Nouveau Gallium3D Now Supports Double-Precision Floating-Point Data Types

[whitecat]

while this is good there still much work needed for good dpm and steam still only sees 256MB vram while my gpu has 1GB.

It's not Mesa fault, it's an application/game issue: Bug 72575 - Several games are unable to detect VRAM size

[smitty3268]

That's a funny way of putting it, Michael

Up until yesterday, only the Gallium3D Softpipe driver has supported GL_ARB_gpu_shader_fp64.

Considering that the softpipe driver only added support 1 day earlier, as part of the preparatory patches adding in Gallium support.

[nanonyme]

Still too much red for my taste. . .
If i were a billionaire i would hire 10 guys dedicated solely for Linux graphics drivers.

I'd wager that listing shows red also if a driver supports only hardware for which the functionality cannot be done at all. IOW misleading matrix

[smitty3268]

I'd wager that listing shows red also if a driver supports only hardware for which the functionality cannot be done at all. IOW misleading matrix

It does, although r300 is really the only driver that affects (and maybe nv50?)

The real reason it's misleading is that some of those lines are for tiny features and some for huge features, and you can't tell any difference between the two.

[dungeon]

Not just yet...
while this is good there still much work needed for good dpm and steam still only sees 256MB vram while my gpu has 1GB.

Well steam gui can look into new glxinfo output - it has that info .

http://cgit.freedesktop.org/mesa/demos/commit/?id=999b6950c644266bb871e79438751bdba2fa2a08

[airlied]

Hm, looks like most of this was already done in July 2014, I wonder what happened to it.

I wrote most of the initial stuff, then some Intel guys took it on, then Ilia decided to push it over the line so I can back to help.

FP64 to me was a tickbox feature to get to GL4 so never a priority.

Dave.

[baffledmollusc]

Now NVIDIA is gonna hate nouveau even more, since doubleprecision performance is intentionally crippled in proprietary driver for GeForce, so that Tesla and Quadro would sell. This patch is threat to their proprietary bussines model.

Is this actually true? I would have expected the reduced double precision rate on consumer-level cards would be a hardware limitation, not a driver one. That is, while the same DP hardware is present on both consumer and workstation cards, it will have been fused off on the consumer cards.

Yes? No?

[log0]

Is this actually true? I would have expected the reduced double precision rate on consumer-level cards would be a hardware limitation, not a driver one. That is, while the same DP hardware is present on both consumer and workstation cards, it will have been fused off on the consumer cards.

Yes? No?

Afaik, you are correct. There is no way to mod the DP performance. But you can enable certain features: http://www.eevblog.com/forum/chat/ha...-counterparts/

[rabcor]

they should just do what AMD is doing, actually with AMD's linux driver strategy combined with freesync, I'm beginning to see nvidia as a shorter straw lately. But they are the ones who are getting ahead of AMD with stacked DRAM, so in the end... we can just hope that nouveau will catch up to nvidias proprietary driver in performance one of these days. although that might just be a pipe dream. :/

[gnufreex]

Is this actually true? I would have expected the reduced double precision rate on consumer-level cards would be a hardware limitation, not a driver one. That is, while the same DP hardware is present on both consumer and workstation cards, it will have been fused off on the consumer cards.

Yes? No?

Older cards are just about driver and firmware. Newer ones have some changes on the PCB wiring, so you cant easily mod it. But GPU is the same. There is a way to mod some cards by soldering some stuff on the PCB.