Media Player .NET (MPDN) - D3D HQ GPU Video Renderer [v2.49.0/v1.31.0 27 Dec 2018]

[burfadel]

With test v11, I am getting around 168 fps. It has a quickly flickering big white square in the middle of the window.

With test v10, I am getting 8638.2 fps, give or take. However, the window is black.

AMD R9-280X.

[huhn]

build 11 works.

169 fps good old interop bug i loving it.

[mrcorbo]

With build 11 I'm getting around 37(!) fps on a 270X.

[huhn]

Originally Posted by mrcorbo

With build 11 I'm getting around 37(!) fps on a 270X.

what pcie version are you using?

[Anime Viewer]

Originally Posted by Zachs

OK guys, here's another OpenCL test app.

Debug log now prints the extensions supported as well as the GPU name of your chosen adapterID.
More importantly, I've added Nvidia GPU support. Could you guys with Nvidia GPU (as well as the Optimus guys) give it a go and report back?
As before, I'm interested in min/max frame rate and your GPU model.

Oh and if you don't see flickering / flashing square then it isn't actually working.

Cheers!

I know other Optimus users have already shared their results, but for the record my system (with the gpu in my signature) is reporting ~338 fps (when run in its default size box ~228 if I maximize the box to the 1920x1080 screen size).

Originally Posted by Garteal

Initially it said OpenCL 1.1, but after doing some research I found that NVIDIA had released a hotfix that adds OpenCL 1.2 support amongst other things.

With that I'm getting around 947FPS on a GTX970 in a PCIe2.0x16 slot.

My original test (above) was with OpenCL 1.1 also, after installing 1.2 which was bundled with driver 350.05 Garteal linked I was getting 345 in the default window and 269 with the window maximized. OpenCL and the new driver were an improvement for me too based on that test.

Edit: 350.12 is now out...

[Scyna]

on build 11 i get 170 fps and build 10 like 8.8k on amd 290

[mrcorbo]

Originally Posted by huhn

what pcie version are you using?

2.0/x16

[huhn]

Originally Posted by mrcorbo

2.0/x16

maybe the reason but pcie 2.0 should be able to do better.

[Zachs]

Build 10 wasn't actually working so it wasn't doing anything at all which explains why it's so fast

Oh BTW don't maximize the window as you'd then be causing windows desktop composition to do a bigger memory copy which would affect the result obviously.

[Zachs]

There's actually no requirement for any vendors to do memory sharing on the GPU itself, so technically it isn't actually a bug on AMD's part. But yeah it's painfully slow.

[huhn]

Originally Posted by Zachs

There's actually no requirement for any vendors to do memory sharing on the GPU itself, so technically it isn't actually a bug on AMD's part. But yeah it's painfully slow.

it's not doing a dx9 <-> openCL interop if you think about it.

[Zachs]

You should blame Khronous for defining it as a proper way of sharing surfaces then. It says it's up to the driver to avoid memory copy but stopped short of saying it should when possible. My feeling is it's a hardware design flaw in AMD's discrete cards that it's the actual problem and it can't be fixed unless they redesign the hardware. Just a guess.

[huhn]

dx9 <-> direct compute was fine as far as i remember.

[Zachs]

Yeah but a different set of API. OpenCL works on cards that don't support DirectCompute.

[huhn]

this is just my guess why this maybe isn't hardware related.

http://abload.de/img/directcomputegtugt.png

at least i remembered correctly.

edit: yeah i totally forgot. some people can't reach 24 FPS so they can't use dx9 <-> openCL for any realtime program. i call this a bug and i'm talking about GPU like r7 or r9 series.

[Zachs]

When I said hardware related, I meant its inability to share memory on the GPU without doing a copyback.

[ryrynz]

Can confirm build 11 working on the R5 system.

[Zachs]

What frame rate did you get with it? Around 170 too?

[ryrynz]

LOL. No. R5 is weak. 15 FPS. I can't get the 8570 detected debugview only says there's one platform.

[huhn]

Originally Posted by Zachs

When I said hardware related, I meant its inability to share memory on the GPU without doing a copyback.

but directcompute share memory on the GPU. yes it's a different APi no question but this still gives hopes doesn't it?

but speculating doesn't fix this issue. the issue is that some people can't reach 100 FPS and all AMD card even r5 should be able to do that easily. even which this copyback madness. even intel iGPU can do this and these GPU are running on the CPU ram in the first place...