FAH on XP x64, CPU & GPU

[GrandAdmiralThrawn]

Hello,

This is my first topic, so please be gentle with me. It's also a really strange request, but I still wanted to ask.

Given that Folding@Home still supports Windows XP according to the website, I tried to join the project today using such an OS, after SETI@Home has finished distributing work. So I downloaded the latest release, only to find that some of the binary programs included are targeting NT 6.0 (Windows Vista+). That includes FAHClient.exe but also workers/cores like the 64-bit version of FahCore_a7.exe and some other binaries. This results in a "not a valid Win32 application error" on older NT systems.

There are no real NT 6.0+ Win32 API dependencies there though. Just fixing up the PE32 and PE32+ headers of the binaries using CFF Explorer did the trick and I'm now actively folding, at least on the CPU, using a 64-bit worker (or "core"?). All one needs to do is to change two fields in the optional PE32/PE32+ headers of all .exe files: MajorSubsystemVersion from 6 to 5 and MinorSubsystemVersion from 0 to 1 (XP 32-bit) or from 0 to 2 (XP x64 or Server 2003). So Windows NT 5.1/5.2 instead of 6.0.

Here's a quick summary of the system:

• CPU: Intel Xeon X5690 (6 cores / 12 threads, ISE up to SSE 4.2)
• RAM: 48GiB DDR3
• GPU: nVidia GeForce GTX Titan Black (Kepler Arch)
• OS: Windows XP Professional x64 Edition SP2

However, there are some things that don't seem to be quite right. An attempt to launch FAHControl.exe either directly or via the tray application results in an error being logged in the console Window: "ERROR:Exception: Failed to create process with: FAHcontrol: This application has failed to start because the application configuration is incorrect. Reinstalling the application may fix this problem." The program fails to start.

Also, while folding on the CPU, it keeps logging tons of these errors: "ERROR:Exception: Accessing './work/00/wuinfo_01-dat': This function is not supported on this system." It keeps making progress though, despite that. Currently I'm at 4% for the given WU.

Occasionally, the same error shows up for "GPU.txt". Not sure if I can even get GPU work on this platform? nVidia Driver version is 368.81, and I've got CUDA 8.0. And... OpenCL 1.2 I think.

I'm aware that this is a very weird and probably deprecated setup for F@H, but I still wanted the machine to contribute.

Does anyone have an idea about how to get FAHControl.exe to work and about how to interpret the error messages in the CLI log? I managed to get the web control part working by manually editing config.xml, but no luck for FAHControl so far.

Also, would you think that making my GeForce GTX Titan Black work for the project using CUDA 8.0 or OpenCL 1.2 on this operating system is possible? Or not? I mean, I do have the APIs, but I don't know if the project can give me compatible (or at least easily binary-hackable) workers or cores that could run here.

Thank you!

[uyaem]

Hello,

This is my first topic, so please be gentle with me.

Why? Thrawn was never gentle with anyone, if I remember the books on my shelf correctly.

Also, while folding on the CPU, it keeps logging tons of these errors: "ERROR:Exception: Accessing './work/00/wuinfo_01-dat': This function is not supported on this system." It keeps making progress though, despite that. Currently I'm at 4% for the given WU.

That doesn't sound good though, I hope someone who's more into the tech of FAH can make an educated guess if that may corrupt the results. Because FAHClient may not be bug-free when it comes to edge cases like this, esp. when the executable has been tampered with.
Can you see that results are uploaded successfully and not rejected (BAD_WORK_UNIT, or similar, I believe is the error message)?

EDIT: Acutally the error message seems to be 'Server did not like results, dumping' (taken from another thread).

[JimboPalmer]

Welcome to Folding@Home!

"ERROR:Exception: Accessing './work/00/wuinfo_01-dat': This function is not supported on this system."

This is almost for sure a file permissions error, most common when you don't use the express install and F@H can't find your work files or lacks permission to change them. It can be the result of installing as one user and running as some other user.

i hope that helps.

https://www.techpowerup.com/gpu-specs/g ... lack.c2549
As hardware, this card supports F@H, I has OpenCL 1.2 and Double Precision floating point math.

I do not see a driver more recent than 368.81 for XP.

(Windows 10 would be a very nice upgrade)

[Joe_H]

I believe you will have problems running on Win XP, even the 64-bit Pro version. As I recall there are a number of libraries and frameworks needed for the client and the folding cores that are not available in the version needed on XP.

[GrandAdmiralThrawn]

Thank you for your replies!

As for Windows 10: I'd rather go for FreeBSD UNIX or some Linux, which I'm more comfortable by now, but not for this machine. The software setup is just too complex, would take me too much time to rebuild (also: too lazy).

I've checked the file permissions and gave my user (the user who executes FAHClient.exe) full control over all F@H folders and files, including the workunit data. That didn't get rid of the messages. Also, during install time I picked "manual" as an option. Well, it said "Expert" there, so maybe I should just go and RTFM...

Also, I haven't finished my first WU yet. I thought I'd try to fix all issues (if there really are any game-breaking ones) before resuming the work.

@Joe_H: Do you know what the exact dependencies are? I've run the client (and core) through a Dependency Walker profiling run (that's essentially a stack + lib trace, like strace/ltrace on Linux or truss on FreeBSD), and found lots of runtime links that seem broken, including kernel32 API calls. As it stands it's probably quite miraculous that it's even crunching.

I'll try to just let it work and see how it goes when the first WU is done.

One more question: Is my client a problem for the project if I do all this? The binary modifications for the .exe files are minimal; They're not changing any actual, functional code, but just some meta data in the headers. But still, I don't want to return bogus data to F@H. Or will that be detected anyway?

[GrandAdmiralThrawn]

My apologies for the double post, but here's what the worker reported upon WU completion:

Code: Select all

17:15:35:WU00:FS00:0xa7:Completed 500000 out of 500000 steps (100%)
17:15:35:WU00:FS00:0xa7:Saving result file ..\logfile_01.txt
17:15:35:WU00:FS00:0xa7:Saving result file frame20.trr
17:15:35:WU00:FS00:0xa7:Saving result file frame20.xtc
17:15:35:WU00:FS00:0xa7:Saving result file md.log
17:15:35:WU00:FS00:0xa7:Saving result file science.log
17:15:35:WU00:FS00:0xa7:Folding@home Core Shutdown: FINISHED_UNIT
17:15:35:WU00:FS00:FahCore returned: FINISHED_UNIT (100 = 0x64)
17:15:35:WU00:FS00:Sending unit results: id:00 state:SEND error:NO_ERROR project:16406 run:1689 clone:3 gen:20 core:0xa7 unit:0x00000016a8f5c67d5e801e5f1e80ef73

I'm assuming that looks pretty ok?

[Neil-B]

… there would be a little more of the log after that (up to and including "Cleaning UP") that would show if the server accepted it without issues.

[GrandAdmiralThrawn]

Ah, you're right! I missed that because there were other messages about no work being available, interrupting the output. Here's the whole thing (removing all log messages not related to the result submission):

Code: Select all

17:15:35:WU00:FS00:0xa7:Completed 500000 out of 500000 steps (100%)
17:15:35:WU00:FS00:0xa7:Saving result file ..\logfile_01.txt
17:15:35:WU00:FS00:0xa7:Saving result file frame20.trr
17:15:35:WU00:FS00:0xa7:Saving result file frame20.xtc
17:15:35:WU00:FS00:0xa7:Saving result file md.log
17:15:35:WU00:FS00:0xa7:Saving result file science.log
17:15:35:WU00:FS00:0xa7:Folding@home Core Shutdown: FINISHED_UNIT
17:15:35:WU00:FS00:FahCore returned: FINISHED_UNIT (100 = 0x64)
17:15:35:WU00:FS00:Sending unit results: id:00 state:SEND error:NO_ERROR project:16406 run:1689 clone:3 gen:20 core:0xa7 unit:0x00000016a8f5c67d5e801e5f1e80ef73
17:15:35:WU00:FS00:Uploading 1.53MiB to 168.245.198.125
17:15:41:WU00:FS00:Upload 28.64%
17:15:47:WU00:FS00:Upload 57.29%
17:15:53:WU00:FS00:Upload 77.75%
17:15:58:WU00:FS00:Upload complete
17:15:58:WU00:FS00:Server responded WORK_ACK (400)
17:15:58:WU00:FS00:Final credit estimate, 935.00 points
17:15:58:WU00:FS00:Cleaning up

That should be the whole thing I assume?

[Neil-B]

Looks like it was uploaded fine … Not showing on the WU status app https://apps.foldingathome.org/wu yet but it can take time for results to get through the process

[GrandAdmiralThrawn]

Hm, the first one still doesn't show up there, but the second WU it completed [does show up], as does [the third one]. Just a matter of time then?

So, does this mean that those completed WUs have been verified somehow as well? My biggest fear here is that I might somehow be sending back bogus result data. I mean, I guess it looks reasonably ok to me, but what would I know.

For SETI@Home they simply sent each WU out multiple times to create a quorum. How does Folding@Home's result verification work?

[Joe_H]

Yes, WUs get a basic verification on upload.

[PantherX]

...For SETI@Home they simply sent each WU out multiple times to create a quorum. How does Folding@Home's result verification work?

F@H is rather unique in this aspect as a WU (under normal conditions) is assigned to a single individual only once. Once it is successfully upload to the F@H Server and verified, you will then be allocated points. If the WU is corrupted (client issues or network issues), the F@H Server will reject it and it will be shown in your log.

[GrandAdmiralThrawn]

Alright, thanks for the input. And one more question: The [MS Windows system requirements] website appears to be quite outdated, as it lists software (like XP) which not really supported anymore. I assume that the GPU driver version information might not be correct anymore either.

Where or how can I find the current, official system requirements of the client for MS Windows?

Still trying to find out how I may be able to fold on my GTX Titan Black with a nVidia v368.81 driver and CUDA 8.0 / OpenCL 1.2. Or do I need OpenCL 2.0?

Thanks, and sorry for all the questions.

[uyaem]

OpenCL 1.2 is fine.
I myself am running an nVidia 1060 GTX, drivers 388.46 though.

Basically, the card needs to support double precision, all projects that can run on single precision floating point have finished last year.
According to nVidia, your card is capable of DP.

And for the sake of completon, quoting from other posts that I've read today:
ATI 7xxx (and some high-end 5xxx and 6xxx) will work too.

[Joe_H]

Just HD 7700 and above for that series, HD 8500 and up for that, and there are still cards in the R5 200 series the t don't support DP.

DP (FP64) is even less supported in the mobile versions of the AMD GPUs. Most outside of the Pro lineup for mobile workstations do not support DP, recently it has started showing as available on some of the recent series.

iGPU implementations on AMD processors pretty much did not support DP until very recently.