A3 SMP and GPU problems

A

APOLLO

[H]ard|DCer of the Month - March 2009
Joined
Sep 17, 2000
Messages
9,089
I thought didn't have any issues with Windows A3 WUs and my nVidia GPU clients, but I was dead wrong. The new SMP2 client is not working properly in conjunction with my GPU clients. The A3 WUs just stall mid way through. This has affected two of my systems and they are very different hardware-wise (Opteron/Xeon). The only thing in common between these systems is Win XP. There is no effect if I shut down the process manager. The SMP client will only resume if I shut down the GPU clients, and they all need to be shut down. Something is in conflict when both clients run.

Out of curiosity, I started a Linux VM with the new SMP client and it works fine, quite a lot slower, but at least it works with the GPU clients. I guess being in its own environment is shielding it. For now, I'm forced to continue using a Linux VM until this problem is resolved. Has anyone else seen a conflict between clients with the new A3 WUs? At first I believed everything was fine, but when returning to my computer after several hours, the WU did not make any progress whatsoever despite FahSpy still reporting the same PPD (if a client stalls this monitor won't inform the user).
 
I was having some trouble getting the a3 console to give up any CPU cycles for my GPUs. It was pegged at 99% CPU usage in task manager and the GPUs were stalled out. The priorities were set via task manager to low for smp and high for gpus. I'll tinker with it some more. I have to shut it off so my VM can finish it's last unit before restarting the new client.
 
I'm having issues as well. I haven't had much time to play around yet but either I drop 1 minute+ from the smp or lose 5000-7000PPD out of each of my video cards. After I get home from work Ill play around some more if someone hasn't figured something out by then.
 
Flogge + Cypher,

What OS are your running?

This could be an XP thing.
I have win7 x64 and it could not be smoother.
 
I am having the same issue as well. I have a 9500GT that is usually good for 1200-1400PPD and after converting to the new 6.29 client in Windows I am only getting 150-400PPD for the same card.

I have had the same results as APOLLO with running it in a Virtualbox VM with no issues. I am going to try to the -smp 7 flag instead of the -smp 8 flag tonight to see if it helps.
 
Hadrien said:
I am having the same issue as well. I have a 9500GT that is usually good for 1200-1400PPD and after converting to the new 6.29 client in Windows I am only getting 150-400PPD for the same card.

I have had the same results as APOLLO with running it in a Virtualbox VM with no issues. I am going to try to the -smp 7 flag instead of the -smp 8 flag tonight to see if it helps.
Click to expand...

What OS?
 
Hadrien said:
Oops, forgot to post that. Windows Vista x64 and Windows 7 x64.
Click to expand...

I would try setting the priorites for the cores in the process manager.
I still have to do that, for the best results.
 
Yep new core is slowing the GPUs. This is on WinXP 32bit and Win7 64bit. Didn't see it with the previous ones.
 
I am on Vista 64 bit. I tried it again after my VM finished. GPUs were running at high priority when I started the new client at idle priority. After about a minute the smp was using 99% and the GPUs were stalled out. I really hate to go SMP2 because I'm on a x3 CPU, and the GPUs can get by with less than 1 full core.
 
Interesting. It seems that I'm the only one that sees a complete freeze with the SMP2 client when the GPU clients are running. The reverse seems to be true from what I'm gathering with the posts in this thread. It may have something to do with the OS since I'm using 32-bit XP, while everyone else is running something more recent. In any event, there is a conflict of sorts between clients.

I'm thinking of either removing the -advmethods argument or reverting to the older executable. Both of these options are a temporary solution at best because A3 WUs will eventually be moved to general release, and the old client will expire. I just wish the Linux SMP2 VM installs performed similarly to native Windows and would just leave it running this way.
 
I tried the -smp 7 flag instead of the -smp 8 flag and my GPU client ran as normal. There was a ~20sec per frame increase in times but that is worth it to be able to run the GPU client.

My i7-920 is now showing about 6500PPD in Windows 7 x64 with the new SMP client in HFM and the 9500GT is back to around ~1400PPD.
 
My first thought is do you have the gpu client locked to one cpu core or do you let it use any cpu core ??
 
I've donea little more fanagling with my system. this is my sig rig. If I start a WinSMP client and it pulls an a3 unit, within a minute of it starting the SMP takes 99% of the CPU time and runs, while the GPUs are stalled out completely regardless of priority settings. If I set affinity so that the GPUs have access to anything and the SMP has access to CPUs 0 and 1, but not CPU2 the GPUs run normally and the SMP makes no progress, even though it is using 100% of the 2 cores it has access to. I have not found any way to get GPUs and an a3 unit to work at the same time. It says it is running, but nothing happens on one or the other client depending on how I set affinity. rpiority settings will not override the a3 client, it just takes the entire CPU regardless of anything else. My system is still responsive, but the GPU clients are completely stopped unless I isolate a core for them.
 
For the record I'm running Windows 7 Pro x64. I just did a bunch of testing and found out it was all skewed because I had bill's process manager set funny. It seems like bill's process manager doesn't want to listen to what I tell it to do anyway so I'm gonna do some more testing later tonight without it after I get some gaming in.
 
Cypher- said:
For the record I'm running Windows 7 Pro x64. I just did a bunch of testing and found out it was all skewed because I had bill's process manager set funny. It seems like bill's process manager doesn't want to listen to what I tell it to do anyway so I'm gonna do some more testing later tonight without it after I get some gaming in.
Click to expand...

In the Options of Bills make sure automatic priority is disabled.
 
I crap canned it. Took the advmethods flags off and went back to running the older cores. It needs more tweaking.
 
I'm going to try setting the client to use only 80 or 90% of the CPU and see if it will allow the GPUs to run and itself as well. If that and one other tweak doesn't work I'll just turn the VM back on and see if someone else can resolve the issue.
 
Well I think I got everything sorted out. I loose about 30 seconds off of my smp client but its better then nothing. Between my i7 and my Q6000 Ive got a 12000ppd increase so I cant complain. :D

I still gotta change over my laptop at some point and see if that helps any.
 
Has anyone tried WinAFC to see if it can set priorities correctly? I was planning to migrate all of my boxes to the A3, but not if we can't figure out priorities. I'm 50/50 SMP vs GPU2 and don't want to lose my GPU2 PpD to convert.

 
I am folding A3's with a GPU running and do not have to use any 3rd party priority management application. See my thread about a new A3 core coming later today.
 
Tobit said:
I am folding A3's with a GPU running and do not have to use any 3rd party priority management application. See my thread about a new A3 core coming later today.
Click to expand...
I'm looking forward to the new core. I am surprised there is no one else experiencing the same things I'm seeing. Everyone here is reporting varying amounts of slowdowns with their GPU clients in conjunction with the SMP2 client. I am seeing the reverse. I tried it again yesterday following Tiger's suggestion of permitting the GPUs access to all cores but to no avail, unfortunately. The client just stops. Another thing I forgot to mention earlier is that Task Manager still shows full CPU utilization! Now someone explain THAT!! :confused:
 
I just tried using both yesterday myself and my SMP went from a PPD of 1070 to 95 once the GPU (Console client.) kicked in. SMP ran all night and only went from 85% to 87% when it should have been well into the next unit.

Admittedly I have just started doing this, and have done nothing with affinity or the like. I simply installed then ran both. I have been running SMP for some time before though.

I was worried my SMP client would not make it's deadline so I killed the newly started GPU client to allow the SMP the power. In the 30 minutes it had 100% control of the PC, it's gone to 93%. And the PPD is >1000 again, so it's back to normal.

I'd really like to run both for maximum effect but it doesn't seem worth it with such a slow down.
 
Little_4nic8er said:
I just tried using both yesterday myself and my SMP went from a PPD of 1070 to 95 once the GPU (Console client.) kicked in. SMP ran all night and only went from 85% to 87% when it should have been well into the next unit.

Admittedly I have just started doing this, and have done nothing with affinity or the like. I simply installed then ran both. I have been running SMP for some time before though.

I was worried my SMP client would not make it's deadline so I killed the newly started GPU client to allow the SMP the power. In the 30 minutes it had 100% control of the PC, it's gone to 93%. And the PPD is >1000 again, so it's back to normal.

I'd really like to run both for maximum effect but it doesn't seem worth it with such a slow down.
Click to expand...
The problem reported in this thread has been fixed with the new A3 core released shortly after. Delete the A3 core in your folder and the client will DL the current version. If that isn't the problem, try using the -smp 3 flag instead (if you have a quad-core). It has helped tons with my quad-core systems. Let us know what happens if you try either of these fixes.
 
I have both my sytems doing this with no problem. one of them is xp the other is win7 i set smp to idle and gpu to low. both gpu's run wide open no slowdown at all with a minor slowdown on smp 3%-5% correction! sense the gpu's went down the smp slowdown is about 11% on my i5-750 @3.8 11,200 vs 9800 ppd
 
Last edited:
What about 4 gpus? Should be the same story right?
Posted via [H] Mobile Device
 
APOLLO said:
The problem reported in this thread has been fixed with the new A3 core released shortly after. Delete the A3 core in your folder and the client will DL the current version. If that isn't the problem, try using the -smp 3 flag instead (if you have a quad-core). It has helped tons with my quad-core systems. Let us know what happens if you try either of these fixes.
Click to expand...

Ok thanks. I just download it but had trouble getting it to finish the WU I was doing previously. Thankfully I always copy it elsewhere before changing so I'm just running it from there with the -oneunit flag to finish the remaining 36%. Then I'll start a new WU with the what I downloaded and see if both play nicely now.
 
Vaulter98c said:
What about 4 gpus? Should be the same story right?
Posted via [H] Mobile Device
Click to expand...

I would expect 4x the slowdown with 4 cards. And a 4 cards on a 4core each card should be on its own core. How to keep them there is another storie.
 
Sadly nothing has changed.

The GPU client continues to fly through the WUs, while the SMP client gets hung up.

I just don't know which is the better option for a 1 client system. The SMP does 1 big, and slower WU, (1% increment every 6 minutes. Roughly) but it's worth more credit were the GPU can do many more smaller WUs (1% increment every 1.5 minutes. Roughly) for less credit.

Wish I could run both. *sigh* :(

P.S.

Doh! Silly me. I forgot to add -smp 3. Seems doing so has made it take twice as long as before. (1% increment every 12 minutes. Roughly.) But, that is a huge difference compared to the other day. (Were it went up only 2% over night.) I just hope it maintains that speed/output. This has only been running since I first posted this, so we'll see. Thanks for the help. :)

P.P.S.

Seems to be maintaining 12-13 minutes/1% increment while the GPU continues to rocket through it's work.

*Joy* :D

P.P.P.S

Ok, final edit update. I decide to make sure it was the -smp 3 flag by taking it off while I went to have a bath. (God I wish we had a shower.) It took 40 minutes to do 1%.

Setting affinity through W7 does nothing to solve this. It has to be -smp 3.

Happy folding.
 
Last edited:
Little_4nic8er said:
I just don't know which is the better option for a 1 client system. The SMP does 1 big, and slower WU, (1% increment every 6 minutes. Roughly) but it's worth more credit were the GPU can do many more smaller WUs (1% increment every 1.5 minutes. Roughly) for less credit.
Click to expand...
If truly presented with a dilemma of choice between two clients, most people would probably choose the option that garners the most production in PPD.

Wish I could run both. *sigh* :(
Click to expand...
You definitely can, I see no reason why it cannot be done with both clients running simultaneously. The only time I was faced with the prospect of not running both SMP and GPU clients was just following the release of the A3 WUs. Subsequent to that, the problem of halted progress with my SMP client was corrected.

Doh! Silly me. I forgot to add -smp 3. Seems doing so has made it take twice as long as before. (1% increment every 12 minutes. Roughly.) But, that is a huge difference compared to the other day. (Were it went up only 2% over night.) I just hope it maintains that speed/output. This has only been running since I first posted this, so we'll see. Thanks for the help. :)
Click to expand...
I'm not sure I'm following you. After adding the -smp 3 flag, the client progresses twice as long but it was a huge improvement from before? :confused:

Seems to be maintaining 12-13 minutes/1% increment while the GPU continues to rocket through it's work.
Click to expand...
This is roughly what I'm observing with my 4-core systems (dual duals). IOW, anywhere between 9-14 minutes per frame using the -smp 3 flag and running concurrent dual GPU clients. The variation in TPF is due to SMP WU type and more importantly, what WUs my GPU clients are running. Some GPU WUs use a lot more CPU cycles than others.

Ok, final edit update. I decide to make sure it was the -smp 3 flag by taking it off while I went to have a bath. (God I wish we had a shower.) It took 40 minutes to do 1%.
Click to expand...
Yes, if you run GPU clients with a quad-core and do not have any flags set, the A3 SMP WUs will run ridiculously slow, at least that has been the observations I made with my systems. It's unfortunate, but people have mentioned that the SMP2 client and A3 WUs are a work in progress and so far benefit S1366 far and above prior released architectures, unfortunately.

Setting affinity through W7 does nothing to solve this. It has to be -smp 3..
Click to expand...
Don't bother setting affinity/priority through the Windows Task Manager. The results are temporary at best, ineffective at worst. Download one of the recommended third-party affinity tools posted here. You can ask any questions on how to use these apps either here or start a new thread about affinity and priority. We've all been through the ins and outs with affinity, and there's a lot of info posted about it. It works but needs fiddling around with sometimes.

Only on very rare occasions will an affinity utility not work as advertised in a system. It has occurred on one of my systems and a few other users reported this happening but like I said, it's very rare. You can have both the GPU and SMP2 client working on your system even if the SMP client is a little slower than normal. The net affect is still better than running only one client.
 
I'm content, if that was not made clear above. Everything is working as I would expect, though a core shorter is all. But hey, if 3 cores can do the work of 4 faster, who am I to complain. :D

Thank you, I appreciate the help getting this sorted. :)
 
Little_4nic8er said:
Sadly nothing has changed.

The GPU client continues to fly through the WUs, while the SMP client gets hung up.

I just don't know which is the better option for a 1 client system. The SMP does 1 big, and slower WU, (1% increment every 6 minutes. Roughly) but it's worth more credit were the GPU can do many more smaller WUs (1% increment every 1.5 minutes. Roughly) for less credit.

Wish I could run both. *sigh* :(

P.S.

Doh! Silly me. I forgot to add -smp 3. Seems doing so has made it take twice as long as before. (1% increment every 12 minutes. Roughly.) But, that is a huge difference compared to the other day. (Were it went up only 2% over night.) I just hope it maintains that speed/output. This has only been running since I first posted this, so we'll see. Thanks for the help. :)

P.P.S.

Seems to be maintaining 12-13 minutes/1% increment while the GPU continues to rocket through it's work.

*Joy* :D

P.P.P.S

Ok, final edit update. I decide to make sure it was the -smp 3 flag by taking it off while I went to have a bath. (God I wish we had a shower.) It took 40 minutes to do 1%.

Setting affinity through W7 does nothing to solve this. It has to be -smp 3.

Happy folding.
Click to expand...

what cpu do you have?
 
Little_4nic8er said:
I'm content, if that was not made clear above. Everything is working as I would expect, though a core shorter is all. But hey, if 3 cores can do the work of 4 faster, who am I to complain. :D
Click to expand...
Yeah, I am faced with the same or similar situation with two of my systems.

Thank you, I appreciate the help getting this sorted. :)
Click to expand...
No problem. :cool:
 
I noticed yesterday my ppd looked very low. I switched things over to the [email protected] ppd where down 12,000 vs usual 18 or 19 had a a-1 running took 12 hours to do that thing for 1200 friggen points. not sure if this is part of what everyone ealse is seeing or not.
 
lehmann said:
I noticed yesterday my ppd looked very low. I switched things over to the [email protected] ppd where down 12,000 vs usual 18 or 19 had a a-1 running took 12 hours to do that thing for 1200 friggen points. not sure if this is part of what everyone ealse is seeing or not.
Click to expand...
No, A1 WUs are just plain slow and worth little. What this thread dealt with was A3 WUs and conflict with GPU clients. A1 WUs don't conflict but they are very slow running.
 
You must log in or register to reply here.
Back
Top