No DX12 Software is Suitable for Benchmarking *spawn*

HOCP confirms what pcgameshardware, sweclockers, comuterbase, gamersnexus, and techspot found about BF1 DX12 mode:

When switched to DX12, BF1 falls flat on its face. There is no performance improvement, and there are no visual quality improvements. In fact, there is a negative return on framerate. On both the AMD Radeon RX 480 and NVIDIA GeForce GTX 1060 the framerates go south and provide worse performance while gaming, at 1080p and 1440p.

It was clear though that the GeForce GTX 1060 took a greater hit in performance running DX12 versus DX11. The AMD Radeon RX 480 took a lesser hit in performance, though it was still slower in DX12.

Note that this is all with the latest drivers as well. We are using the AMD Crimson Edition 16.10.2 Hotfix driver and the NVIDIA GeForce 375.63 driver, both newly released for this game.

DX12 introduces a stutter in BF1 on both video cards. We found it incredibly noticeable during mission load, and multiplayer map load. For the first 15 seconds of loading a new map or multiplayer map there would be random noticeable stutter. Then there would be random stutter as you played, especially more noticeable in multiplayer maps. DX11 was butter smooth with no issues on either video card in multiplayer.

Therefore, the answer to our question right now is simple; run BF1 in DX11. There is no reason to run BF1 in DX12, unless you just like to punish yourself. Both the Radeon RX 480 and GeForce GTX 1060 will be better off in DX11 in multiple ways.
http://www.hardocp.com/article/2016..._card_dx12_performance_preview/6#.WBBW_iTTzIU
 
Multi-GPU on DX12 is now final on Deux Ex: Mankind Divided for AMD cards, and the developer says they're working with nvidia to make it final for their cards too, within the next few days.

I got a small boost with my setup&setttings on average results, up to 52 FPS but minimum FPS went from 16 to 25. There's practically no perceived stuttering during the benchmark. The in-game benchmark is definitely more punishing than any in-game zone. I know because I just finished the game and most of the time I was at the framepacing's 74FPS limit lol.

Now that the implementation is final, does anyone have any hints on what software I should use to measure frametimes?
I could do that with DX11 single vs DX12 single vs. DX11 dual vs. DX12 dual.
 
Battlefield 1 Performance Analysis

All of the graphics cards included in this test were run with the latest drivers installed. In Nvidia’s case that would be Driver Version 375.63 and on the AMD side, we are using Radeon Software 16.10.3. Both of these drivers were released with optimisations for Battlefield 1 and Titanfall 2. Our results were obtained using FRAPS to record average and minimum frame rates. We chose the first level of the campaign to benchmark. The game is running using the Ultra preset and the DirectX 12 API.

http://www.kitguru.net/components/graphic-cards/matthew-wilson/battlefield-1-performance-analysis/
 
Surprisingly, The Division now supports a DX12 renderer, someone tested it using a FX 8350 @4.4Ghz and a GTX 1070, he gained 20fps using the "internal benchmark" @1080p and mostly maximum settings! I could test this but I need sometime to download that hefty patch!

 
Last edited:
Finally, today's Windows 10 update fixed D3D9 device enumeration issues! (which causes some games to crash to desktop). So yes, it was a d3d9.dll bug (just checked, it has been updated right now).
Alessio,
any chance of a link with the info as I am really curious.
I thought the issue was linked to the WDDM 2.1 update (mentioned it earlier in the thread) as part of the AU for Windows 10, albeit very indirectly I agree, like you say it is to do with d3d9.dll but introduced by AU and WDDM 2.1
The issues started with the more recent drivers linked with the WDDM 2.1 change.
Thanks

Edit:
Yep just checked some did prove using pre WDDM 2.1 drivers with Win10 AU worked for the games, using correct associated drivers with WDDM 2.1 and AU broke the games, but using those newest drivers without AU (rolled back Win10) also worked.
 
Last edited:
Chek your Windows
Alessio,
any chance of a link with the info as I am really curious.
I thought the issue was linked to the WDDM 2.1 update (mentioned it earlier in the thread) as part of the AU for Windows 10, albeit very indirectly I agree, like you say it is to do with d3d9.dll but introduced by AU and WDDM 2.1
The issues started with the more recent drivers linked with the WDDM 2.1 change.
Thanks

Edit:
Yep just checked some did prove using pre WDDM 2.1 drivers with Win10 AU worked for the games, using correct associated drivers with WDDM 2.1 and AU broke the games, but using those newest drivers without AU (rolled back Win10) also worked.
The issue came with Redstone and WDDM 2.1 drivers, Redstone + WDDM 2.0 drivers combo was not affected. With the last Redstone Update, your D3D9.DLLs (both system32 and syswow64) should be versioned as 10.0.14393.447.

I am able to run those games (F3 and FNV) with this update and the last AMD drivers.
 
Chek your Windows

The issue came with Redstone and WDDM 2.1 drivers, Redstone + WDDM 2.0 drivers combo was not affected. With the last Redstone Update, your D3D9.DLLs (both system32 and syswow64) should be versioned as 10.0.14393.447.

I am able to run those games (F3 and FNV) with this update and the last AMD drivers.

I tried downloading the update and restarting, but I'm on 10.0.14393.351. I'm on the latest nvidia drivers...

Did the update just not go through or something?

Edit : Just now it began properly updating. It must have lied about the download percentage. Fucking Windows.
 
The Division: Patch 1.5 with DirectX 12 rendering path


The-Division-1.5-PTS-DirectX-12-vs.-DirectX-11-Performance-Test-Benchmark-1-pcgh.png

http://www.pcgameshardware.de/The-Division-Spiel-37399/News/Patch-15-DirectX-12-1212814/
 
On a fast CPU and Nvidia the performance is equal or worse, as usual. Now we need an old i5 w/AMD to see what it does.
 
On a fast CPU and Nvidia the performance is equal or worse, as usual. Now we need an old i5 w/AMD to see what it does.

I'm nearly embarrassed by the great performance I have with my RX 480. I did not quite realize how much the rendering differences would make.

It is sweet as can be. Sorry to be a fangirl, but it is readily apparent to me.
 
i've tried DX12 in BF1 and it isn't great right now, some maps are ok but then other have consistent massive cpu spikes. But one thing i found interesting is that it really felt like input latency was lower as i result a was over correcting. I flicked back and forth between DX11 and DX12 and it didn't seem like placebo effect, but i was hardly scientific.
 
There's often little point in adding support to a new API to an existing engine as there's most often too much code to change the underlying logic which mismatch the new API logic.
Hell I know commercial D3D9 based engines using D3D11 API in old D3D9 style, which leads to no performance benefit (obviously it still benefits from new features such as Compute...)
So not surprised, I'd be surprised to see much improvement, especially in an initial convertion.
 
Chek your Windows

The issue came with Redstone and WDDM 2.1 drivers, Redstone + WDDM 2.0 drivers combo was not affected. With the last Redstone Update, your D3D9.DLLs (both system32 and syswow64) should be versioned as 10.0.14393.447.

I am able to run those games (F3 and FNV) with this update and the last AMD drivers.
Not disagreeing, just saying the source of the problem was introduced by the Windows AU update with WDDM 2.1 combined with the supported drivers, which I appreciate they now fixed in latest update.
Glad to hear you tested it ok to show the update now works, was an annoying problem for quite a few people on both Nvidia and AMD.
Makes me think this is one benefit with the Win10 Pro version if have the cash to burn as a retail consumer, as it lets you delay/choose updates (to a certain extent).

Cheers
 
Last edited:
If that's the case how would comparison's be made
There's often little point in adding support to a new API to an existing engine as there's most often too much code to change the underlying logic which mismatch the new API logic.
Hell I know commercial D3D9 based engines using D3D11 API in old D3D9 style, which leads to no performance benefit (obviously it still benefits from new features such as Compute...)
So not surprised, I'd be surprised to see much improvement, especially in an initial convertion.
So how would performance comparisons be performed against older API to confirm there has in fact been performance increases due to the newer API?
 
Back
Top