Windows 10 [unsupported for now] performance

[ipkh]

The folding client on windows 10 is significantly slower than the windows 8.1. My GTX 760 shows an average is 2 days to complete a Work Unit, my 560Ti and CPU show a more normal 12 hours. Could the work units be that different? Any ideas to track down the issue world new helpful.
I have both installed and am dual booting for various games. Windows 10 shows extremely long completion times and is not drawing the same power as 8.1.
Setting up the 2 GPUs and 2 of my CPU cores used about 220 watts total from the battery backup monitor.
Win 8.1 sees this increase to 350-380 watts.
Both systems have the same drivers.

[Joe_H]

There are a number of different projects and their WU's are of different sizes and complexities. So unless you are comparing completion times for WU's from the same projects, the run times may very well be different.

Beyond that, the GPU drivers for Windows 10 may not yet have been fully optimized.

[bruce]

The drivers that are distributed with the (unsupported) version of Win-10 include a number of new features. Unfortunately, new features are often unoptimized so a rational guess is that something has been added that isn't optimum for FAH. WDDM 2.0 and DX12 are a pretty radical changes so that would be my first guess but that's very likely an incorrect guess. The Win-10 previews have probably been tested sufficiently to avoid crashes when displaying normal screen graphics and probably had rather limited performance testing of OpenCL.

Some of the drivers for Win-7 include WDDM 2.0. Has anybody tried them and tested FAH Performance? (Personally, I've avoided them so I don't have any 1st hand experience).

[uddarts]

windows 10 updates are said to be mandatory when released. the common thought is that will also include drivers.
at the moment there is nothing to confirm that but if so, that could present a problem.

edit: there is a setting to halt driver updates. system/advanced system settings/hardware/device installation settings



ud

[bruce]

We can hope that NVidia recognizes the problem and is able to optimize their code sufficiently that the PPD returns to the expected values in some later driver. I predict that Microsoft will not accept drivers that don't meet the specifications that they set forth for Win-10, whatever they are. I really doubt that Stanford will be able to do anything about the issue.