No DX12 Software is Suitable for Benchmarking *spawn*

The current state of DirectX 12 in PC gaming, was it worth the hype?

With the ridiculously high multi-GPU scaling I've had in RoTR, Deux Ex MD and AotS, compared to the DX11 implementation, my anecdotal experience says yes, it's definitely worth the hype.
 
Unfortunately when DX11 is ditched DX12 performance reviews will have little meaning with nothing to compare against.
 
Yeah give it another year and we'll see. I'm also interested in seeing how it affects current hardware in the long run. Mantle needed special care as architectures changed or performance dived.
 
Last edited:
How silly. By that time we'll be able to compare DX12 with DX13. :devilish:
That's true ... in about 5 years or so. :D Meaning developers will always have to cross pollinate games to run with different versions of DirectX?
 
Last edited by a moderator:
How silly. By that time we'll be able to compare DX12 with DX13. :devilish:
I have a feeling this could happen sooner rather than later, a more mature version of DX12 could be released (DX12c?, DX12.3?) or even a DX13. Which will become more relevant and affect our games experience more than what we have today.
 
Will SM6 with DX 12 create more of a distinction vs DX11 with SM5?

Is anything in SM6 reliant on features added by DX12?

If SM6 is a good exponent of the potential extra performance allowed by DX12 why has it taken MS so long to have it ready?
 
Direct3D 12 covers a larger hardware support than Shader Model 6.0 does (at least if we consider only FL12 GPUs where SM6.0 support will be mandatory). I think this was one of the reasons of SM6.0 delay. There are also many other cool features that are not supported in SM6.0 since they are restricted to few and specific architectures (barycentric, depth bounds test, some nvidia indirect stuff extensions).
 
Just tried out the recently released final DX12 Renderer for The Division on my 1070, I witnessed some palpable increase in fps under DX12.

Gameplay, walkaround Brooklyn (near Plymouth Rock safehouse):
DX11: ~ 55- 68 fps
DX12: ~ 65-72 fps

Internal Benchmark:
DX11: 61 fps
DX12: 74 fps

Edit: I used max settings @1080p, with HFTS shadows, my theory is that running HFTS under DX12 produced a large fps gain than under DX11, this could be the main reason why we see this increase under DX12.
 
Last edited:
Scratch that, the gains are already present even at Night (where HFTS is replaced with PCSS), there is generally a 10fps difference between DX12 and DX11 at Max settings.
 
Is this the case? I thought they were supported.
Performance: The cost of Contact Shadows varies from location to location and with the width and depth of the scene you're viewing, and whether you're using NVIDIA HFTS or NVIDIA PCSS. Using High Shadow Quality we observe a 0.6 FPS cost for enabling "Sun Low", a 2.9 FPS cost for enabling "All Low", and a 6.5 FPS cost for enabling "All High". Switching to NVIDIA HFTS or NVIDIA PCSS, however, sees the cost per level sit around 2 FPS, for a total of 6 FPS when enabling "All High".
http://www.geforce.com/whats-new/gu...de#tom-clancys-the-division-ambient-occlusion
 
Ran some tests on my rig (I7-3770k @4.1GHz and Fury X @ 1100/545MHz

"1080p" in windowed (game crashes in full screen)

I also was getting 15fps boost , but it turned out to be because I had PCSS and HBAO+ on. After running AO at Ultra and Shadows at High:

DX11

Avg FPS: 67,8 fps
Typical FPS: 68,2 fps
Avg CPU: 34%
Avg GPU: 99%

DX12:

Avg FPS: 73,8 fps
Typical: FPS 74,2 fps
Avg CPU: 36%
Avg GPU: 99%

So getting roughly 6 fps upgrade in benchmark.
 
Thx for the bench. What about VRAM usage ? I notice that in Deus EX MD, DX12 use more VRAM (which is "bad" cause the game already eat vram a lot in dx11 ...)
 
Thx for the bench. What about VRAM usage ? I notice that in Deus EX MD, DX12 use more VRAM (which is "bad" cause the game already eat vram a lot in dx11 ...)

DX11

Benchmark
Decicated peak: 3447MB
Dynamic peak: 69MB

Game
Dedicated peak 3747MB
Dynamic peak 72MB

DX12

Benchmark
Dedicated peak 3672MB
Dynamic peak 428MB

Game
Dedicated peak 3693MB
Dynamic peak 695MB

Overall pretty same, I didn't run exacltly same places in game, but overall I can't say either renderer using more memory from these short tests. It looks like DX11 starts with less memory, but catches up very fast.
 
Back
Top