computerbaseAshes of the Singularity Beta1 DirectX 12 Benchmarks

[ShintaiDK]

I dont know if thats true, i am telling what i read. Perhaps thats just theoretical numbers and as you say its never going to scale perfectly with number of cores or whatever...but i still expected at least some improvement, i dunno, at least 2x perf increase, not to see the same numbers as with DX11. So all this talk about how superior and novel it is, and how DX11 is archaic and does not use your hardware fully, only to find out that your fully utilized hardware under DX12 produces same number of frames...

You are not going to see 2x increase. Unless you already sit with a very weak CPU perhaps and the game is weak on game logic. Theory and draw calls is one thing as you are into. Actual games another matter.

So its broken then? Seems like its broken then on some deep level, given the fact these technological improvements related to DX12 are sort of mail selling point of the game. After playing the game i can tell the gameplay/game design certainly is not (for now).

Its broken in the terms that there are still big issues. The game is beta1 stage after all. But its not going to release the fabled performance when fixed.

 

[TheELF]

...thus making the GPU the bottleneck, right? Any point to DX12 then, when it comes to PC gaming? Do we expect upcoming next-gen GPUs become so powerful, that they will be able to utilize multiple cores of upcoming (Intel) CPUs?

Yes GPU is the bottleneck.
Dx12 will be great if they start using it correctly,games will stop to have so many frame drops/stutters from texture streaming or from moving your view from a low count draw call point to one with a high count,it will make games smoother but no increased performance other then with very weak cores and a strong GPU.

Look at this video it explains a lot of what dx12 should do.
https://www.youtube.com/watch?feature=player_detailpage&v=H1L4iLIU9xU#t=941

 

[JDG1980]

Because DX11 is the bottleneck and AMD decided a long time ago rather than try and overcome the archaic API they'd push past it with Mantle. Now we have DX12, Vulkan, etc... Hopefully DX11 dies sooner rather than later. It's been around way too many years now.

That won't happen until 2020 (when Windows 7 goes out of support).

 

[JDG1980]

I still dont understand how are Nvidia cards doing more or less as good under DX11 as they are doing under DX12/as AMD cards doing under DX12.

Was the DX12 not supposed to unlock all the CPU cores utility, thus increasing the performance several times in comparison to DX11, where only one CPU core is usually taxed? But now i am seeing basically the same frame-rates... can anyone elaborate please?

Nvidia has multi-threaded DX11 drivers. AMD doesn't. This explains a lot of the performance differences in DX11, and why AMD does a lot of catching up in DX12.

 

[Timmah!]

Yes GPU is the bottleneck.
Dx12 will be great if they start using it correctly,games will stop to have so many frame drops/stutters from texture streaming or from moving your view from a low count draw call point to one with a high count,it will make games smoother but no increased performance other then with very weak cores and a strong GPU.

Look at this video it explains a lot of what dx12 should do.
https://www.youtube.com/watch?feature=player_detailpage&v=H1L4iLIU9xU#t=941

Thank you!

 

[Silverforce11]

When are they planning to release?

I look forward to the multi-GPU adapter support they are bringing soon. Really showcases DX12's flexibility allow devs to do crazy stuff like pair AMD + NV GPU!

Also for the bench, the interesting analysis is the 970 vs 390.

In DX11, the 390 is close but behind, but it gains some in DX12. Fury X looks like it is really gimped for DX11.

The reason?

Same front-end, 8 ACE engine, 1 CP hardware scheduler is feeding ~2.8K shaders in Hawaii is asked to feed 4K shaders in Fiji. It simply cannot do it in DX11 with serial ops and single threaded rendering. But it shines in DX12 because those ACEs are now scheduling tasks for the shaders to work.

 

[KaRLiToS]

I think you had too much DX12 koolaid

Their implementation is still broken. But you will never get what you expect.

I would prefer Dx12 koolaid than the green koolaid.

 

[Madpacket]

Warps vs. Wavefronts, different approaches really. 980Ti ties Fiji out of sheer brute force. Given DX12 is essentially Mantle++, none of this should be surprising to anyone. Rest of Nvidia's Maxwell line will likely fall behind AMD's Tahiti through Polaris as more DX12 games are released. Nvidia really needs Pascal to compete in DX12 as AMD set the standards. Should be an interesting year but it'll take more than Ashes of the Singularity to really get people to notice how far behind Maxwell Geforce cards really are. Is DICE or Crytek planning on releasing a DX12 game this year?

 

[Silverforce11]

Warps vs. Wavefronts, different approaches really. 980Ti ties Fiji out of sheer brute force. Given DX12 is essentially Mantle++, none of this should be surprising to anyone. Rest of Nvidia's Maxwell line will likely fall behind AMD's Tahiti through Polaris as more DX12 games are released. Nvidia really needs Pascal to compete in DX12 as AMD set the standards. Should be an interesting year but it'll take more than Ashes of the Singularity to really get people to notice how far behind Maxwell based Geforce cards really are. Is DICE or Crytek planning on releasing a DX12 game this year?

Fable DX12 also shows NV's entire stack below the 980Ti is non-competitive (R290 reference >= 980!), and yes the 980Ti is brute forcing it with its superior front-end, lots of ROPs.

Fiji is a gimped design that needs DX12 to shine more-so than the other GCN chips due to recycling the same front end made for ~2.8K shaders.

 

[guskline]

I ran Ashes of Singiularity benchmark Extreme preset at 2560x1440 on both of my machines below with the following results.

On my 5960x @4.4Ghz and a single EVGA GTX980TI SC (vcore 1102 vs 1000 on test model)
I had overall average frame rate of 54.8 for DX11 and 54.6 for DX12. This is almost identical so it's hard to see what affect, if any the difference in DX11 vs DX12 makes.

On my 4790k rig @4.7Ghz and 2 R9 290s (Sapphire OC Tri-Xs at 1000 core)

I had overall frame rate of 36.4 for DX11 and 39.4 for DX12. Obviously Ashes of Singularity doesn't yet support CrossFire since the scores for 2 R9 290s should be much higher. I noticed the R9-390 is higher but remembered that even if vcore is nearly the same for 290 to 390, the memory is now 6 GHz for 390 vs 5.2Ghz for the Sapphire R9 290OC. I suspect that makes a difference.

In the AMD 290 I had a 3 point jump in fps going from DX11 to DX12.

 

[selni]

Nvidia has multi-threaded DX11 drivers. AMD doesn't. This explains a lot of the performance differences in DX11, and why AMD does a lot of catching up in DX12.

Yes, but it doesn't explain why nvidia's dx11 results are still better than 12. They're either not rendering the same thing in the same way (likely, driver shader replacement etc) or the dx12 implementation in this particular game is less optimal for nvidia's hardware than what their dx11 driver is doing automatically.

Or more likely some combination of both - lower level APIs aren't magic.

 

[Silverforce11]

Yes, but it doesn't explain why nvidia's dx11 results are still better than 12. They're either not rendering the same thing in the same way (likely, driver shader replacement etc) or the dx12 implementation in this particular game is less optimal for nvidia's hardware than what their dx11 driver is doing automatically.

Or more likely some combination of both - lower level APIs aren't magic.

Looks like Oxide enabled Async Compute which they disabled back last year at NV's request (as per Dan Baker's statements on OCnet).

 

[selni]

Looks like Oxide enabled Async Compute which they disabled back last year at NV's request (as per Dan Baker's statements on OCnet).

Doesn't seem important in the nvidia DX11 vs 12 comparison. The version last year had it off (as you mention at nvidia's request) and still showed the same thing.

 

[Silverforce11]

Doesn't seem important in the nvidia DX11 vs 12 comparison. The version last year had it off (as you mention at nvidia's request) and still showed the same thing.

That was the initial build, it showed the same perf loss in DX12, then Oxide released another build later and AMD GPUs perform worse, such as Fury X losing to the 980Ti which no longer show perf loss in DX12. This was after Dan Baker posted mentioning they disabled Async Compute at the request of NV.

http://www.computerbase.de/2015-10/...ashes-of-the-singularity-directx-12-1920-1080

Now this beta build comes out, and we see the similar thing, where NV regresses in performance in DX12, I would assume Async Compute is now enabled again.

Note this is similar to what Crystal Dynamics said about Tomb Raider. They have Async Compute running in XBone & DX12, but its disable for DX11 PC release because they saw no performance gains and even some perf loss. I can only guess they tested with NV GPUs because NV sponsored that game.

 

[guskline]

SilverForce11, why is my DX11 vs DX12 score for my GTX980TI almost identical ? My score jumped 3 PTs from DX11 to DX12 for my R9 290. It was 54.6 vs 54.8 for DX11 to DX12 on the GTX980TI. Perhaps Madpacket's explanation of the sheer brute force of the GTX980TI making up for the change from DX11 to DX12 has some validity.

 

[Silverforce11]

SilverForce11, why is my DX11 vs DX12 score for my GTX980TI almost identical ? My score jumped 3 PTs from DX11 to DX12 for my R9 290. It was 54.6 vs 54.8 for DX11 to DX12 on the GTX980TI. Perhaps Madpacket's explanation of the sheer brute force of the GTX980TI making up for the change from DX11 to DX12 has some validity.

Ars article on Ashes show the 980Ti perform difference on 4 vs 8 core CPUs back then, but I was referring to the Ashes build where NV cards had a performance degradation in DX12, vs the 2nd build when they did not. Now its back to it was like in the first alpha results (presumably with AC enabled, based on what Dan Baker had said at the time).

 

[guskline]

Sorry SilverForce11, I'm not following your reply?

 

[lilltesaito]

I ran Ashes of Singiularity benchmark Extreme preset at 2560x1440 on both of my machines below with the following results.

On my 5960x @4.4Ghz and a single EVGA GTX980TI SC (vcore 1102 vs 1000 on test model)
I had overall average frame rate of 54.8 for DX11 and 54.6 for DX12. This is almost identical so it's hard to see what affect, if any the difference in DX11 vs DX12 makes.

On my 4790kn rig @4.7Ghz and 2 R9 290s (Sapphire OC Tri-Xs at 1000 core)

I had overall frame rate of 36.4 for DX11 and 39.4 for DX12. Obviously Ashes of Singularity doesn't yet support CrossFire since the scores for 2 R9 290s should be much higher. I noticed the R9-390 is higher but remembered that even if vcore is nearly the same for 290 to 390, the memory is now 6 GHz for 390 vs 5.2Ghz for the Sapphire R9 290OC. I suspect that makes a difference.

In the AMD 290 I had a 3 point jump in fps going from DX11 to DX12.

Could it be different drivers being used?

 

[Silverforce11]

Sorry SilverForce11, I'm not following your reply?

There's some differences in the CPU that may cause your and Computerbase.de results to not match? Just a guess.

 

[guskline]

Sure, Computerbase.de uses a 4770k @4.4Ghz while I use a 5960x @ 4.4Ghz and a 4790k @4.7Ghz.

 

[dogen1]

SilverForce11, why is my DX11 vs DX12 score for my GTX980TI almost identical ? My score jumped 3 PTs from DX11 to DX12 for my R9 290. It was 54.6 vs 54.8 for DX11 to DX12 on the GTX980TI. Perhaps Madpacket's explanation of the sheer brute force of the GTX980TI making up for the change from DX11 to DX12 has some validity.

Your 980Ti performs ~ the same in DX11 and 12 simply because nvidia's DX11 drivers are very fast. The biggest benefit from D3D12 comes from the reduction in driver overhead. If no performance increase is seen, then you can only assume that nvidia's driver overhead is not enough to bottleneck performance.

 

[3DVagabond]

What died is AMD's marketshare. Maybe they should have tried to optimize better for DX11. After all, there are still no final DX12 games on the market.

Market share has nothing to do with the subject here. Maybe you could start a thread about how AMD should have embraced DX11 more. Then we could discuss why that would have been worse for you and me. Because why would anyone care if it cost AMD market share.

DX12 is a better, more powerful, higher performing API. I'm thankful they didn't take the route of just following the leader and instead advanced the state of the art.

 

[3DVagabond]

Disregarding consoles, who buys nowadays low end AMD CPU with high end GPU? People who have money for 980Ti/Fury X, are very likely to have the means to get i7, which, if i understand correctly, is fast enough to feed the GPU the data just using single of its 4/6/8 cores...thus making the GPU the bottleneck, right? Any point to DX12 then, when it comes to PC gaming? Do we expect upcoming next-gen GPUs become so powerful, that they will be able to utilize multiple cores of upcoming (Intel) CPUs?

Actually with DX11 CPU's are about at their limit (Not every game, of course). They will be able to feed more to the GPU with these new API's. It's just early days yet.

 

[3DVagabond]

Nvidia has multi-threaded DX11 drivers. AMD doesn't. This explains a lot of the performance differences in DX11, and why AMD does a lot of catching up in DX12.

Oxide said there is no AMD optimized code but there is nVidia optimized code. I doubt AMD has done anything with the drivers to optimize it for DX11.

I'm always hearing how bad AMD DX11 drivers are but if you look at comparative benchmarks in reviews, which are all DX11, AMD does quite well. The 980ti and Titan X are the only cards that outperform AMD. And then it's not really by much most times.

 

[tential]

Oxide said there is no AMD optimized code but there is nVidia optimized code. I doubt AMD has done anything with the drivers to optimize it for DX11.

I'm always hearing how bad AMD DX11 drivers are but if you look at comparative benchmarks in reviews, which are all DX11, AMD does quite well. The 980ti and Titan X are the only cards that outperform AMD. And then it's not really by much most times.

Because when you scale it to real world and don't use a top of the line processor it tells a different story.

AMD has had the price/perf gpu crown, but then hampers that by needing a faster CPU than the nvidia equivalent to use it due to poor DX11 drivers.

Not that this matters to me personally, I think people should get an i7 as their intel processor will last along time. But, that's the argument many people present about AMD and DX11.