Folding Forum

by **Napoleon** » Mon Jun 24, 2013 12:39 pm

I recently set up an 9400GT to fold with wine, using Installation instructions for v6.41 GPU in Linux - thanks, Jesse!
I didn't follow those instructions to the letter, but I got v6.41 working just fine, running in ~/gpufah folder. Which led me to think if I could apply those instructions to v7 Windows tarball release as well. I figured Windows tarball release of v7 is as close as it gets to the v6 GPU console client. So...

Code: Select all: wget https://fah-web.stanford.edu/file-releases/beta/release/fahclient/windows-xp-32bit/v7.3/fahclient_7.3.6-32bit-release.tar.bz2 tar xjf fahclient_7.3.6-32bit-release.tar.bz2 mv fahclient_7.3.6-32bit-release winfah cd winfah/win32/ rm *.dll cp ~/gpufah/*.so . ln -s cudart.dll.so cudart32_30_14.dll ln -s cudart.dll.so cudart.dll ln -s cudart.dll.so nvcuda.dll ln -s cufft.dll.so cufft32_30_14.dll ln -s cufft.dll.so cufft.dll

I figured the first hurdle would be to get "wine FAHClient.exe --lspci", in order to detect the pre-Fermi GPU properly. At first launch, WINE needed some specific library in a specific place - problem solved following the instructions at http://askubuntu.com/a/295931. But then:

Code: Select all: wine FAHClient.exe --lspci fixme:toolhelp:CreateToolhelp32Snapshot Unimplemented: heap list snapshot fixme:toolhelp:Heap32ListFirst : stub fixme:toolhelp:CreateToolhelp32Snapshot Unimplemented: heap list snapshot fixme:toolhelp:Heap32ListFirst : stub VendorID:DeviceID:Vendor Name:Description

I presume the above means some (FAHClient.exe) compatibility issue with WINE? Native Linux lspci works just fine, of course. FWIW, I can do CPU only folding in WINE easily enough. I have a native Linux tarball client for CPU in ~/fah, so after quitting the native client - "cp ~/fah/config.xml ~/winfah" and "wine FAHClient.exe":

Code: Select all: *********************** Log Started 2013-06-24T12:19:20Z *********************** 12:19:20:************************* Folding@home Client ************************* 12:19:20: Website: http://folding.stanford.edu/ 12:19:20: Copyright: (c) 2009-2013 Stanford University 12:19:20: Author: Joseph Coffland <joseph@cauldrondevelopment.com> 12:19:20: Args: 12:19:20: Config: Z:/home/napoleon/winfah/config.xml 12:19:20:******************************** Build ******************************** 12:19:20: Version: 7.3.6 12:19:20: Date: Feb 18 2013 12:19:20: Time: 15:25:17 12:19:20: SVN Rev: 3923 12:19:20: Branch: fah/trunk/client 12:19:20: Compiler: Intel(R) C++ MSVC 1500 mode 1200 12:19:20: Options: /TP /nologo /EHa /Qdiag-disable:4297,4103,1786,279 /Ox -arch:SSE 12:19:20: /QaxSSE2,SSE3,SSSE3,SSE4.1,SSE4.2 /Qopenmp /Qrestrict /MT /Qmkl 12:19:20: Platform: win32 XP 12:19:20: Bits: 32 12:19:20: Mode: Release 12:19:20:******************************* System ******************************** 12:19:20: CPU: Intel(R) Pentium(R) Dual CPU E2220 @ 2.40GHz 12:19:20: CPU ID: GenuineIntel Family 6 Model 15 Stepping 13 12:19:20: CPUs: 2 12:19:20: Memory: 3.86GiB 12:19:20: Free Memory: 3.35GiB 12:19:20: Threads: WINDOWS_THREADS 12:19:20: Has Battery: false 12:19:20: On Battery: false 12:19:20: UTC offset: 3 12:19:20: PID: 46 12:19:20: CWD: Z:/home/napoleon/winfah 12:19:20: OS: Microsoft Windows XP Service Pack 3 12:19:20: OS Arch: AMD64 12:19:20: GPUs: 0 12:19:20: CUDA: Not detected 12:19:20:Win32 Service: false 12:19:20:*********************************************************************** 12:19:20:<config> 12:19:20:  12:19:20: <max-slot-errors v='1'/> 12:19:20: <max-unit-errors v='1'/> 12:19:20: 12:19:20:  12:19:20: <checkpoint v='30'/> 12:19:20: 12:19:20:  12:19:20: <max-packet-size v='big'/> 12:19:20: <power v='full'/> 12:19:20: 12:19:20:  12:19:20: <allow v='192.168.0.1-192.168.0.199'/> 12:19:20: 12:19:20:  12:19:20: <log-rotate-max v='1000'/> 12:19:20: 12:19:20:  12:19:20: <proxy v=':8080'/> 12:19:20: 12:19:20:  12:19:20: <command-allow-no-pass v='192.168.0.1-192.168.0.199'/> 12:19:20: <password v='***'/> 12:19:20: 12:19:20:  12:19:20: <pause-on-battery v='false'/> 12:19:20: <pause-on-start v='true'/> 12:19:20: 12:19:20:  12:19:20: <passkey v='********************************'/> 12:19:20: <team v='191980'/> 12:19:20: <user v='GREYHOUND_SMP'/> 12:19:20: 12:19:20:  12:19:20: <next-unit-percentage v='100'/> 12:19:20: 12:19:20:  12:19:20: <slot id='0' type='CPU'> 12:19:20: <cpus v='2'/> 12:19:20: </slot> 12:19:20:</config> 12:19:20:Trying to access database... fixme:ntdll:NtLockFile I/O completion on lock not implemented yet 12:19:21:Successfully acquired database lock 12:19:21:Enabled folding slot 00: PAUSED cpu:2 (paused) 12:19:22:Set client configured 12:21:16:FS00:Unpaused 12:21:16:WU00:FS00:Connecting to assign3.stanford.edu:8080 12:21:17:WU00:FS00:Connecting to assign3.stanford.edu:8080 12:21:18:WU00:FS00:News: Welcome to Folding@Home 12:21:18:WU00:FS00:Assigned to work server 171.67.108.35 12:21:18:WU00:FS00:Requesting new work unit for slot 00: READY cpu:2 from 171.67.108.35 12:21:18:WU00:FS00:Connecting to 171.67.108.35:8080 12:21:20:WU00:FS00:Downloading 1001.68KiB 12:21:25:WU00:FS00:Download complete 12:21:25:WU00:FS00:Received Unit: id:00 state:DOWNLOAD error:NO_ERROR project:8702 run:12 clone:29 gen:8 core:0xa4 unit:0x000000086652edb35196ea80c295806b 12:21:25:WU00:FS00:Downloading core from http://www.stanford.edu/~pande/Win32/AMD64/Core_a4.fah 12:21:25:WU00:FS00:Connecting to www.stanford.edu:80 12:21:26:WU00:FS00:FahCore a4: Downloading 2.89MiB 12:21:32:WU00:FS00:FahCore a4: 21.64% 12:21:38:WU00:FS00:FahCore a4: 45.45% 12:21:44:WU00:FS00:FahCore a4: 69.25% 12:21:50:WU00:FS00:FahCore a4: 95.22% 12:21:51:WU00:FS00:FahCore a4: Download complete 12:21:51:WU00:FS00:Valid core signature 12:21:51:WU00:FS00:Unpacked 9.59MiB to cores/www.stanford.edu/~pande/Win32/AMD64/Core_a4.fah/FahCore_a4.exe 12:21:51:WU00:FS00:Starting 12:21:51:WU00:FS00:Running FahCore: Z:\home\napoleon\winfah/FAHCoreWrapper.exe Z:/home/napoleon/winfah/cores/www.stanford.edu/~pande/Win32/AMD64/Core_a4.fah/FahCore_a4.exe -dir 00 -suffix 01 -version 703 -lifeline 46 -checkpoint 30 -np 2 12:21:51:WU00:FS00:Started FahCore on PID 57 12:21:52:WU00:FS00:Core PID:60 12:21:52:WU00:FS00:FahCore 0xa4 started 12:21:53:WU00:FS00:0xa4: 12:21:53:WU00:FS00:0xa4:*------------------------------* 12:21:53:WU00:FS00:0xa4:Folding@Home Gromacs GB Core 12:21:53:WU00:FS00:0xa4:Version 2.27 (Dec. 15, 2010) 12:21:53:WU00:FS00:0xa4: 12:21:53:WU00:FS00:0xa4:Preparing to commence simulation 12:21:53:WU00:FS00:0xa4:- Looking at optimizations... 12:21:53:WU00:FS00:0xa4:- Created dyn 12:21:53:WU00:FS00:0xa4:- Files status OK 12:21:53:WU00:FS00:0xa4:- Expanded 1025209 -> 2544640 (decompressed 248.2 percent) 12:21:53:WU00:FS00:0xa4:Called DecompressByteArray: compressed_data_size=1025209 data_size=2544640, decompressed_data_size=2544640 diff=0 12:21:53:WU00:FS00:0xa4:- Digital signature verified 12:21:53:WU00:FS00:0xa4: 12:21:53:WU00:FS00:0xa4:Project: 8702 (Run 12, Clone 29, Gen 8) 12:21:53:WU00:FS00:0xa4: 12:21:53:WU00:FS00:0xa4:Assembly optimizations on if available. 12:21:53:WU00:FS00:0xa4:Entering M.D. 12:21:59:WU00:FS00:0xa4:Mapping NT from 2 to 2 12:21:59:WU00:FS00:0xa4:Completed 0 out of 250000 steps (0%) 12:30:57:WU00:FS00:0xa4:Completed 2500 out of 250000 steps (1%)

There was a "fixme:ntdll:NtLockFile I/O completion on lock not implemented yet" line in the console from WINE (not present in the log file), but apparently that wasn't critical for FAHClient. Alas, for some weird reason my Windows FAHControl wasn't able to connect to the remote WINE client, but I was able to "unpause" the cpu:2 slot using telnet. Anyway, IMHO the --lspci issue would have to be solved first, the remote client connectivity issue is for later.

Ideas?dll19

by **Joe_H** » Mon Jun 24, 2013 3:10 pm

My understanding from posts in the past from persons who looked into doing this is that the WINE wrapper needs updating before it will work with V7. The original coder of the WINE wrapper has not announced any plan to do so, I don't know what is involved in updating the source code.

by **7im** » Mon Jun 24, 2013 3:31 pm

The V7 client also supports GPUs natively now, and these work units were just released to public status (advanced) on Friday.

by **Napoleon** » Mon Jun 24, 2013 4:10 pm

Yep, but core_17 does not support pre-Fermi GPUs. Granted, core_11 work might not be around for long, so this exercise isn't particularly important. Anyway, I'll experiment some more.

I figured since the FahCore_11.exe is the same for v6 and v7 both, there'd be a theoretical chance at least. FahClient.exe WINE compatibility seems to be 1st problem to solve. There's the --lspci issue, and WINE'd FahClient doesn't detect CUDA either. Note that v6 won't work without -forcegpu switch, so the GPU autodetection is something of a problem with it as well.

Occurred to me as I'm writing this - I haven't tried <gpu-vendor-id v="0xXXXX"/> and <gpu-device-id v="0xXXXX"/> yet. I'll have to see if I can override GPU detection with those, equivalently to -forcegpu switch in v6. There's also gpu-species, but I have no idea what that does.

by **Jesse_V** » Mon Jun 24, 2013 4:36 pm

It's probably "entirely impossible" to get FahCore 17 to run through Wine for pre-Fermi cards at the moment. You'd need to have one of two things happen: 1) correctly emulate of a Visual Studio installation, 2) discover that Nvidia's CUDA JIT stuff is only driver-side and that the driver also brings CUDA JIT to pre-Fermi cards. I think both are unlikely.

7im wrote:The V7 client also supports GPUs natively now, and these work units were just released to public status (advanced) on Friday.

The V7 client has indeed supported FahCore 17 before the FahCore actually came out, but it's a moot point for pre-Fermi GPUs, such as my GT 240m. Proteneer isn't supporting AMD cards in Linux either at the moment, and there's no Wine hacks for those cards either. The only way to get GPU folding in Linux on the old Nvidia cards is to use Wine. It's an old client running an old core, but at least it's being utilized.

Eventually all the projects for these older cores will be finished off, support for pre-Fermi cards will end, and my GPU will be unable to get any work whether I'm running Linux or Windows, but until then, it's folding.

by **Napoleon** » Mon Jun 24, 2013 5:19 pm

No luck with

Code: Select all: ... <gpu-vendor-id v="0x10DE"/> <gpu-device-id v="0x0641"/> ...

since v7.3.6 warns that it doesn't recognize those options. I suppose some older v7 clients did, since they are (still) mentioned in the user guide?

Thus, things come back to the problem in my OP:

wine FAHClient.exe --lspci
fixme:toolhelp:CreateToolhelp32Snapshot Unimplemented: heap list snapshot
fixme:toolhelp:Heap32ListFirst : stub
fixme:toolhelp:CreateToolhelp32Snapshot Unimplemented: heap list snapshot
fixme:toolhelp:Heap32ListFirst : stub
VendorID:DeviceID:Vendor Name:Description

No VendorID:DeviceID listing at all, so no GPU(s) detected.

Even if I got --lspci to work, FahClient might still ignore the GPU slot because no CUDA version is detected. If only I could somehow convince v7 FahClient to launch FahCore_11.exe blindly without any checking...

Sledge Hammer wrote:Trust me, I know what I'm doing.

I'm out of ideas, so case closed as far as I'm concerned.

by **7im** » Mon Jun 24, 2013 5:25 pm

Jesse_V wrote:
The V7 client has indeed supported FahCore 17 before the FahCore actually came out, but it's a moot point for pre-Fermi GPUs, such as my GT 240m. Proteneer isn't supporting AMD cards in Linux either at the moment, and there's no Wine hacks for those cards either. The only way to get GPU folding in Linux on the old Nvidia cards is to use Wine. It's an old client running an old core, but at least it's being utilized. Eventually all the projects for these older cores will be finished off, support for pre-Fermi cards will end, and my GPU will be unable to get any work whether I'm running Linux or Windows, but until then, it's folding.

It's not a matter of Proteneer supporting AMD or not, it's having suitable software and drivers from AMD that hinders that. And there are no Wine hacks for ANY AMD cards, so Wine is no better in that regards either.

Don't take me wrong, I'm not against this. But to answer the OP's question, IMO, yes, it's likely not ever going to be possible to run V7 under Wine, and development in that area has less and less incentive by the day as fah is able to better support GPUs under Linux.

Your 3 solutions are:
1. Stick with the v6.41 for older NV cards. The client will be supported as long as the work units are around. No need to push for V7 here as the newer WUs won't run on these cards anyway.
2. For newer NV cards upgrade to V7. It works.
3. Run Windows while we wait for better AMD support from AMD.

I wish there were better solutions available.

by **TheWolf** » Mon Jun 24, 2013 6:10 pm

7im wrote:Your 3 solutions are:
1. Stick with the v6.41 for older NV cards. The client will be supported as long as the work units are around. No need to push for V7 here as the newer WUs won't run on these cards anyway.

Not true, core 15 and core 17 will run on older hardware, its just not allowed to
Its to force people to upgrade there hardware if they wish to run those WU's.
Folding at home is no longer about folding with what you have but rather upgrade if you wish to fold.
I've ran core 15 on 8600GT, 8800GT/GTS, 9800GT, 240GT, 260GTX, and 295GTX without any problems.
Now running core 17 on a 260GTX and so far without problems. So its not like older hardware can't run them
its just not allowed too.

by **7im** » Mon Jun 24, 2013 6:43 pm

TheWolf wrote:
7im wrote:Your 3 solutions are:
1. Stick with the v6.41 for older NV cards. The client will be supported as long as the work units are around. No need to push for V7 here as the newer WUs won't run on these cards anyway.

Not true, core 15 and core 17 will run on older hardware, its just not allowed to
Its to force people to upgrade there hardware if they wish to run those WU's.
Folding at home is no longer about folding with what you have but rather upgrade if you wish to fold.
I've ran core 15 on 8600GT, 8800GT/GTS, 9800GT, 240GT, 260GTX, and 295GTX without any problems.
Now running core 17 on a 260GTX and so far without problems. So its not like older hardware can't run them
its just not allowed too.

What's true is that these configurations are untested for data accuracy as yet, and you could be sending back corrupted data and not even know it. It's happened before!

by **TheWolf** » Mon Jun 24, 2013 7:02 pm

7im wrote:
TheWolf wrote:
7im wrote:Your 3 solutions are:
1. Stick with the v6.41 for older NV cards. The client will be supported as long as the work units are around. No need to push for V7 here as the newer WUs won't run on these cards anyway.

Not true, core 15 and core 17 will run on older hardware, its just not allowed to
Its to force people to upgrade there hardware if they wish to run those WU's.
Folding at home is no longer about folding with what you have but rather upgrade if you wish to fold.
I've ran core 15 on 8600GT, 8800GT/GTS, 9800GT, 240GT, 260GTX, and 295GTX without any problems.
Now running core 17 on a 260GTX and so far without problems. So its not like older hardware can't run them
its just not allowed too.

What's true is that these configurations are untested for data accuracy as yet, and you could be sending back corrupted data and not even know it. It's happened before!

And just why were they not tested? If there were no test then there is no way to know
if the data would be accurate or not. As said it was a way to try and force people to
upgrade to what PG thinks the standards should be, not what the hardware is capable of.
Opinions are like a holes everyone got one and that's mine an I stand behind mine 100%.

by **PantherX** » Mon Jun 24, 2013 8:38 pm

TheWolf wrote:...Not true, core 15 and core 17 will run on older hardware, its just not allowed to
Its to force people to upgrade there hardware if they wish to run those WU's.
Folding at home is no longer about folding with what you have but rather upgrade if you wish to fold.
I've ran core 15 on 8600GT, 8800GT/GTS, 9800GT, 240GT, 260GTX, and 295GTX without any problems.
Now running core 17 on a 260GTX and so far without problems. So its not like older hardware can't run them
its just not allowed too.

Please note that there are two different versions for FahCore_15:
Per-Fermi version
Fermi or higher version
Thus, you can expect that projects using FahCore_15 to run on the supported hardware. However, it has been a very long time since I got a FahCore_15 WU on my GTX 260. Whatever FahCore meets the demands of the researcher would be used for that project.

Regarding FahCore_17, that shouldn't happen so could you please PM me the details? AFAIK, the lack of support was due to missing hardware features.

by **Napoleon** » Tue Jun 25, 2013 1:49 am

TheWolf wrote:Not true, core 15 and core 17 will run on older hardware, its just not allowed to
Its to force people to upgrade there hardware if they wish to run those WU's.
Folding at home is no longer about folding with what you have but rather upgrade if you wish to fold.
I've ran core 15 on 8600GT, 8800GT/GTS, 9800GT, 240GT, 260GTX, and 295GTX without any problems.
Now running core 17 on a 260GTX and so far without problems. So its not like older hardware can't run them
its just not allowed too.

There's no stopping anyone from (ab)using -forcegpu nvidia_fermi with v6 client. But I've seen too much weird stuff with pre-Fermi GPUs to buy a simple "conspiracy theory" answer.

For example, I could do so with my vanilla 9400 GT card (apparently would work under WINE, too). And it's frigging ancient, only has Compute Capability 1.0! But ION... essentially it's just a mobilized version of 9400 (m G), even upgraded to CC 1.1. With FahCore_11, performance is equal to 9400 GT, but FahCore_15 throws a GPU_MEMTEST_ERROR right at startup... :eo

Another example: that very same ION GPU used to fail instantly with P66xx WUs. Everything else apparently was OK, but not those particular projects. I had to put ION folding on hold until 66xx projects were finished in order to avoid dumping a boatload of WUs.

That sort of unexpected flakiness with pre-Fermi GPUs actually caused PG to pull back core_15 work for pre-Fermi GPUs and reanimate core_11 work back in December: see http://folding.typepad.com/news/2012/12 ... ssues.html. But they did try...

What works for you may not work for others, and there's a chance it only seems to work. Sure it would be nice if somebody could test every GPU/card vendor/driver version/localization/phase-of-the-moon/you-name-it combination under the sun, but let's be realistic: it just can't be done.

There's quite a gap in HW features between CC 1.x (pre-Fermi) and 2.x (Fermi and newer), so that seems like a natural split. As far as specific HW features go, proteneer mentioned atomics at #fah, IIRC. If you take a look at http://en.wikipedia.org/wiki/CUDA#Versi ... ifications, you'll see that CC 1.x is pretty vague about supporting atomic operations, ranging from none to some. No syncthreads features at all, whatever they may be.

I'm getting very speculative here, but what the heck, apparently I'm not the only one...
Let's pretend that locks and mutexes don't exist, and even if they did, they'd be just cosmetic stuff. Just code away, it's not like a few race conditions here and there are ever going to cause any problems in real life. Stuff just works, 100%, never mind the naysayers. :wink:

Pre-Fermi stuff seems rather messy to me... but can we really blame (only) PG for it?

PS: Pentiums with FDIV bug can run scientific calculations, too. The question is, should they be allowed to?

by **7im** » Tue Jun 25, 2013 4:37 am

Raising the client min ver to 7 for core_17 projects eliminates all potential abuse. But the really crappy PPD on older hardware for core_17 WUs already does a good job in that regard.

by **bruce** » Tue Jun 25, 2013 5:13 am

7im wrote:Raising the client min ver to 7 for core_17 projects eliminates all potential abuse. But the really crappy PPD on older hardware for core_17 WUs already does a good job in that regard.

It all depends on your perspective. Doesn't core_17 get the same PPD that older hardware got on core_15/core_16? I think it's just that Core_17 provides a much higher PPD by supporting the added capabilities of the newer hardware.

by **P5-133XL** » Tue Jun 25, 2013 5:50 am

bruce wrote:It all depends on your perspective. Doesn't core_17 get the same PPD that older hardware got on core_15/core_16? I think it's just that Core_17 provides a much higher PPD by supporting the added capabilities of the newer hardware.

No! Core_15/16 got all their points via base points which is a linear function of performance. Core_17 tries to equivalence the old base points by a combination of base+QRB but QRB is exponential. Older HW is generally lower performance than newer and thereby will get less QRB relatively and thus the PPD received will be much lower for lower-end HW or in this case older HW. The exponential portion functions in both directions and not just upwardly.

The way it is currently setup for p8900 it is a disaster for lower-end HW (which includes old HW) and a true coup for newer High-end HW.

Folding Forum

WINE and v7 GPU folding - entirely impossible?

WINE and v7 GPU folding - entirely impossible?

Re: WINE and v7 GPU folding - entirely impossible?

Re: WINE and v7 GPU folding - entirely impossible?

Re: WINE and v7 GPU folding - entirely impossible?

Re: WINE and v7 GPU folding - entirely impossible?

Re: WINE and v7 GPU folding - entirely impossible?

Re: WINE and v7 GPU folding - entirely impossible?

Re: WINE and v7 GPU folding - entirely impossible?

Re: WINE and v7 GPU folding - entirely impossible?

Re: WINE and v7 GPU folding - entirely impossible?

Re: WINE and v7 GPU folding - entirely impossible?

Re: WINE and v7 GPU folding - entirely impossible?

Re: WINE and v7 GPU folding - entirely impossible?

Re: WINE and v7 GPU folding - entirely impossible?

Re: WINE and v7 GPU folding - entirely impossible?

Who is online