Folding Forum

I came to the Forum to see if anyone had any results for their Folding rigs running a patched version of Windows or using the latest NVIDIA drivers due to the Meltdown and Spectre exploits. I guess most people who are serious are running dedicated machine(s) that are not used for general purpose computing and are not subject to running untrusted code, but Windows 10 does have that annoying tendency to enforce patching when you least expect it to. I myself do not plan on patching my standalone Ubuntu rigs as all they do is Fold, but some day I will need to install new Drivers if I upgrade my GPUs.

For more information and a compilation of manufacturers' links, see https://meltdownattack.com

Intel: https://security-center.intel.com/advisories.aspx
AMD: https://www.amd.com/en/corporate/speculative-execution
NVIDIA: http://nvidia.custhelp.com/app/answers/detail/a_id/4611
(note AMD doesn't seem to address GPUs at the time I posted this. They may add a new link for GPU drivers later.)

Microsoft: https://portal.msrc.microsoft.com/en-US ... /ADV180002

Meltdown: In the F@H space, only Intel CPUs will be susceptible to Meltdown. There a POWER CPUs that also are, but no F@H code for POWER. Some ARM designs are susceptible but they have not yet found their way into products. Fixes for Meltdown will slow OS calls, but F@H does not make many (Checkpoints and time stamps are both going to be affected, but are rare compared to the time spent folding) Linux has the GPU drivers in user space, so there is no slow down unique to GPU WUs. My understanding is that Windows Graphics drivers are partially in Kernel space, which may slow them somewhat. But recall, that communication between the CPU and GPU is already so slow that the F@H code minimizes the number of times it calls those routines.

Spectre: It is not clear that anyone even knows how to prevent Spectre. So no code to slow us down, yet. Attempt to reduce Spectre over the web have been inside browser code, which may slow down the Chrome client. I would be surprised if the Chrome client tried anything that triggered mitigation, but I am very naive on it.

Windows 10 got the update this week, and I didn't see any differences in GPU folding performance on my three systems ...

nvidia driver 390.xx will also have fixes for Meltdown, so this could make a difference too.

I tried 390.65 with security fixes on Windows 7 with gtx 1080ti but there is no difference compared to 388.xx

Windows general benchmarks Games and Storage
https://www.techspot.com/article/1556-m ... e-windows/

I'm seeing a big discrepancy between my total estimated PPD reported by FAHControl (48% higher) versus EOC stats. Anything to do with these updates or are the stats hinky???
https://folding.extremeoverclocking.com ... 1&t=224497

Stats down this holiday weekend.

Aurum wrote:I'm seeing a big discrepancy between my total estimated PPD reported by FAHControl (48% higher) versus EOC stats. Anything to do with these updates or are the stats hinky???
https://folding.extremeoverclocking.com ... 1&t=224497

There are a few WUs that give slightly higher PPDs than others.

Theodore wrote:

Aurum wrote:I'm seeing a big discrepancy between my total estimated PPD reported by FAHControl (48% higher) versus EOC stats. Anything to do with these updates or are the stats hinky???
https://folding.extremeoverclocking.com ... 1&t=224497

There are a few WUs that give slightly higher PPDs than others.

When they put FAH on BOINC I'll be able to see for myself.

A project to put FAH on BOINC was started years ago and was abandoned due to differences in the way deadlines and bonus points were handled.

The projects processed by the Chrome client could probably be adapted for BOINC since (A) they don't earn a bonus and (B) the deadlines are extremely short so if BOINC decides to share the CPU resources with some other project, FAH will abandon the assigned WUs that have been suspended. I don't expect a project to do that will be met with enthusiastic support.