AMD Catalyst 15.x (15.200.1040.0 June 8) Win 7/8.1/10 x64

The big difference is using Windows 10 drivers in general, Windows 10 + Windows 10 driver isn't much different if any. The overhead improvements didn't come from WDDM 2.0

 

Seems that they might have come from WDDM 1.3 though. I'm not 100% sure yet.

Once Windows 10 releases, I'll do as many benchmarks as I can with the latest "Windows 10" driver on Windows 8.1, and then repeat on Windows 10.

 

Windows 10 build 10147
Display driver: 9.14.10.01128

AMD Radeon R9 290X(2x) @ 2560x1440 and Intel Core i7-3930K Processor

DX11 Multi-threaded draw calls per second
1 529 321

DX11 Single-threaded draw calls per second
1 296 273

DX12 draw calls per second
19 508 669

Mantle draw calls per second
25 555 848

Link to my test:

http://www.3dmark.com/3dm/7532332

 

It's the first time that the MT test is higher than the ST one. And Mantle also seems to be much faster than DX12... Hm...

 

interesting, never knew this. How exactly do you do this, and can you play mkv files?

also, will updating my system to windows 10 require me to redownload all my games? i have over 700gb i would have to download.

 

Mantle has actually been pretty much always fasters with fast 6 core cpu's. It has been slower only on 4c/8t and lesser cpu's. That's what I recall anyways.
Now difference is very huge actually. (Like you said)

MT looks good, finally there are some multithreading happening.

 

For the videos just download Smooth Video Project (SVP) which comes with both mediaplayer classic and madvr. SVP does the frame interpolation for higher fps and madvr does the downsampling. I wouldn't recommend modded windows 10 drivers for this as OpenCL and DVXA are unreliable even with the fixes provided.

To keep your games just repartition your hard drive with something like MagicTool Partition Wizard. This way you can do a clean install of windows while keeping all game directories intact.

 

Shows up OS as 8.1 and the MT test is interesting, can you run it without crossfire to see if they show up with single card as well.
Still they are an improvement from these tests,

http://www.legitreviews.com/looking...mance-3dmark-api-overhead-feature-test_160936

 

It still detects as windows 8.1 but the build matches 6.3.10147 & windows 8.1 will detect as 6.3.9600. I will run a single GPU test im a few hrs.

 

Good result!

 

AMD Radeon R9 290X(1x) @ 2560x1440 & Intel Core i7-3930K Processor

Disabled crossfire without reboot, just restarted AB
DX11 Multi-threaded draw calls per second
638 945

DX11 Single-threaded draw calls per second
689 454

DX12 draw calls per second
18 500 515

Mantle draw calls per second
16 772 122

http://www.3dmark.com/3dm/7548169

AMD Radeon R9 290X(1x) @ 2560x1440 & Intel Core i7-3930K Processor

With a reboot after disabling crossfire
DX11 Multi-threaded draw calls per second
942 854

DX11 Single-threaded draw calls per second
1 005 666

DX12 draw calls per second
19 668 119

Mantle draw calls per second
21 624 854

http://www.3dmark.com/3dm/7548254

 

So there are MT improvements with multigpu?

 

what dxdiag shows in 10158 for 290x/285 ?

Ed: still shows 11.1 as max level.

 

I notice a bit of texture flickering (random spots) in Witcher 3 with these 1040 drivers. Other than that, all is good, though i can't really say it is any better/faster than the 15.5betas. Running a single 290 here.

 

i got it to work in the end as it would not overwrite 1 of the files so had to delete that 1st

thank you

 

I guess they use one submission thread per GPU.

With MT and one GPU: The multiple game thread draw calls get bundled to a single submission thread.

With MT and two GPUs: The multiple game thread draw calls get divided between two submission threads.

I'm still pretty sure the trick from NVidia is something simple like this, the question is why AMD can't do the same?

Edit: Maybe the problem is the order, and NVidia is already using Rasterizer Ordered Views internally? How does the Draw Call performance compare between Maxwell 2.0 and the older versions?

 

DX11 has been out for awhile, I'm sure if they were able to do it like that they would have. DX12 is more suited to AMD GPU hardware it seems. Like they predicted it

 

Sounds the same to me. I have a feeling that what AMD is doing for multi GPU, NVIDIA is doing for each GPU cluster in a single GPU. That could explain a lot of things.

 

http://www.3dmark.com/fs/5272315

My 3dmark score with 290x @ 1100/1400... on 1040 and w10 10130. So 200 score behind 390x with 1100/1500

 

its more like 400 score difference