Problems with GPU folding, is my card not fast enough?

[sqook]

I've got an ubuntu linux workstation which is already CPU folding along quite happily, and I'd like to add a GPU folding slot to the mix. It has a GeForce GT 630 card, which isn't the fasted GPU around but is more than powerful enough to drive dual 30" monitors. However, I have yet to complete a WU with this card, it seems to be an impossible task for it. The WU expiration just seems too tight for this card, even WUs of 8000 points (the lowest I have ever seen for GPU folding) have an expiration of around 4 days, and the folding client (correctly) estimates that it will take this GPU around 5 days to complete. In fact, every GPU WU I have attempted has an expiration which is either around or before the estimated ETA.

Do I just have bad luck with WUs that have too short expiration periods? Or is this GPU not fast enough to fold?

[7im]

Some cards are not fast enough unless they fold 24/7.

[Zagen30]

How far into a WU are you getting that 5 day estimate? The client doesn't provide an accurate estimate in the first couple of percent done because it doesn't have enough data to make a good prediction. If you're seeing such lengthy estimates, say, 10% in, and it's been folding non-stop, then the estimate is much more likely to be right.

[ChristianVirtual]

Also base on your comment CPU folding was already on. Each GPU need one supporting CPU core. You might want to reduce the setting for the CPU-slot by one core count.

Could you share your first few lines of log file until end of config section ?

Also what diver version you have installed ? It happen in the past that certain combination of GPU/driver are more productive. E.g. I run 319.x for my 660Ti, as newer driver slowed it down.

[tondo5]

hi all, i have the same issue... the eta is always bigger than the available time...
i have a NVIDIA GEFORCE 610M with the latest driver (340)... the laptop in on 24/7...
any idea? maybe is really not powerful enough for this... can i flag it to ask easier WU?
thank you all...

here the log:

Code: Select all

16:11:27:17:127.0.0.1:New Web connection
16:11:42:WU00:FS01:0x17:Completed 0 out of 5000000 steps (0%)
16:11:42:WU00:FS01:0x17:Temperature control disabled. Requirements: single Nvidia GPU, tmax must be < 110 and twait >= 900
16:12:01:FS01:Shutting core down
16:12:01:WU00:FS01:0x17:WARNING:Console control signal 1 on PID 5508
16:12:01:WU00:FS01:0x17:Exiting, please wait. . .
16:12:02:WU00:FS01:0x17:Lost lifeline PID 5488, exiting
16:12:02:WU00:FS01:0x17:ERROR:103: Lost client lifeline
16:12:02:WU00:FS01:0x17:Folding@home Core Shutdown: CLIENT_DIED
16:12:02:WU00:FS01:FahCore returned: INTERRUPTED (102 = 0x66)
16:12:09:WU01:FS00:Starting
16:12:09:WARNING:WU01:FS00:Changed SMP threads from 2 to 3 this can cause some work units to fail
16:12:09:WU01:FS00:Running FahCore: "C:\Program Files (x86)\FAHClient/FAHCoreWrapper.exe" C:/Users/tondo/AppData/Roaming/FAHClient/cores/web.stanford.edu/~pande/Win32/AMD64/Core_a4.fah/FahCore_a4.exe -dir 01 -suffix 01 -version 704 -lifeline 2604 -checkpoint 15 -np 3
16:12:09:WU01:FS00:Started FahCore on PID 3084
16:12:09:WU01:FS00:Core PID:3240
16:12:09:WU01:FS00:FahCore 0xa4 started
16:12:09:WU00:FS01:Starting
16:12:09:WU00:FS01:Running FahCore: "C:\Program Files (x86)\FAHClient/FAHCoreWrapper.exe" C:/Users/tondo/AppData/Roaming/FAHClient/cores/web.stanford.edu/~pande/Win32/AMD64/NVIDIA/Fermi/Core_17.fah/FahCore_17.exe -dir 00 -suffix 01 -version 704 -lifeline 2604 -checkpoint 15 -gpu 0 -gpu-vendor nvidia
16:12:09:WU00:FS01:Started FahCore on PID 3444
16:12:09:WU00:FS01:Core PID:3828
16:12:09:WU00:FS01:FahCore 0x17 started
16:12:10:WU01:FS00:0xa4:
16:12:10:WU01:FS00:0xa4:*------------------------------*
16:12:10:WU01:FS00:0xa4:Folding@Home Gromacs GB Core
16:12:10:WU01:FS00:0xa4:Version 2.27 (Dec. 15, 2010)
16:12:10:WU01:FS00:0xa4:
16:12:10:WU01:FS00:0xa4:Preparing to commence simulation
16:12:10:WU01:FS00:0xa4:- Looking at optimizations...
16:12:10:WU01:FS00:0xa4:- Files status OK
16:12:10:WU01:FS00:0xa4:- Expanded 117963 -> 266400 (decompressed 225.8 percent)
16:12:10:WU01:FS00:0xa4:Called DecompressByteArray: compressed_data_size=117963 data_size=266400, decompressed_data_size=266400 diff=0
16:12:10:WU01:FS00:0xa4:- Digital signature verified
16:12:10:WU01:FS00:0xa4:
16:12:10:WU01:FS00:0xa4:Project: 6368 (Run 3, Clone 13, Gen 120)
16:12:10:WU01:FS00:0xa4:
16:12:10:WU01:FS00:0xa4:Assembly optimizations on if available.
16:12:10:WU01:FS00:0xa4:Entering M.D.
16:12:10:WU00:FS01:0x17:*********************** Log Started 2014-08-26T16:12:09Z ***********************
16:12:10:WU00:FS01:0x17:Project: 9201 (Run 489, Clone 0, Gen 129)
16:12:10:WU00:FS01:0x17:Unit: 0x000000a06652edc45399e9407b2439f3
16:12:10:WU00:FS01:0x17:CPU: 0x00000000000000000000000000000000
16:12:10:WU00:FS01:0x17:Machine: 1
16:12:10:WU00:FS01:0x17:Digital signatures verified
16:12:10:WU00:FS01:0x17:Folding@home GPU core17
16:12:10:WU00:FS01:0x17:Version 0.0.52
16:12:10:Removing old file 'configs/config-20140821-230818.xml'
16:12:10:Saving configuration to config.xml

Mod Edit: Added Code tags to log file

[Joe_H]

First, if possible go back to an older version of the drivers. Versions after 3.27 run cards using chips before the GK110 run folding at about half the speed.

As for flags, there is currently none usable to indicate the desire to get smaller GPU WU's. Persons with similar cards may comment on whether a 610M is capable of handling currently available projects.

[7im]

snip

Persons with similar cards may comment on whether a 610M is capable of handling currently available projects.

I did. Part time folding won't make the deadlines on my GT430 with Core_17 using 327.xx drivers. I have to fold on it 24/7. 1 hour 51 minute frames on P13000 currently.

[sqook]

I have an 8-core i7, it is split into 2 slots, one with 2 cores and the other with 6 cores. I pause the 2 core slot when attempting GPU folding. The GPU has been crunching away on a 8000 credit WU for 3 straight days, estimated time 4.44 days left, timeout is in 2 days (29 Aug). My Windows machine at home, in contrast, is folding a 17000 credit unit but has until 9 September (14 days), and will easily finish this job before the timeout.

It seems that on my Linux machine, the GPU always gets assigned WUs with very short timeouts, usually around 4-5 days. Since the machine is unusable with GPU folding on (see my other thread), I usually do GPU folding part time, but for the purposes of this thread I ran an experiment over the weekend and have been folding full time.

I guess the conclusion is that this card will never finish GPU WUs unless I run it 24/7, which isn't an option here. Oh well. Thanks for the input everybody.

[bruce]

Don't even try to compare a CPU-based WU with a GPU-based WU, but there's some real validity to comparing a WU running on one GPU with a WU running on a different GPU.

First there's the hardware. A GT 630 (aka a GT 430) has 96 unified shaders which run at 875 GHz. Yes, there are a few supported GPUs that have fewer shaders or run at slower speeds, but not many. Mid-range GPUs are easily 10x as productive as yours. That doesn't imply your GPU can't do useful work, but it does mean that meeting the deadlines are going to be a challenge.

Since screen lag is a problem, you're not going to run 24x7 but I have four suggestions. First, find a way to backdate the drivers. Second, don't try to turn on or off folding manually; be sure the slot is configured for idle=true so it automatically run any time your mouse/keyboard are inactive. Third, make sure your power savings settings are configured for 24x7 maximum performance (no sleeping). Fourth, make sure at least one CPU is available to the GPU whenever it's running. (And 5th: Don't trust V7's projected completion time except maybe first thing in the morning when it has just been running continuously for a number of hours.)

you've already tried several of these suggestions, but unless you use all of them, you're just going to be wasting time/power using the GPU to fold. If you still can't make the deadlines, then shut down the GPU and reallocate those extra CPUs to your CPU slot.

[Napoleon]

As for flags, there is currently none usable to indicate the desire to get smaller GPU WU's. Persons with similar cards may comment on whether a 610M is capable of handling currently available projects.

No flags in client v7, but GPU client v6 will get only older core15 work for Fermi and newer NVidia GPUs. I've seen a looong streak of P762x WUs which all have a very generous deadline - Timeout (aka Preferred Deadline) is almost 38 days. Low end GPUs will also get better PPD with v6 client, even though core15 work doesn't have any Quick Return Bonus.

The "no nonsense" v6 console client can also be used in Linux with WINE. Takes a bit of tweaking, but I've had success with Mint14 and Ubuntu 14.04 LTS Server. Very recently I started folding core15 WUs under Mint17 LTS. For some hints, see Installation instructions for v6.41 GPU in Linux. That topic might need some minor updating for Ubuntu14 LTS / Mint17 LTS - I did certain things in a more GUI-oriented way - but it still is a very nice guide.

[tondo5]

thank you all for the information!! i will try to use old driver, then with the v6 or simply use the cpus...
ps the minimum 10 days allowed to all wu doesn't consider the GPU WU i guess? right? thanks again...

[Joe_H]

Different projects have different deadlines, they are not all 10 days. You can look up the deadlines on the Project Summary page that is linked to at the top of the forum. The preferred deadline is needed to be met to get the QRB points for those projects that get bonuses for quick return. WU's turned in after that deadline but before the final deadline get just the base points. No points are credited for WU's returned after the final deadline. The version 7 client should automatically dump WU's after the final deadline passes.