Quick test: DirectX 12 API Overhead Benchmark results

Published by

Click here to post a comment for Quick test: DirectX 12 API Overhead Benchmark results on our message forum
#5038662
https://forums.guru3d.com/data/avatars/m/55/55855.jpg
Results from both PCWorld and PCPer are in. PCWorld’s results show the difference in API draw call handling between both the Nvidia GeForce GTX Titan X and the AMD Radeon R9 290X. With DX11 the Titan X manages a maximum of 740 thousand drawcalls per second and a whopping 13 million, 419 thousand calls with DX12. In comparison the 290X manages a maximum of 935 thousand drawcalls with DX11, so more than the Titan X to begin with and again an unbelievable 13 million, 474 thousand calls with DX12. So the 290X slightly edges out the Titan X in DX12 drawcall handling.Interestingly enough AMD’s latest drivers have improved DX12 performance to the point where it’s actually a head of Mantle by about 8%. PCPer’s testing is more traditional, showing the frames per second instead of the number of drawcalls per second. The GTX 980 delivers 15.67 FPS in DX12, a massive improvement compared to the 2.75 FPS it manages in DX11. On the other hand AMD’s R9 290X delivers 19.12 frames per second in DX12 a significant jump over the GTX 980. And perhaps what’s more interesting is that the next iteration of 3DMark which will debut with Windows 10 also supports Mantle. And the 290X delivers 20.88 FPS in the same benchmark running Mantle. A 33% performance lead over the GTX 980 running DX11 and a 9% lead over the same 290X running DX11.
290X edging out the Titan X, LOL.
#5038681
https://forums.guru3d.com/data/avatars/m/260/260629.jpg
My Results: 720p DX11 ST 865 781 DX11 MT 799 304 Mantle 14 463 069
#5038749
data/avatar/default/avatar25.webp
I cannot benchmark.When it comes to test DX12 it says API not available.
#5038754
https://forums.guru3d.com/data/avatars/m/80/80129.jpg
It does look like AMD cards gain alot more than Nvidia, and not just because how poor AMD DX11 performance is, a 290x is around 25% higher in DX12 than a 980. Though, reading the last few pages again it seems the results vary so much you can't actually accurately compare them, not even to identical hardware.
When all is said and done it really wouldn't surprise me if you saw 290x's beating Titan X's in benchmarks like this. The proper workload to test DX12 is essentially massive basic parallelism. AMD's architecture accels when it can properly split up the workload to fill the pipeline. In this case the workload is designed to be as simple as possible, which makes it a natural fit for their architecture. Not to mention AMD's bandwidth/fillrate. In some games I doubt DX12 will even have a performance impact.
#5038759
https://forums.guru3d.com/data/avatars/m/258/258688.jpg
That difference is just crazy. And strange how dx11 single threaded is faster than multi, i wonder why is that
HH explained that in the article. In current D3d games (DX11, and earlier), most of the performance used from the cpu derives from the first core, regardless of how many cores the cpu has. In properly coded D3d games, several cores in each cpu are hit for performance, and suddenly single-core performance loses its luster and multicore gets big (finally!) Ironically, a lot of Intel's cpu performance deltas over AMD in benchmarks come from the fact that AMD's single-core performance is not as robust, but AMD's multicore performance is much better when compared to Intel's multicore performance. Kind of an interesting distinction, I think...;)
#5038762
https://forums.guru3d.com/data/avatars/m/258/258688.jpg
It does look like AMD cards gain alot more than Nvidia, and not just because how poor AMD DX11 performance is, a 290x is around 25% higher in DX12 than a 980. Though, reading the last few pages again it seems the results vary so much you can't actually accurately compare them, not even to identical hardware.
Draw calls do not translate into frame rates in a 1-to-1 ratio...;) 25% better in draw calls might translate into a 5% advantage in frame rates--or less, or more, depending on the game, and depending on the scene.
#5038771
https://forums.guru3d.com/data/avatars/m/242/242471.jpg
Here its dx11 single threaded part with most variations ram @ 2400mhz 2.4mill 2.38mill 2.28mill 2.32mill 2.42mill One time it quit in the middle and said dx11 ST part not supported, although I changed threaded optimization and forced 1 frame to render ahead, I think it was mostly 1 frame render ahead fault.. btw in nvidia driver its marked as 3dmark14
#5038781
https://forums.guru3d.com/data/avatars/m/34/34585.jpg
Would kinda like to know how the AMD FX handles it also, i got rid of my FX a little while ago to go Core i5 now i am wondering if i should have stuck with it lol. You can see that the i7 gets a massive increase with HT enabled than when turned off so thats good to see, i can the FX maybe on par or bettering the i7 if that's the case even with it's weaker IPC. The results vary so much due to the processors and their clock speeds once overclocked they go through the roof compared to stock clocks.
#5038804
https://forums.guru3d.com/data/avatars/m/163/163068.jpg
Draw calls do not translate into frame rates in a 1-to-1 ratio...;) 25% better in draw calls might translate into a 5% advantage in frame rates--or less, or more, depending on the game, and depending on the scene.
No, but in most of the games that Intel dominates, it's probably for that reason. No game uses such advanced AI or CPU physics, that it would slam AMD CPUs over Intel's.
#5038814
data/avatar/default/avatar27.webp
I would love to see dx12 on Crossfire/sli systems. Wouldl ike to see how it scales compares to 1 card. i hope we can see much better smoothness. I hope the drawcalls will be almost double of the single card cause for me I'm super excited to see crossfire/SLI performance. I just hope they can get their act together and allow to have different cards in system cause then I can maybe use my old cards without selling it or having it lie around. Anyway can someone test with SLI/crossfire setups so we can see how it scales? I will appreciate it as I don't have pro/advanced edition or windows 10 preview
#5038855
data/avatar/default/avatar26.webp
The DX12 test runs fine if I disable Crossfire. Also tons of other 7970 results with the DX12 test done. As for the test that are working in Crossfire.. 1 card: DX11 single and multithreaded are ~1.2M Mantle 13m 2 cards: DX11 single threaded 1.44M DX11 multithreaded 1.33M Mantle 23.7M
Thanks Cyclone3d, that explaines why my SLI wasn't running the DX12 test. I forgot it was single GPU for DX12. http://www.3dmark.com/3dm/6388063 DX11 Multi-threaded draw calls per second - 2,244,712 DX11 Single-threaded draw calls per second - 2,107,254 DX12 draw calls per second - 12,045,421
#5038866
data/avatar/default/avatar05.webp
The DX12 test runs fine if I disable Crossfire. Also tons of other 7970 results with the DX12 test done. As for the test that are working in Crossfire.. 1 card: DX11 single and multithreaded are ~1.2M Mantle 13m 2 cards: DX11 single threaded 1.44M DX11 multithreaded 1.33M Mantle 23.7M
Really from 1.44m to 23.7m thats insane. If true dx12 will be a game changer. that's 16.45x improvement. i know it's mantle not DX12 but should be more or less the same. I might just buy this now to see what I can get with mantle at least
#5038927
data/avatar/default/avatar37.webp
im confused.. my system scored 12 million calls in the mantle run yesterday, today only 10 million?! has the calculation been modified over night?
#5038952
https://forums.guru3d.com/data/avatars/m/220/220214.jpg
im confused.. my system scored 12 million calls in the mantle run yesterday, today only 10 million?! has the calculation been modified over night?
Since this is all CPU bound it only takes one or to other tasks running on your PC at the same time to slow it. Anyway this is what I got (but CPU was at least 10% busy on other tasks) http://www.3dmark.com/3dm/6381817 DX11 Multi-threaded draw calls per second 431524 DX11 Single-threaded draw calls per second 421451 Mantle draw calls per second 9034655 Seems abysmally bad in DX11 - there was barely anything drawing before it hit 30FPS. This is with previous AMD flagship dual GPU 7990!??!
#5038957
data/avatar/default/avatar21.webp
Since this is all CPU bound it only takes one or to other tasks running on your PC at the same time to slow it.
the faster run yesterday i did with my system running since 5 hours and several tasks running. the run today (tried more than once! even reboots) was pretty clean. so i dont think that this is the answer. btw: this is a cpu benchmark, it doesnt rly matter if u habe a 7950, 7990 or r9 290 running.
#5038964
https://forums.guru3d.com/data/avatars/m/196/196284.jpg
You guys haven't read the main post have you? You need to be using Windows 10 Preview build 10041 to have access to DX12
I have read the entire thread. Both threads actually.
btw: this is a cpu benchmark, it doesnt rly matter if u habe a 7950, 7990 or r9 290 running.
The GPU does in fact make a difference. The CPU won't issue any more draw calls than the GPU can handle. If you're running an i3 2100 paired with a GTX980, the CPU will issue every drawcall that it possibly can. If you're running an i7 5930K paired with an R7 240, the CPU will only issue as many drawcalls as the GPU can handle.
#5038973
https://forums.guru3d.com/data/avatars/m/227/227853.jpg
Would kinda like to know how the AMD FX handles it also, i got rid of my FX a little while ago to go Core i5 now i am wondering if i should have stuck with it lol. You can see that the i7 gets a massive increase with HT enabled than when turned off so thats good to see, i can the FX maybe on par or bettering the i7 if that's the case even with it's weaker IPC. The results vary so much due to the processors and their clock speeds once overclocked they go through the roof compared to stock clocks.
How would an fx surpass an i7, when i7s are better performers?
#5038980
https://forums.guru3d.com/data/avatars/m/220/220214.jpg
The GPU does in fact make a difference. The CPU won't issue any more draw calls than the GPU can handle. If you're running an i3 2100 paired with a GTX980, the CPU will issue every drawcall that it possibly can. If you're running an i7 5930K paired with an R7 240, the CPU will only issue as many drawcalls as the GPU can handle.
Exactly. There is some form of "issue call->wait until call done->issue another call" going on. So the total number done will be dependent on the slowest of the CPU and GPU in a system.
#5039024
https://forums.guru3d.com/data/avatars/m/220/220214.jpg
The little tech nuby i am with stuff like this can i ask a question what this will mean for gaming in general ? Will my 2x 7950 see good improvements in DX12 ? or will it stay the same ? or does DX12 gives a flat performance boost to all games and gpu / cpu performance ? What will this do for games that do not use DX12 ? will you still get a performance boost ?
I'm wondering something about this too... Since in this new benchmark we can see the massive improvement in draw calls in Mantle compared to DX11, then why is it that in Battlefield 4 - which has a specific Mantle mode (instead of DX11) - then why is there more or less NO difference in performance (and actually the last time I tried it, it was way too unstable and crashed mostly) If this Mantle mode for BF4 is anything to go by, then these draw call differences are pointless for current games and we wouldn't see any improvement at all in framerates.

Previous Page

Next Page