DirectX9, slower performance than DX11

[Diathorus]

I noticed [does not matter which game] that DirectX9 is running much slower than DirectX11. I am using the same settings for the plugins. I think this is a revision regression.
Please look into the screenshots; computer specifications are noted in my signature.
I have tried other builds from different compilers, but the performance still suffers...
    (DirectX9)
    (DirectX11)

[dannzen]

it could be that your integrated GPU(CPU) is used with dx9
and because it doesn't support dx11, the nvidia 520 gpu is activated...

thats probably a setup problem...
verify if the nvidia gpu is used with dx9

rightclick dolphin.exe and setup highperformance gpu

[CiroConsentino]

(06-10-2012, 03:25 AM)dannzen Wrote: it could be that your integrated GPU(CPU) is used with dx9
and because it doesn't support dx11, the nvidia 520 gpu is activated...

thats probably a setup problem...
verify if the nvidia gpu is used with dx9

rightclick dolphin.exe and setup highperformance gpu

on my system DX11 runs better than DX9. On DX9 games stutter a lot, but they play fine on DX11.

Currently I'm using OpenGL though. Always get 100% performance on all games. Running at 1920x1080 resolution and "2x native internal resolution". Dolphin is at 3.0 build 716 (latest as of June 02, 2012).

my specs are:
Intel Core I7 2600 3.4GHz
8GB RAM DDR-3
GeForce GTX 570HD 2560MB of VRAM
Windows 7 Ultimate 64-bits
FullHD monitor, at 1920x1080 resolution

[NaturalViolence]

@Diathorus

Settings?
Build?
Games that you've tried?
Hardware?

[etking]

I can confirm this, DX11 is always 1-2 frames faster in SMG and Zelda SS compared to DX9. This was not the case with build 3.0-424 but is with 3.0-716.

System DellXPS17 3D Core i7 Sandy Bridge, GT555M
Settings: Lock Tread to core = on, JIT, HLE, Vsync = on, PerPixel Lighting = On, EFB to Texture, Cache = fast, OpenMP Texture Decoder, 1080p 3x IR, 16xAF, no AA

[MayImilae]

Huh. I just tested it. They're right!? I just gave Super Mario Galaxy a quick test. 3.0-716, 1x Native resolution, EFB to Tex, Tex Cache on Fast, at the observatory start (no moving), HLE with Xaudio, frame limit off.

D3D9 - 66-67fps
D3D11 - 71-72fps

What on earth is going on? Alright, another test. At 3x native, same settings otherwise, I got this.

D3D9 - 64fps
D3D11 - 61fps

Well, that's normal I guess. I tend to keep a bunch of old revs around, so, let's try 3.0-600. Same settings as the first test, with Fast Mipmaps off for fairness.

3.0-600

1x Native
D3D9 - 72fps
D3D11 - 67fps

3x Native
D3D9 - 69fps
D3D11 - 60-61fps

Huh. Looks like that from 600 to 716, DX11 performance has improved for 1x, but DX9 performance took a big hit for both 1x and 3x. I have a few revisions inbetween those, let's try them.

3.0-631

1x Native
D3D9 - 76-77fps
D3D11 - 70-71fps

3x Native
D3D9 - 72fps
D3D11 - 60fps

3.0-682

1x Native
D3D9 - 66fps
D3D11 - 71-72fps

3x Native
D3D9 - 63-64fps
D3D11 - 60-61fps


So what did we learn? 3.0-631 is freakin fast, for some reason. And while DX11 has more or less been stable (even improved some), DX9 took a 10fps hit somewhere between 3.0-631 and 3.0-682. And, despite the performance hit, at 3x native resolution, DX9 is still faster.

Note that I kept a GPU monitor up throughout, and at no point in these tests was my GPU maxed out. This is entirely CPU related. And uh, I didn't pay enough attention to get GPU usage stats, sorry.

[NaturalViolence]

The performance ratio between the two will vary slightly from one build to another, from one game to another, and from certain settings. It will vary a lot from one gpu architecture to another.

[isamu]

Yeah but that isn't the case with Gamecube games....right?

[ExtremeDude2]

(07-12-2012, 09:08 AM)isamu Wrote: Yeah but that isn't the case with Gamecube games....right?

What?

[isamu]

(07-12-2012, 10:06 AM)ExtremeDude2 Wrote:

(07-12-2012, 09:08 AM)isamu Wrote: Yeah but that isn't the case with Gamecube games....right?

What?

It's true. Gamecube games such as F-ZeroGX runs terribly in DX11 but runs great in DX9.