Woah, AMD Releases OpenGL 4.0 Linux Support!

[Heiko]

stupid edit timer...

I recently heard that nVidia scaled the DP performance down for consumer cards to 1/8th on GF100... For the professional gpgpu cards it is still 1/2 though.

see: http://www.evga.com/FORUMS/tm.aspx?m=136362

and also: http://techreport.com/articles.x/18332/5

I should pause to explain the asterisk next to the unexpectedly low estimate for the GF100's double-precision performance. By all rights, in this architecture, double-precision math should happen at half the speed of single-precision, clean and simple. However, Nvidia has made the decision to limit DP performance in the GeForce versions of the GF100 to 64 FMA ops per clock?one fourth of what the chip can do. This is presumably a product positioning decision intended to encourage serious compute customers to purchase a Tesla version of the GPU instead. Double-precision support doesn't appear to be of any use for real-time graphics, and I doubt many serious GPU-computing customers will want the peak DP rates without the ECC memory that the Tesla cards will provide. But a few poor hackers in Eastern Europe are going to be seriously bummed, and this does mean the Radeon HD 5870 will be substantially faster than any GeForce card at double-precision math, at least in terms of peak rates.

http://www.hexus.net/content/item.php?item=24000&page=3

Delve a little deeper, handily not mentioned in any briefing, and NVIDIA is limiting the double-precision speed of the desktop GF100 part to one-eighth of single-precision throughput, rather than the one-fifth speed of the Radeon HD 5000-series. We'll have to wait for the Tesla parts before that's restored to the one-half speed the GF100 is capable of.

[BlackStar]

I recently heard that nVidia scaled the DP performance down for consumer cards to 1/8th on GF100... For the professional gpgpu cards it is still 1/2 though.

Yeah, just found out today on TechReport's review.

Another interesting/strange fact is that R800 has between 2-4 times the raw flops of Fermi. We won't be seeing that kind of difference in actual use (R800 is said to be more difficult to schedule efficiently), but this is interesting nonetheless.

[cutterjohn]

Hmmm... Tom's hw article seems to imply on this page that DP is still 1/2...

...have to look overall the entirety of all the articles again when I have a chance, i.e. download them to my reader...

[V!NCENT]

PPLLEEAASSEE pay me 1200? for an Opteron 6000!

*I work for hardware*

AMD: "Ehm... PR has been taken, sorry. Any other areas you wanted to apply for?"

[Heiko]

Hmmm... Tom's hw article seems to imply on this page that DP is still 1/2...

...have to look overall the entirety of all the articles again when I have a chance, i.e. download them to my reader...

I have been searching through the GF100 and the Fermi whitepapers. Not a word about single precision or double precision performance in the GF100 whitepaper. The Fermi whitepaper clearly states that DP does half the performance of SP.

There is also a topic on the nVidia gpgpu forums which suggests that DP on GF100 is only 1/8th of the performance of SP:

http://forums.nvidia.com/index.php?showtopic=164417

(some nVidia marketing spokesman is also quoted who states that GF100 does not have the same DP performance as Fermi).

It seems nVidia is very silent about this fact (nowhere mentioning the actual performance of the chip, only suggesting it is the same chip as Fermi... which it is as far as I know, but it is crippled at least with DP performance).

[b15hop]

At first I thought this whole AMD beating nvidia to OpenGL 4 was a April fools joke. Congratulations to AMD if they can pull this off!

[BlackStar]

At first I thought this whole AMD beating nvidia to OpenGL 4 was a April fools joke. Congratulations to AMD if they can pull this off!

They already have. The drivers are out there and the hardware has been available for 7 months. Nvidia released their GL4-capable cards just 5 days ago and they aren't available on the market yet.

[cutterjohn]

Anand tech article also still list 64b operations as 1/2 but AMD's as 1/5...

http://www.anandtech.com/show/2977/nvidia-s-geforce-gtx-480-and-gtx-470-6-months-late-was-it-worth-the-wait-/3

[cutterjohn]

Stupid edit, anyways still haven't had time to go over the other 3 articles. (Haven't even put them on the reader yet either)

[DarkFoss]

I hate to be Mr. Negative around here, and I know this is a beta driver, but...

Since the driver doesn't work on Lucid, I popped in an old Ubuntu 8.10 Live CD I had lying around and installed the driver in there. All the Unigine benchmarks give a black screen and spam the console with shader compiler errors.

Fragment shader failed to compile with the following errors:
ERROR: 0:313: error(#132) Syntax error: 'sample' parse error
ERROR: error(#273) 1 compilation errors.  No code generated

Anyone else tested this driver with any of the Unigine benchmarks on an Evergreen card?

I just assembled my new rig this week with a gigabyte 5770... all of the unigine heaven bench was blackscreened using the beta OpenGL 4.0 driver konsole was filled with errors like

11:06:23 GLShader::loadGeometry(): unknown geometry shader vertex count "" in "core/shaders/common/geometry_base_wireframe.shader" file
11:06:23 GLShader::loadFragment(): error in "core/shaders/common/fragment_base_wireframe.shader" file
11:06:23 defines: ATI,ATI_R800,QUALITY_LOW,QUALITY_MEDIUM,QUALITY_HIGH,MULTISAMPLE_0,USE_INSTANCING,USE_GEOMETRY_SHADER,USE_TEXTURE_ARRAY,USE_SRGB,USE_PARALLAX,USE_OCCLUSION,USE_REFLECTION,OPENGL,USE_PSEUDO_INSTANCING,USE_PSEUDO_TRANSFORM,USE_ARB_SAMPLE_SHADING,USE_ARB_TEXTURE_SNORM,USE_ARB_TEXTURE_MULTISAMPLE,HAS_ARB_DRAW_INSTANCED,BASE_WIREFRAME,MESH_SKINNED
11:06:23 Fragment shader failed to compile with the following errors:
11:06:23 ERROR: 0:113: error(#132) Syntax error: 'sample' parse error
11:06:23 ERROR: error(#273) 1 compilation errors. No code generated

I uninstalled them and installed the Cat 10.3 only had test 10 of 26 go black

GLShader::loadGeometry(): unknown geometry shader vertex count "" in "core/shaders/common/geometry_base_wireframe.shader" file
GLShader::loadGeometry(): unknown geometry shader vertex count "" in "core/shaders/grass/geometry_base_shadow_world.shader" file
GLShader::loadGeometry(): unknown geometry shader vertex count "" in "core/shaders/meshes/geometry_base_shadow_world.shader" file

Benchmark finished
Time: 259.706
Frames: 8497
FPS: 32.7177
Min FPS: 12.9554
Max FPS: 64.0583
Scores: 824.159