DavidGraham
Veteran
They used an older driver as well, P100 scored 133 vs Vega using 375.39 driver, but scored 122 vs Intel using the newer 375.51 driver, which AMD didn't use and opted for an older version!
Which drivers should OEMs use in comparisons? *spawn*
- Thread starter DavidGraham
- Start date
Did you also notice that the 122ms and 133ms systems are radically different which can easily make a big difference on the results?
DavidGraham
Veteran
And what does this have to do with using the latest drivers available?
Maybe they were the latest drivers available when the comparison was done?
DavidGraham
Veteran
375.51 was released on the 5th of April, AMD showed a live comparison just yesterday! 40 days later.
http://www.nvidia.com/download/driverResults.aspx/117178/en-us
In fact, on the 9th of May, NV released a newer 376.66 driver!
http://www.nvidia.com/download/driverResults.aspx/118962/en-us
http://www.nvidia.com/download/driverResults.aspx/117178/en-us
375.66 seems more a bug fixes over the 375.51 (HDMI problem, EGL apps crash and other) ( long lived branch ). this said i dont know if theres difference in performance between both. ( cant say im really in touch with Linux Nvidia drivers )
The drivers used are probably fine as professional drivers are not really comparable to consumer ones in terms of frequent major optimisation updates.They used an older driver as well, P100 scored 133 vs Vega using 375.39 driver, but scored 122 vs Intel using the newer 375.51 driver, which AMD didn't use and opted for an older version!
I am more wary that these complex environments are setup by competitor (when considering CUDA/libraries/framework/code utilised and settings/and to a lesser extent but still influential others such as rather selective batch and matrices size/etc), I feel this way when Intel did it to Nvidia and when Nvidia does it back, and you can add Google as well with their comparisons to Intel and Nvidia.
Cheers
