Folding with a SMP Slot AND a GPU slot (AMD Based Cards only)

[Whaler_99]

One issue that has been coming up - if you install the v7 client and you have a supported AMD card, it will auto configure the SMP slot to use all the cores available (lets say 4 for arguments sake) and it will configure a GPU slot.

The problem - AMD GPU actually requires a single core - and when the SMP slot is "expecting" 4 but only gets 3, your SMP progress, and PPD, go in the toilet.

How To Fix

1 - With the FAHControl program up, on the right hand side, change the drop down from Novice to Expert

2 - Look in the Windows called "Folding Slot". This should list your SMP and GPU(s) slots. Highlight the GPU(s) slot, right-click and choose Pause. This will pause GPU folding and speed up SMP.

3 - Highlight the SMP slot, right-click and choose Finish. This means once finished the current WU, the SMP will then Pause and not download another one.

Clicking Fold/Pause/Finish in the main Window - options beside the Configure and Preferences, affect all the slots.

Now, we will change the number of SMP cores. Do NOT do this without doing the above and finishing the current SMP WU or you can corrupt you current WU.

4 - Click Configure, highlight the SMP slot and click Edit button

5 - Where it says "-1" for the SMP change that to 1 less then the total you have. 3 in the case of a 4 core system, 5 in the case of a 6 core, etc. Essentially one less then the current value of SMP in the main window. (Don't worry where it says multiple of 2 - this only applies to very high core count systems)

6 - Click OK till back to main window.

7 - Highligh each slot, right-click again and choose Fold

Folding resumes and should see much better performance from the SMP slot.

I am currently testing on the dual AMD (so two GPU slots) on if reducing by one core fixes this.

[Mentalguy]

didnt work for me

[RailfanS]

It didn't work for me either. I'm running a 7870 and I've tried 12.8 and 12.11 drivers. The most I can get the GPU client to load my 7870 is about 30%. Even after an hour of running the GPU client progress bar hasn't moved off of 0.00%. Any ideas?

[spleenbegone]

I fold on two amd cards with my home computer. I've found that it's best to keep the SMP at the right number (4 for my i5) and through task manager set the affinity of each gpu to a core. I.E. one on core 0, and one on core 1. Each should take up less of the core, 7-10% each in my experience, and allow the cards to work at full power.

Now, this is done on 6 series cards, so I can't say how it works with the 7 series, but it's worth a try.

[Whaler_99]

Testing away... I have paused SMP folding, enabled GPU folding and low and behold, all 4 cores are running at over 60%... Bloody AMD cards... So, just reducing it by one or even two may not make any difference. Going to try spleenbegone's suggestion.

Also - I am user 12.10 - there is a HUGE bug in the 12.11 drivers, which were beta. Check this link:

http://www.overclock.net/t/1323729/amd-gpu-folding-on-12-11-beta-drivers/130

You can read all of it, I did most. :) Basically need to clean your system and go back to 12.8 for the best or use one of the cutomized 12.11 versions. This affects GPU usage. On 12.10 I am gettting 95-100 usage - just wacks the heck out of my CPU.

[Whaler_99]

I fold on two amd cards with my home computer. I've found that it's best to keep the SMP at the right number (4 for my i5) and through task manager set the affinity of each gpu to a core. I.E. one on core 0, and one on core 1. Each should take up less of the core, 7-10% each in my experience, and allow the cards to work at full power.

Now, this is done on 6 series cards, so I can't say how it works with the 7 series, but it's worth a try.

I it is the fahcore_16.exe*32 (2 of them) instances that you modify the affinity on correct?

[spleenbegone]

I'm using 12.10 as well. Also, I think it may depend on the work unit, as I just got a 11292 and it's shot up.

[Whaler_99]

I have a 11292 and a 11293 - still waiting to see what my GPU gets.

[Whaler_99]

OK, changed the affinity on Process 1 to use core 3, on Process 2 to use core 2 - saw CPU usage on core 0 and 1 drop to zero. Started up folding on the SMP at 4 cores and getting about 5k PPD, a lot better then before. Now, to see if it stays there. :)

[spleenbegone]

5k more, or 5k total? Either way, glad to see there's an improvement.

[Whaler_99]

Right now, one GPU gets about 6K, second GPU about 6.5K and the CPU about 5.5k - it was normally getting like 500 total in this config, so a nice improvement.

[spleenbegone]

WOO!

​Spleenbegone, helping competing teams since two hours ago.

If you have a spare pci-e slot, I suggest getting a 630. I picked up one for $50 for an old work computer turned folding rig, and it turns in a 14k WU every other day without a problem.

[RailfanS]

Okay,

I finally managed to get my 7870 working. I've had the 12.11 drivers installed since they came out, so when I tried the 12.8 drivers earlier I failed to clean the registry after uninstalling 12.11. Anyways after doing that, running ATI Man (mentioned in the link) and installing 12.8 I was able to get the GPU to run at between 68% and 95% After setting the SMP to run on 7 out of 8 cores on my 8120 I was able to get 89%-99% GPU usage without slowing down my SMP too much. Thanks for the help guys, this is the first time I've been able to try GPU folding :)

[Whaler_99]

WOO!

​Spleenbegone, helping competing teams since two hours ago.

If you have a spare pci-e slot, I suggest getting a 630. I picked up one for $50 for an old work computer turned folding rig, and it turns in a 14k WU every other day without a problem.

This is my main rig / gaming system So my 6870's are in xfire - so cannot really add any other GPU's to it. Have a few other systems though I am looking to get NVidia cards for...

[Whaler_99]

WOO!

​Spleenbegone, helping competing teams since two hours ago.

If you have a spare pci-e slot, I suggest getting a 630. I picked up one for $50 for an old work computer turned folding rig, and it turns in a 14k WU every other day without a problem.

And thanks for the help! :)

[spleenbegone]

WOO!

​Spleenbegone, helping competing teams since two hours ago.

If you have a spare pci-e slot, I suggest getting a 630. I picked up one for $50 for an old work computer turned folding rig, and it turns in a 14k WU every other day without a problem.

If you have 3 pci-e slots you can run crossfire and then an nvidia card in the third for folding/physx.

You're very welcome, good luck with your future folding.

[Boosted_i6]

Thanks for the the info. I did the same thing but was only running two cores instead of three

[gasho]

OK I cleaned my system from 13.1 drivers on my 6870 and put it back on 12.8 and put the SMP on 3 following the post but somehow it doesn't work for me. PPD is on about 7K like it was before with only the SMP set on 4 with 13.1 drivers.

Does the GPU need longer to turn up ?

[Whaler_99]

OK I cleaned my system from 13.1 drivers on my 6870 and put it back on 12.8 and put the SMP on 3 following the post but somehow it doesn't work for me. PPD is on about 7K like it was before with only the SMP set on 4 with 13.1 drivers.

Does the GPU need longer to turn up ?

First off, yes, can take a while for the PPD to ramp up. Also, check utilization on the GPU - is it at 99%?

[gasho]

OK I cleaned my system from 13.1 drivers on my 6870 and put it back on 12.8 and put the SMP on 3 following the post but somehow it doesn't work for me. PPD is on about 7K like it was before with only the SMP set on 4 with 13.1 drivers.

Does the GPU need longer to turn up ?

look at the top right corner. there are the stats of my utilization . the GPU is and stayed on 0% ... it just won't turn on. I tried several combinations with the SMP and GPU configuration but nothing seems to work for me. I tried the SMP to put it on 1 because I have a i3 with HT. I tried it on 3 too but nothing.

Question. Do you count in the HT cores too?

[Whaler_99]

OK I cleaned my system from 13.1 drivers on my 6870 and put it back on 12.8 and put the SMP on 3 following the post but somehow it doesn't work for me. PPD is on about 7K like it was before with only the SMP set on 4 with 13.1 drivers.

Does the GPU need longer to turn up ?

OK - I not sure if you have posted, but what GPU do you have? Also - in the sceen shot, the unknown ETA - I wonder if it is even downloading any work units. If you click on the LOG tab. In the there you will see an option to add a check mark to show the log from specific slots. Check it and check your slot. Then take a look at the log file and see what error you are getting?

[gasho]

OK I cleaned my system from 13.1 drivers on my 6870 and put it back on 12.8 and put the SMP on 3 following the post but somehow it doesn't work for me. PPD is on about 7K like it was before with only the SMP set on 4 with 13.1 drivers.

Does the GPU need longer to turn up ?

I'm using a 6870 with 12.8 drivers

Unstable machine is what I got from the log

decompressed_data_size=171163 diff=0

13:39:23:WU00:FS01:0x16:- Digital signature verified

13:39:23:WU00:FS01:0x16:

13:39:23:WU00:FS01:0x16:Project: 11293 (Run 22, Clone 438, Gen 42)

13:39:23:WU00:FS01:0x16:

13:39:23:WU00:FS01:0x16:Assembly optimizations on if available.

13:39:23:WU00:FS01:0x16:Entering M.D.

13:39:24:WU00:FS01:0x16:Tpr hash 00/wudata_01.tpr: 676696103 3529356191 598819968 198289263 3451390094

13:39:24:WU00:FS01:0x16:Working on ALZHEIMER DISEASE AMYLOID

13:39:24:WU00:FS01:0x16:Client config unavailable.

13:39:25:WU00:FS01:0x16:Starting GUI Server

13:39:30:WU00:FS01:0x16:Setting checkpoint frequency: 500000

13:39:30:WU00:FS01:0x16:Completed 3 out of 50000000 steps (0%).

13:39:33:WU00:FS01:0x16:Run: exception thrown in GuardedRun -- cannot continue further.

13:39:33:WU00:FS01:0x16:Going to send back what have done -- stepsTotalG=50000000

13:39:33:WU00:FS01:0x16:Work fraction=0.0004 steps=50000000.

13:39:37:WU00:FS01:0x16:logfile size=0 infoLength=0 edr=0 trr=23

13:39:37:WU00:FS01:0x16:+ Opened results file

13:39:37:WU00:FS01:0x16:- Writing 640 bytes of core data to disk...

13:39:37:WU00:FS01:0x16:Decompress: Memory problem 2

13:39:37:WU00:FS01:0x16: ... Done.

13:39:37:WU00:FS01:0x16:DeleteFrameFiles: successfully deleted file=00/wudata_01.ckp

13:39:37:WU00:FS01:0x16:

13:39:37:WU00:FS01:0x16:Folding@home Core Shutdown: UNSTABLE_MACHINE

13:39:38:WARNING:WU00:FS01:FahCore returned: UNSTABLE_MACHINE (122 = 0x7a)

13:39:38:WU00:FS01:Sending unit results: id:00 state:SEND error:FAULTY project:11293 run:22 clone:438 gen:42 core:0x16 unit:0x000000b16652edbc4d94b9713ef697dc

13:39:38:WU00:FS01:Uploading 512B to 171.67.108.44

13:39:38:WU00:FS01:Connecting to 171.67.108.44:8080

13:39:38:WU00:FS01:Upload complete

13:39:38:WU00:FS01:Server responded WORK_QUIT (404)

13:39:38:WARNING:WU00:FS01:Server did not like results, dumping

13:39:39:WU00:FS01:Cleaning up

[Whaler_99]

From that:

This appears in various forms but appears to be directly related to calculation errors detected by a GPU. Whether the errors are GPU hardware errors or are inherent in the WU is currently unknown. One known cause is switching user sessions in Windows whilst folding or the use of Microsoft's Remote Desktop (both of which disable and cause a reset the GPU, crashing the FahCore). Another known cause is allowing the Windows power-savings mode to "turn off the monitor" for certain driver versions. (This problem has been fixed in may driver versions.)

So, make sure your Power Options are set to something like performance - ensure you are not turning off monitor, and don't use Remote Desktop or user switching if you do. I know here, I was trying to setup a FAH client through RDP and the GPU slot simply failed. That was issue. Walked into my server room, logged in locally, redid it, worked fine. Basically, don't remote into any system running FAH. Sends the GPU into a ttizzy. :)

Any of these help?

[gasho]

no remote desktop no switching users no power savings on.

I don't get it ...

It starts running and starts downloading a new WU for the GPU but then just pops up "failed"

[Whaler_99]

no remote desktop no switching users no power savings on.

I don't get it ...

It starts running and starts downloading a new WU for the GPU but then just pops up "failed"

I have had this work - reinstall FAH over the existing - but, do a custom. When it comes to the spot to use existing config, that is fine. That is your ID, passkey, etc. When it comes though to HOW to run it - by default it is chosen to autostart it and to run FAH Control when you log in. Some GPU have issues with this (My GT 640). So, uncheck the auto start FAH Control and choose, I think, the third option - manually start.

Now, when you "log" in, nada. You have to manually run FAH Control software. And, it doesn't auto start the folding. Click on Configure. Near bottom of pop up, will see a button to Start. Click it then click Save. See if this helps.