A New Patch To Improve Radeon Gallium3D Performance

[phoronix]

Phoronix: A New Patch To Improve Radeon Gallium3D Performance

A new patch has surfaced on the Mesa development list that allows for further performance improvements to the R600 Radeon Gallium3D driver for some OpenGL workloads.

http://www.phoronix.com/vr.php?view=17560

[airlied]

Phoronix: A New Patch To Improve Radeon Gallium3D Performance

A new patch has surfaced on the Mesa development list that allows for further performance improvements to the R600 Radeon Gallium3D driver for some OpenGL workloads.

http://www.phoronix.com/vr.php?view=17560

There is also a good chance this patch will negate the pcie gen2 boost people see in certain benchmarks (not in others), as there was cases where the depth texture was being sampled from GART instead of VRAM, and thus making GART access faster using gen2 made benchmark like lightsmark faster.

Doing the depth texturing from VRAM should make it all just work properly

Of course there may be some benchmarks where gen2 matters still.

Dave.

[Pontostroy]

Nice news.
http://openbenchmarking.org/result/1...BY-HD6770FLU82
hd6770 1920x1080
lightsmark shows 204 fps, without patch 170-180 fps
heroes of newerth 33 fps, without patch 29-30 fps (36 fps with hyperz patch)
It would be nice if someone updated the hyperz patch for current mesa master + improve flushed depth texture handling patch

[curaga]

This is one of the times that shows where Phoronix has value. Benching several cards and several tests showed it did worse for many benches, which would otherwise have come up much later if at all.

[crazycheese]

There is also a good chance this patch will negate the pcie gen2 boost people see in certain benchmarks (not in others), as there was cases where the depth texture was being sampled from GART instead of VRAM, and thus making GART access faster using gen2 made benchmark like lightsmark faster.

Doing the depth texturing from VRAM should make it all just work properly

Of course there may be some benchmarks where gen2 matters still.

Dave.

Maybe its time to implement profiles? Just thinking, one can extend driconf to support them and then implement patches as this as a turnable option (just like tiling works) which could be turned on if application has a profile. I think one cannot evade profiles if there are even slightest plans regarding crossfire/multicard anyway.

[NomadDemon]

per app profile is shit, never implement this shit.

developers must optimise they code, to work on all hardware, not by profiles with uber optimisation...