AMD R600g Performance Patches Yield Mixed Results

[Lemonzest]

Of course Xonotic is going to be texture starved on a low end 512MB card! they should run it again with a midrage 1GB card, like the 67xx/68xx cards

[tsuru]

....

However, we're fighting a battle we can't win. S3TC textures need 4x to 8x less memory and would help a lot with this problem. Any driver with S3TC support has a great advantage over a driver without one.

We could also cheat by using the BC7 format for plain RGBA8 textures. That would be a win if we implemented the BC7 encoding on the GPU.

Could ASTC be the go-to algo for the texture compression instead of BC7? My understanding is that it's royalty free and a part of the more recent OpenGL specs. IIRC, it's also "better" than S3TC in many respects. If so, could calls for S3TC use be silently made ASTC calls? If not, what makes BC7 a better candidate?

[marek]

I don't think the hardware can do ASTC. BC7 is a good choice, because it has the same compression ratio as DXT5 and it's not encumbered by patents.

[Rigaldo]

Of course Xonotic is going to be texture starved on a low end 512MB card! they should run it again with a midrage 1GB card, like the 67xx/68xx cards

What's your point? It's performance significantly decreased on the same card, how does it have to do with the graphics RAM amount?

[marek]

Here's some update. We believe the huge performance regression is actually caused by TTM. Put simply, it always synchronizes the CPU with the GPU before a buffer is moved, which is a total performance killer. I think it's a huge mistake that TTM does the synchronization at all, because it's completely unnecessary in our case.

[89c51]

Here's some update. We believe the huge performance regression is actually caused by TTM. Put simply, it always synchronizes the CPU with the GPU before a buffer is moved, which is a total performance killer. I think it's a huge mistake that TTM does the synchronization at all, because it's completely unnecessary in our case.

BTW -and yes i am hijacking- you radeon people broke something in the latest stable kernels related to TTM/GEM.

https://bugs.archlinux.org/task/32095#comment100850

(and there is also another bug report in the kernel bugzilla related to this)

[curaga]

Of course Xonotic is going to be texture starved on a low end 512MB card! they should run it again with a midrage 1GB card, like the 67xx/68xx cards

When did 512mb VRAM become low-end?

[bridgman]

When did 512mb VRAM become low-end?

Roughly 2-3 years ago -- I don't remember exactly when, but there was a point where most graphics card mfgs moved en masse to shipping 1GB as the standard configuration even though it wasn't really required at the time.

[Drago]

Here's some update. We believe the huge performance regression is actually caused by TTM. Put simply, it always synchronizes the CPU with the GPU before a buffer is moved, which is a total performance killer. I think it's a huge mistake that TTM does the synchronization at all, because it's completely unnecessary in our case.

Marek, are there any plans for TTM refactoring, so it doesn't do these synchronizations?

[marek]

Marek, are there any plans for TTM refactoring, so it doesn't do these synchronizations?

Not at the moment, though it's possible there are more issues with TTM. The memory fragmentation comes to mind right now, though I'm not really sure that's the issue.