Quick test: DirectX 12 API Overhead Benchmark

Oh I thought it replaced DX11 completelly when game changed to that mode.

 

think this is the first time that I am really excited about the next version of Windows. lets hope devs will be ready for this!

 

Double post

 

Pretty good results now that I've tested mine.

i get the following:

DX11 Single: 1.1M
DX11 Multi: 1.08M
Mantle: 19.2M

I guess 6 cores does make a huge difference but all in all mine isn't that bad.

No mantle has nothing to do with DX11 but in the game you can run either mantle or dx11.

 

Windows update detect my asus 280x direct cu like an amd 7970 and can't run the dx 12 test, any ideas to fix it?

 

Its not a gpu performance test.. so comparing gpu's performance based on it is really not a good idea.

Result will vary a lot, a 960 a little bit overclocked end 10% faster of a 980 lol.

I dont know if theres bug or just bad metrics.

 

Just tested this out, here's my results

http://www.3dmark.com/3dm/6395560

DX11 Single-threaded draw calls per second 907373
DX11 Multi-threaded draw calls per second 2727155

 

Windows 10 is looking promising. Hopefully games fully support DX12 quickly.

 

Yeah, I saw that afterwards.

I didn't see the one with a 960 being faster than a 980 though, did see an overclocked 960 match a 980.

 

My CPU is also overclocked to 4.7Ghz, my RAM is running at 2388, and my 7970s are running at 1050/1575.

Oh, and single card vs dual card, mantle score is almost double, while DX11 only goes up by about 17%.

edit: by seeing that, I think CF and SLI setups are going to see a lot larger gain in performance vs single card than they currently do.

 

This is not a benchmark. It's a comparison test of increased drawcall capabilities of dx12 over dx11. They made it very clear this test was designed to remove hardware performance from the equation and simply compare apis.

https://www.youtube.com/watch?v=O_6PxugCBa0

 

At the same time, some things can be derived from the results of this test.

 

It will need to be confirmed in practice, and im not sure what can be derived today will be so evident then.

And as "multi threading DX11", dont misunderstood "driver command list" and DX11 multi-threading support. ( Funny thing it seems that in TitanX, it is not enabled ).. This " feature " is not used so much for games, it reduce the performance in most cases. ( only AssassinCreed and CIV5 " benchmark" was show it )...

 

My results:
DX11 Multi-threaded draw calls per second 692 527
DX11 Single-threaded draw calls per second 691 889
Mantle draw calls per second 13 630 279

 

I wonder why their single threaded DX11 for GTX 980 is so slow... I get over 2 million. Also their multi threaded result is kinda low too.

Now you have to give source for that. To me that doesn't make any sense nor should it even be possible and even if it would be it would likely just be counter productive.

 

I think he mix 2 things.. BF4 have not been coded with Mantle in mind, it have serve as a test / base support to Mantle API test / developpement .

BF4 will not show massive increase in performance ( even if it is there ), nor drawcalls, as the DX11 and the Mantle version are nearly identical in the number of drawcall..

The game have not been developped for Mantle "only"... using what can bring Mantle potentially for increase " performance " or redraw the quality level.

But understand that Mantle developpement have start way before on the dice team than BF4 was on developpement. ( DICE leader graphics, developpement teams have been a real pillar for the developpement, and evolution of Mantle, as they have not only use it, but it was an essential work for use / test / developp the API in a real practical game / engine ) they have been fully involved in the developpement of Mantle, nearly from the start. Its maybe the reason, that have give to AMD the possibility to bring this complete API in a real short time ( maybe only 4years of developpement, who is extremely short for something like that, without release multiple version over the time ).

 

Can anyone confirm with me that the file size for this update is a whopping 1.6GB, is that accurate or am I trying to download the wrong file? Thanks guys in advance for helping me out with this!

 

Yep it is. They are always around this size aren't they? It included shed loads of textures needed for the tests, that's why AFAIK.

 

Thank you very much for taking the time to let me know that geogan, its greatly appreciated! I'm in the process of DL'ing now!