Contribute
Register

OS X 10.8.3 and NVIDIA 6xx OpenCL Benchmarks

Status
Not open for further replies.
Article: OS X 10.8.3 and NVIDIA 6xx OpenCL Benchmarks

todays test with 7950:
1757

and gtx680
794

both on stock 10.8.3 kexts

I am confident thatApple/nVidia will remedy the issue of poor OpenCl performance; My old 5770HD eats my GTX 660 Ti SC for breakfast in regards to OpenCl performance. Not only does all of the current Mac models use nVidia exclusively, but the release of the QK5000 and GTX 680 Mac editions would lead me to believe that a fix is around the corner.
 
asus sabertooth z77 + ati 7970 vapor x score 1970
 

Attachments

  • Capture d’écran 2013-04-21 à 16.41.46.jpg
    Capture d’écran 2013-04-21 à 16.41.46.jpg
    199.9 KB · Views: 227
those explanation of poor performance in 10.8.3 doesn't explain how it was possible to reach high performance on 10.8.2! ok, luxmark was wrong about graphics card, but somehow showed good performance! Don't you find this is strange?
My Gigabyte 660ti shows 670 in 10.8.3
 
Article: OS X 10.8.3 and NVIDIA 6xx OpenCL Benchmarks

those explanation of poor performance in 10.8.3 doesn't explain how it was possible to reach high performance on 10.8.2! ok, luxmark was wrong about graphics card, but somehow showed good performance! Don't you find this is strange?
My Gigabyte 660ti shows 670 in 10.8.3
What high performance in 10.8.2?

If you read the article correctly, the only score that changed between 10.8.2 and 10.8.3 was LuxMark due to broken OpenCL drivers in 10.8.2
 
Article: OS X 10.8.3 and NVIDIA 6xx OpenCL Benchmarks

What high performance in 10.8.2?

If you read the article correctly, the only score that changed between 10.8.2 and 10.8.3 was LuxMark due to broken OpenCL drivers in 10.8.2

So, with broken video drivers LuxMark showed better performance? Isn't it funny?
It DID show higher performance!

No matter that in 10.8.3 it now correctly detects calculations units count - in 10.8.2 LuxMark was able to do faster calculations (please tell me how it was possible by your theory)!

What I want to say, is that in 10.8.3 something wrong with drivers/luxmark even they shows correct units count now.
 
Article: OS X 10.8.3 and NVIDIA 6xx OpenCL Benchmarks

So, with broken video drivers LuxMark showed better performance? Isn't it funny?
It DID show higher performance!

No matter that in 10.8.3 it now correctly detects calculations units count - in 10.8.2 LuxMark was able to do faster calculations (please tell me how it was possible by your theory)!

What I want to say, is that in 10.8.3 something wrong with drivers/luxmark even they shows correct units count now.
If you followed the links in the article you would have read that Kepler cards are WORSE at OpenCL than Fermi cards due to the architecture change. So if you want the best OpenCL performance then get a Fermi card or an AMD 7xxx card.
 
Article: OS X 10.8.3 and NVIDIA 6xx OpenCL Benchmarks

If you followed the links in the article you would have read that Kepler cards are WORSE at OpenCL than Fermi cards due to the architecture change. So if you want the best OpenCL performance then get a Fermi card or an AMD 7xxx card.
I agree with this fact (fermi - kepler), but not agree with arguments exposed in the article about drivers in 10.8.2 and 3.
How can bad drivers in 8.2 increase performance in bench?
 
gt 640 geforce score 607
 
Why is it that when I replace a GTX 660ti card with a GTX 560ti card that the Luxmark score(default Sala test) goes up from 682 to 1162? Wouldn't you expect that the older 560ti card to have an inferior Luxmark performance? The 560ti has 448 cores and the 660ti has 1344 cores... In all other tests(Geekbench, Novabench, Cinebench OpenGL) the scores are virtually the same. All this with a Z77X-UD5H + i7-3770K and Mountain Lion 10.8.4.
 
Status
Not open for further replies.
Back
Top