650M 10.8.4 GPU drivers

[leman]

Just wanted to let you know that the 10.8.4 seems to bring another significant GPU driver upgrade. My score in Unigine Valley is up 8% (from 590 to 640). Now to compare it to the scores in Windows... will do it some time later.

Update: Windows 7 x64 benchmarks, Unigine Valley, same settings (1680x1050, Ultra, no AA)

DX9: 771
DX11: 649
OpenGL: 551

As we can see, the OS X drivers are still quite a bit slower than the DX9 version of the benchmark under Windows (~20%). Although, OS X performs much better than the OpenGL version under Windows. Weird thing: when running the GL version of the benchmark, the GPU temperatures are significantly lower on both OSes, so is the VRAM clock. I wonder whether there is some problem with the code of the benchmark, or with the driver - which prevents the automatic overclocking from working. We are talking about 500Mhz (25% OC from the base), so its a pretty big deal! Would also explain the discrepancy between the DX and GL benchmarks...

 

[B...]

Hopefully 10.9 is even better.

 

[leman]

Hopefuly 10.9 will have OpenGL 4.1... or even better, 4.3 - just for the OpenGL ES compatibility. I have no idea why Apple takes so long with this stuff, it would be a great opportunity for them - games that work on both OS X and iOS!

 

[B...]

Or at the very least have those updated 4000 graphics.

 

[jafingi]

Apple really needs to add full OpenGL 4.1 support, to make it the gaming platform they want.

 

[theuserjohnny]

Or at the very least have those updated 4000 graphics.

Yes please! My scores on CineBench have improved with the updated drivers but I would like for them to at least get those Intel 4000 drivers out for OSX as well!

 

[nontroppo]

Does anyone know that the newer Intel 4000 drivers (v15.31) didn't make it into 10.8.4? Not at home to check on my MBP...

Oh, thanks leman for the info!

 

[leman]

Does anyone know that the newer Intel 4000 drivers (v15.31) didn't make it into 10.8.4? Not at home to check on my MBP...

Oh, thanks leman for the info!

These drivers have nothing to do with OS X, they are windows drivers. And version numbers do not correspond either.


P.S. Added benchmarks under Windows

 

[nontroppo]

These drivers have nothing to do with OS X, they are windows drivers. And version numbers do not correspond either.

Not that any of us know, but I'm always curious how Apple's GPU drivers are coded, surely they get direct support from Intel/NVidia/ATI, and that they must basically port driver code to work in Darwin? I would thus hope that Apple's hardware partners share updated code that makes their hardware work better in Apple's product.

On the other hand we've seen features on Windows/Linux (hardware accelerated video for example) that made it to OS X substantially later (years in the case of video)...

 

[nontroppo]

Did the 650M support OpenCL 1.2 in 10.8.3? It does seem that the Intel 4000 does support OpenCL V1.2 which was one of the "features" of the new Windows drivers:

 

[Mackan]

Not that any of us know, but I'm always curious how Apple's GPU drivers are coded, surely they get direct support from Intel/NVidia/ATI, and that they must basically port driver code to work in Darwin? I would thus hope that Apple's hardware partners share updated code that makes their hardware work better in Apple's product.

On the other hand we've seen features on Windows/Linux (hardware accelerated video for example) that made it to OS X substantially later (years in the case of video)...

Some people insist that Apple themselves code all their drivers. But that seems like a lot of work. Probably there's something between, they get some kind of driver from ATI/Nvidia/Intel that they improve/port/optimize upon.

 

[leman]

Did the 650M support OpenCL 1.2 in 10.8.3? It does seem that the Intel 4000 does support OpenCL V1.2 which was one of the "features" of the new Windows drivers:

Image

Well, this is what clGetDeviceInfo() returns on my 10.8.4 installation:

Device Intel(R) Core(TM) i7-3615QM CPU @ 2.30GHz supports (8 execution units)OpenCL 1.2 
Device GeForce GT 650M supports (2 execution units)OpenCL 1.1

It seems that 1.2 OpenCL is CPU-only (so no HD4000 support at all).

The OpenGL Extension Viewer seems to be misleading, as it just returns the highest OpenCL device it can find... even if there is no GPU acceleration.

 

[netkas]

Just wanted to let you know that the 10.8.4 seems to bring another significant GPU driver upgrade. My score in Unigine Valley is up 8% (from 590 to 640). Now to compare it to the scores in Windows... will do it some time later.

Update: Windows 7 x64 benchmarks, Unigine Valley, same settings (1680x1050, Ultra, no AA)

DX9: 771
DX11: 649
OpenGL: 551

As we can see, the OS X drivers are still quite a bit slower than the DX9 version of the benchmark under Windows (~20%). Although, OS X performs much better than the OpenGL version under Windows. Weird thing: when running the GL version of the benchmark, the GPU temperatures are significantly lower on both OSes, so is the VRAM clock. I wonder whether there is some problem with the code of the benchmark, or with the driver - which prevents the automatic overclocking from working. We are talking about 500Mhz (25% OC from the base), so its a pretty big deal! Would also explain the discrepancy between the DX and GL benchmarks...

you CAN'T compare DX9, DX11 and OPENGL scores, because of different picture quality. For example DX11 and window's openGL do tesselation, which DX9 or mac opengl can't do.

So your scores tell nothing. you can only compare prev osx results to current one

 

[leman]

you CAN'T compare DX9, DX11 and OPENGL scores, because of different picture quality. For example DX11 and window's openGL do tesselation, which DX9 or mac opengl can't do.

So your scores tell nothing. you can only compare prev osx results to current one

Hi Netkas, you are right of course. Comparing the scores doesn't only compare the driver quality but also the benchmark implementation details, sorry if my previous post was not very clear on this point. I do believe, however, that comparing the OS X and the DX9 scores is more or less legitimate (of course, the question remains how efficient the benchmark code is). And - I find it puzzling that both GL versions run with a significantly cooler GPU (and reduced VRAM clocks on Windows) compared to the DX versions of the benchmark.

 

[jafingi]

Hi Netkas, you are right of course. Comparing the scores doesn't only compare the driver quality but also the benchmark implementation details, sorry if my previous post was not very clear on this point. I do believe, however, that comparing the OS X and the DX9 scores is more or less legitimate (of course, the question remains how efficient the benchmark code is). And - I find it puzzling that both GL versions run with a significantly cooler GPU (and reduced VRAM clocks on Windows) compared to the DX versions of the benchmark.

How can you say that comparing apples with bananas is a legitimate benchmark?

 

[leman]

How can you say that comparing apples with bananas is a legitimate benchmark?

Because the API capabilities are more or less equivalent. Of course, I have no idea how these benchmarks are implemented - but I think its safe to assume that the image quality is comparable (i.e., thet DX9 and OS X GL version do not do things too differently).

 

[JordanNZ]

Because the API capabilities are more or less equivalent. Of course, I have no idea how these benchmarks are implemented - but I think its safe to assume that the image quality is comparable (i.e., thet DX9 and OS X GL version do not do things too differently).

It shouldn't matter anyway, because Valley doesn't use Tessellation.
Some interesting numbers with Valley here...

http://www.g-truc.net/post-0552.html#menu

 

[Serban]

Apple said that in 10.9 will have a better view of things.
They compare Diablo 3 on 1080p high with 650M on 10.8.3 has about 55-59 fps and with 10.9 will have over 70fps..so if i think it will on bar with windows