1440p is a good example of that the 1070 creates a new emergent class of 1440p ready cards for 60fps plus gaming and this is something that the 1080 did for 4k as well where it's finally exceeded 60fps on a single GPU with the 1080 and the 1070 does that for 1440p as for the founders edition Carter taking the same stance of this as we took with 1080 founders Edition card and that is you should wait for the AV models so the AIB models we priced and the 350 to 450 dollar range brothers the fe card is priced at $450 so that price difference is enough to justify the purchase of an AI v card because it's a good good amount of different you'll save more money put it towards something else the IB cards of course will also be better overclockers in some regard when looking at things like heat versus noise so if you're trying to reduce your noise levels but also overclock that sometimes difficult with the reference cars because the blower fan has to increase and its RPM so that's a big thing to look at I would suggest waiting for the AIB cards other than that the 1070 is a good chip it performs well for us where you would want it to be and if you're playing at 1440p that makes it exceptionally good because nothing really exists at this price point I can do 1440p right now so in that regard this is setting a new standard for Nvidia I would not buy it for 1080p you're good buying something cheaper for that because this is just overkill for 1080p in almost all instances so that is our gtx 1070 review as always thank you for watching full article links in the description below for more information pay traveling the poster hall video if you wanna helps out directly with your contributions over time and of course a big banks to Mike dag Lyon Andy Berke Ethan gallic Andrew Coleman who all helped pull this thing together remotely across continents while we're in Taipei Taiwan for Computex so thank you for watching I'll see you all next time