DX12 case studies (NVidia GDC 2017)

[IllogicalGlory]

Yeah DX12 is definitely going to have some teething. The creators update added WDDM 2.1 so could be related to that or drivers needing to be updated.

I'm sad how little we get in ways of updated testing from the review sites. Why don't we have any "June 2017 DX11 vs DX12 Revisited/Showdown" type articles?

Ideal:

3 GPUs each: 1060, 1070, 1080, 470, 580, Fury (X)

5 CPUs each: Pentium, i5, R5, i7, R7

3 Resolutions: 1080p, 1440p, 4k

2 Settings: Medium, Very High

Games: Hitman, Deus Ex MD, Doom, Sniper Elite, The Division, Rise of the Tomb Raider, Battlefield 1 (any others I missed?)

Someone please!!

6 GPUs * 5 CPUs * 3 Resolutions * 2 settings * 7 games = 1260 tests

There's a reason we don't see thorough testing like that.

 

[Bacon1]

6 GPUs * 5 CPUs * 3 Resolutions * 2 settings * 7 games = 1260 tests

There's a reason we don't see thorough testing like that.

Hey I did say Ideal

 

[BFG10K]

Cause it's not that amazing to outperform a compiler. It's a program vs a human being.

Nonsense. If you think so you, should have no trouble providing examples to prove it.

Show us examples of any significant-sized program (e.g AAA game, Linux/BSD Kernel, GPU device driver, etc) rewritten entirely in assembler that performs better than the compiled version. On multiple CPU types.

 

[dogen1]

Show us examples of any significant-sized program (e.g AAA game, Linux/BSD Kernel, GPU device driver, etc) rewritten entirely in assembler that performs better than the compiled version. On multiple CPU types.

I wasn't really talking about writing a whole program that way. (If that's what you meant, then sure that's a different, probably with few examples) It's actually not always that difficult to outperform a compiler in small cases though, which is about the only way that asm is used nowadays.

 

[antihelten]

6 GPUs * 5 CPUs * 3 Resolutions * 2 settings * 7 games = 1260 tests

There's a reason we don't see thorough testing like that.

Your average TPU review contains 22 games * 3 resolutions * 15 GPUs = 990 tests. Computerbase.de reviews contains 23 games * 3 resolutions * 17 GPUs = 1173 tests

Of course they reuse results from previous reviews, so the tests are not all done at once (TPU basically only perform 22 games * 3 resolutions * 1 GPU = 66 tests per review, and Computerbase.de does 23 games * 3 resolutions * 1 GPU = 69 tests per review).

 

[frozentundra123456]

I wasn't looking for any "biased articles" at all when I found this one. I just thought I would post it since it has some interesting info regarding DX12 optimization. Also Hitman isn't exactly a gameworks title now is it? And the AnvilNext DX12 engine from Ubisoft hasn't been released yet, but it likely will with the new Assassin's Creed game.



Yeah I can't believe that myth still persists. There is a section in Doom where I got a 150% increase in performance vs OpenGL, so yeah, low level APIs are useful

Quite an ironic request coming from that particular poster, is it not?

Trolling and Threadcrapping are not allowed
Markfw
Anandtech Moderator

 

[beginner99]

It also confirms some suspicions myself and others have had in regards to DX12 performance on NVidia; that matching the DX11 driver is very difficult to do manually through DX12 when it comes to certain aspects, especially memory management.

It's only a problem because the dx12 games/engine we have seen so far all just tack-on dx12 on the old probably dx9 based engine. A pig with lipstick is still a pig. If you really want to go dx12 and make use of it's features you need to drop backwards compatibility. So yes it's for sure hard to tack-on dx12 features and not be slower than dx11. We see that also with BF1 where dx12 also sucks on AMD for example.

Of course if you ship your GPU with crippled memory config like GTX 970 and lie about it then yes it^s hard for devs to optimize for it (eg. only use 3.5 of the advertised 4 GB)

 

[TheELF]

It's only a problem because the dx12 games/engine we have seen so far all just tack-on dx12 on the old probably dx9 based engine.

Isn't that how APIs work anyway?
Back in ye olden days you would get games with different executables depending on your GPU hardware, then they invented directx/3d so you could make engines that are hardware independent.
Every API is just tacked on just as any gpu support (driver) is just tacked on,otherwise we wouldn't be able to play any console ports of the past many years on windows.

 

[SirDinadan]

Isn't that how APIs work anyway?

The abstraction level what really matters in the context of DX11 and DX12. The devs can access a lot of functionality "closer to the metal" in the new low-level, explicit APIs. With that comes more work, greater flexibility and greater responsibility.
Because of the inherently different nature of the APIs, we don't see many titles that really gain with DX12 & Vulkan - Ashes the solo superstar as its engine was a clean sheet design. DOOM's gain on AMD is twofold - AMDs OpenGL driver is that bad / inefficient / doesn't fit GCN and somehow AMD managed to get some of their low level optimizations into the engine like Shader Intrinsics, etc.

Don't think we will see many DX12 exclusive games this year, maybe there won't be any as the time to develop a new engine for AAA games are enormous.