DX12 on fermi

you should be able to run driver 365.19 and Windows 10 insider build 14332 the dx12 is enabled. you might even be able to use the most current insider build with 365.19 or maybe even current drivers and get dx12 on fermi.

 

i was looking and the only open source drivers for Nvidia i can find are for Linux and OpenGL

 

Already tried this with SLI enabled and disabled and it won't run the DX12 Futuremark API overhead test nor the Unreal 4 DX12 elemental tech demo, using builds 14342, 14352, 14366, 14367.

 

Me too, I tried with 365.19 on Insider, no DX12. It appears DirectX 12 on main tab of Dxdiag, but on the Direct3D DDI stays 11.3. On the 358.70, Direct3D shows 12, and I was able to run the API Overhead test. The Unreal DX12 always crash, but a least don't say to remove -d3d12 from options, like on the 365.19.

 

that is really sad to hear.

 

Except DirectX implementation are not open source, but shared source for hardware partners of the graphics advisor boards only.

 

We will see if something will change after Redstone release. In any case we will receive driver with WDDM 2.1 and Directx 11.4 (driver 368.20 enabled it in insider build).

But I still have little hope that we will receive wddm 2.1 directx 12 driver for fermi after redstone release 2th August.

UPDATE.

I found something about wddm 2.1 directx 12 driver on this page and driver 368.20:
http://blog.naver.com/sqauill
It is in the Korean language I think but follow this two links above:
I transalted this site to my mother languge with google and here is what I have for you:

1.
http://blog.naver.com/OpenMagazineV...88&categoryNo=12&parentCategoryNo=0&viewDate=

You can see on the photos that he was able to have wddm 2.1 and dx12 ddi 11.4 with QUADRO 4000m which is GF104 fermi gpu.

2. Here you have difference between normal driver and modded

http://blog.naver.com/OpenMagazineV...88&categoryNo=22&parentCategoryNo=0&viewDate=

3. Here you have a link for modded 368.20 driver:

OPTIMUS BASED SYSTEM MOD
https://mega.nz/#!LJxzDS5D!N9LIKQmaLYGecTLedjVk0CzWzZAmjgBKpRj2O7GrQ9Y

MS HYBRID BASED SYSTEM MOD
https://mega.nz/#!SFpi1JLC!YhI-yFWrltpONsbH9aY7mmPRM6Q2BfguIN4GESakLDs

I checked those files with KIS 2016 and modded driver is clean.

Password for the 7z archive is the site adress: http://blog.naver.com/sqauill

4. He had windows 10 insider preview (redstone)

5. I think this modded driver is for mobile gpu only, but you can add gpu string if are not presented to the inf like always and try.
Also if someone have mobile fermi and insider preview he can try with this driver.

I don't have possibility now to instal latest insider windows 10 but someone can give a try with this modded driver.

Remember to enablein cmd with admin privilages:
bcdedit -set loadoptions DISABLE_INTEGRITY_CHECKS
bcdedit -set TESTSIGNING ON

as it is modded driver

to go back do the same, but put in cmd:
bcdedit -set loadoptions ENABLE_INTEGRITY_CHECKS
bcdedit -set TESTSIGNING OFF

 

What is the password ?

 

Hi, WDDM 2.1 should be related to HDR (BT2020) support (not sure why a different version number, maybe it's related to windowed mode). It should come with the anniversary update (1607?). First wave of tool for Shader Model 6.0 should be also released soon.

Direct3D 11.4 was released with the November update of Windows 10 (1511), DDI should be the related display device interface, nothing more (it just add a method for device removal on DirectX 11).

All this has nothing to do with Fermi.

 

I know, you know but i haven't tried that driver. I just need a password to it.

 

Neat conspiracy theory...of course it ignores the main advantage of DX12 and vulcan promoted as being reduced CPU overhead. So, how about some sort of data where DX12 makes Fermi GPU's perform much better.

 

thanks to your post, i just found out 3dmark had this test.

not quite sure i get the results though?

dx11 single thread is 2,865,054, dx11 multi is 2,373,902, and dx12 is 17,898,073. ... ... wtf? an increase of 15 million?

http://www.3dmark.com/aot/147166

 

Yes, it's about what I would expect from a 970. Look at a 1080 results for example, dx12 just above 25 million.

http://www.3dmark.com/aot/147101

 

The password is the site adress: http://blog.naver.com/sqauill

 

Anyone tried the driver in w10 rs1 preiview?

 

Maybe at the weekend. I cannot make any promise.

 

siriq, I tried many times, can't reach on DX11 more than ~1.100.000 with Phenom II. Look at this result with a 1090T:

http://www.3dmark.com/3dm/9211957

I don't know exactly why this happens. Single thread performance isn't similar between FX and Phenom? Or maybe could this results be related to HT Link/NB improvements?

 

Your results are strange and quite low.

On my i7 2670QM and gf 108 Geforce GT630M ( GPU Clock - 890MHz, Memory 1200MHz), with ssd and 16 GB RAM I have reached:

DirectX 11 single-thread 1 203 128

DirectX 11 multi-thread 1 651 527

 

Yes, I know, but I'm beginning to think that this is really a limitation from the Phenom platform. I tried tests with two motherboards that I have, the M2N-SLI, that features a Nforce 560 chipset, and the ECS 8200, that have a nForce 720a chipset.

The M2N has worse performance on the API Overhead tests, around 600 000 to 700 000 on DX11 results, but same results on DX12, around 3 400 000 draw calls. I think that this worse results can be explained by the limited HT frequency on the nForce 560 chipset, that has a max of 1000MHz (HT 5x) and also the Northbridge, that is locked at 1600MHz, can't overclock it.

On the ECS 8200, I have this results that I said, around 1 100 000, 1 200 000 on DX11, and the same 3 400 000 on DX12. The HT clock on this board I can overclock up to 2600MHz, and Northbridge up to 2400MHz, I can't go higher on northbridge because I can't increase the voltage, the VRM has overvoltage shutdown feature.

In both tests, I used the same CPU, my 940BE, at 3600MHz and same memories, DDR2 800, unganged, DC. I also tried with my two 560, ans many drivers, including 314.22 (dark souls 3 fermi driver), 358.70 (beta DX12), 365.19 and the last 368.39. I also used the latest 4 builds of the Windows Insider 10.

If anyone here has Phenom platforms, please if its possible, submit API Overhead test results, maybe can help-me to understand this results.

 

later, will try