As always we’ll also take a quick look at synthetic performance, though as GTX 780 is just another GK110 card, there shouldn't be any surprises here. These tests are mostly for comparing cards from within a manufacturer, as opposed to directly comparing AMD and NVIDIA cards. We’ll start with 3DMark Vantage’s Pixel Fill test.

Pixel fill is traditionally bound by ROP and memory throughput, but with enough of both the bottleneck can shift back to the shader blocks. In this case that’s exactly what happens, with the GTX 780 trailing GTX Titan by about the theoretical difference between the two cards. On the other hand it’s very odd to see the GTX 680 get so close to the GTX 780 in this test, given the fact that the latter is more powerful in virtually every way possible.

Moving on, we have our 3DMark Vantage texture fillrate test, which does for texels and texture mapping units what the previous test does for ROPs.

Unlike pixel fill, texel fill is right where we expected it to come in compared to cards both above and below the GTX 690.

Finally we’ll take a quick look at tessellation performance with TessMark.

NVIDIA’s tessellation performance is strongly coupled to their SMX count, so the high number of SMXes (12) on the GTX 780 helps it keep well ahead of the pack. In fact we’re a bit surprised it didn’t fall behind GTX Titan by more than what we’re seeing. On the other hand the lead over the GTX 580 is right where we’d expect it to be, showcasing the roughly trebled geometry performance of GTX 780 over GTX 580.

Crysis 3 Compute


View All Comments

  • kilkennycat - Thursday, May 23, 2013 - link

    Fyi: eVGA offers "Superclocked" versions of the GTX780 with either a eVGA-designed "ACX" dual-open-fan cooler, or the nVidia-designed "titan"blower. Both at $659 are ~ $10 more than the default-speed version. The overclocks are quite substantial, 941MHz base, 993MHz boost (vs default 863/902) for the "titan" blower version, 967/1020 for the ACX-cooler version. The ACX cooler is likely to be more noisy than the "titan", plus it will dump some exhaust heat back into the computer case. Both of these eVGa Superclocked types were available for a short time on Newegg this morning, now "Auto Notify" :-( :-( Reply
  • littlebitstrouds - Thursday, May 23, 2013 - link

    Being a system builder for video editors, I'd love to get some video rendering performance numbers. Reply
  • TheRealArdrid - Thursday, May 23, 2013 - link

    The performance numbers on Far Cry 3 really show just how poorly Crysis was coded. There's no reason why new top-end hardware should still struggle on a 6 year old game. Reply
  • zella05 - Thursday, May 23, 2013 - link

    Just no. crysis looks way better than farcry 3. dont forget, crysis is a pc game, farcry is a console port Reply
  • Ryan Smith - Thursday, May 23, 2013 - link

    On a side note, I like Far Cry 3, but I'd caution against using it as a baseline for a well forming game. It's an unusually fussy game. We have to disable HT to make it behave, and the frame pacing even on single GPU cards is more variable than what we see in most other games. Reply
  • zella05 - Thursday, May 23, 2013 - link

    there has to be something wrong with your testing? how on earth can 2560x1440 only shave 1fps of all those cards? impossible. I have dual 580s on a dell 1440p monitor and I can say with complete conviction that when playing Crysis 3 you lose at LEAST 10% frame rate. Explain yourselves? Reply
  • WeaselITB - Thursday, May 23, 2013 - link

    There are two 1080p graphs -- one "High Quality" and one "Very High Quality" ... the 1440p graph is "High Quality."
    Comparing HQ between the two gives 79.4 to 53.1 for the 780 ... seems about right to me.

  • BrightCandle - Thursday, May 23, 2013 - link

    Both of your measures taken from FCAT have issues which I will try to explain below.

    1) The issue with the 95% point

    If we take a game where 5% of the frames are being produced very inconsistently then the 95% point wont capture the issue. But worse is the fact that a 1 in 100 frame that takes twice as long is very noticeable when playing to everyone. Just 1% of the frames having an issue is enough to see a noticeable problem. Our eyes don't work by taking 95% of the frames, our eyes require a level of consistency on all frames. Thus the 95% point is not the eqvuialent of minimum FPS, that would be the 100% point. The 95% point is arbitary and ultimately not based on how we perceive the smoothness of frames. It captures AMDs current crossfire issue but it fails to have the resolution necessary as a metric to capture the general problem and compare single cards.

    2) The issue with the delta averaging

    By comparing to the average frame time this method would incorrectly categorise clearly better performing cards. Its the same mistake Tomshardware made. In essence if you have a game and sometimes that game is CPU limited (common) and then GPU limited the two graphics cards will show similar frame rates at some moments and the faster of them will show dramatically higher performance at other times. This makes the swing from the minimum/average to the high fps much wider. But it could be a perfectly consistent experience in the sense that frame to frame for the most part the variation is minimal. Your calculation would tell us the variation of the faster card was a problem, when actually it wasn't.

    The reason that measure isn't right is that it fails to recognise the thing we humans see as a problem. We have issue with individual frames that take a long time. We also have issues with inconsistent delivery of animation in patterns. If we take 45 fps for example the 16/32/16/32 pattern that can produce in vsync is highly noticeable. The issue is that frame to frame we are seeing variation. This is why all the other review sites show the frame times, because the stuttering on a frame by frame basis really matters.

    We don't particularly have issues with a single momentary jump up or down in frame rate, we might notice them but its momentary and then we adapt rapidly. What our brains do not adapt to rapidly is continuous patterns of odd delivery of frames. Thus any measure where you try to reduce the amount of data needs to be based on that moment by moment variation between individual or small numbers of frames, because big jumps up and down in fps that last for 10s of seconds are not a problem, the issue is the 10ms swing between two individual frames that keeps happening. You could look for patterns, you could use signal frequency analysis and various other techniques to tune out the "carrier" signal of the underlying FPS. But what you can't do is compare it to the average, that just blurs the entire picture. A game that started at 30 fps for half the trace and then was 60 fps for half the trace with no other variation is vastly better than one that continuously oscillates between 30 and 60 fps every other frame.

    Its also important to understand that you analysis is missing fraps. Fraps isn't necessarily good for measuring what the cards are doing but it is essentially the best current way to measure what the game engine is doing. The GPU is impacting on the game simulation and its timing and variation in this affects what goes into the frames. So while FCAT captures if the frames come out smoothly it does not tell us anything about whether the contents is at the right time, fraps is what does that. NVidia is downplaying that tool because they have FCAT and are trying to show off their frame metering and AMD is downplaying it because their cards have issues but it is still a crucial measure. The ideal picture is both that the fraps times are consistent and the FCAT measures are consistent, they after all measure the input into the GPU and the output and we need both to get a true picture of the sub component.

    Thus I am of the opinion your data doesn't currently show what you thought it did and your analysis needs work.
  • rscsrAT - Thursday, May 23, 2013 - link

    As far as I understood the delta averaging, it adds the time difference between two adjacent frames.
    To make it clear, if you have 6 frames with 16/32/16/32/16/32ms per frame, you would calculate the value with (5*16)/((3*16+3*32)/6)=333%.
    But if you have 6 framse with 16/16/16/32/32/32ms per frame, you would have 16/((3*16+3*32)/6)=67%.
    Therefore you still have a higher value for a higher fluctuating framerate than with a steady framerate.
  • WeaselITB - Thursday, May 23, 2013 - link

    For your #1 -- 95th percentile is a pretty common statistical analysis tool ... I'm assuming that they're assuming a normal distribution, which intuitively makes sense given that you'd expect most results to be close to the mean. I'd be interested in seeing the 3-sigma values, as that would further point out the extreme outliers, and would probably satisfy your desire for the "1%" as well.

    For your #2 -- they're measuring what you're describing, the differences between individual frametimes. Compare their graphs on the "Our First FCAT" page between the line graph of the frametimes of the cards and the bar graph after they massaged the data. The 7970GE has the smallest delta percentage, and the tightest line graph. The 7990 has the largest delta percentage (by far), and the line graph is all over the place. Their methodology of coming up with the "delta percentage" difference is sound.


Log in

Don't have an account? Sign up now