Strange problem with TPF slowdowns

[nimbyfaygo]

This problem is a little tricky to figure out. The system is Q6600, GTX275 running Win 7 64bit. I'm running SMP client with -smp -advmethods set. For example I'm running project 6014 (although I see the same thing with all projects)

After a restart, I load the CPU and GPU clients, after 30min I am currently getting 5:11 TPF, which gives 7.9k PPD. It will stay around the same TPF for a couple hours, then gradually start to slow down. After 8 hours or so my TPF will be around 7:32 or so.

I can't explain it, the core temps are staying around 54-57c, and there isn't any processes that are using CPU time. From Task manager, I see about 96% utilization from FAH client, and 3-4% from the GPU client. All 4 cores are staying pegged. I have tried disabling every process that isn't required, and closing everything except the 2 clients. It doesn't help at all, I'm still seeing the same slowdowns.

I installed intel burn test, and ran 500 instances of that. Through the entire test, I saw 26.8 - 28.4sec each time.

From my findings, it's either the Client acting screwy or windows. I can't pinpoint the problem though.

 

[Kendrak]

malware?

Something eating CPU cycles in the background?

 

[leagle]

Try running a complete WU with the GPU client turned off. I don't know if that is the problem, but it will help with diagnosing the issue.

 

[nimbyfaygo]

malware?

Something eating CPU cycles in the background?

Nope, I have 44 processes listed. Nothing else is even touching the CPU. I'll add screenshot of task manager, maybe someone will spot something odd. I'll look again in the morning and see if anything else is running.

 

[nimbyfaygo]

Try running a complete WU with the GPU client turned off. I don't know if that is the problem, but it will help with diagnosing the issue.

Ya, I'll try that and see what the results are. I have a new WU coming up in 16min. I'll just close the GPU client now.

I set the priority of SMP to "normal" maybe that will help.

I'll just ignore the problem if the GPU client is doing it, I see a steady 9k ppd from it.

 

[lehmann]

thats a lot of stuff running in the background. at least compared to what i run which is half that much stuff in task manager. spool can be set to manuel, search indexer disabled, driver helper service turned off both of them. realsched off, wmpnetwk off, hpswp manuel, logitech, firefox. that stuff eats more power than what it looks like.

 

[nimbyfaygo]

Regardless of all the things I have running, it didn't slow down at all last night. It seems that the GPU client is causing the problems.

Since it's the GPU client, I'll just ignore it since the GPU produces more PPD than the quad.

 

[APOLLO]

The GPU client is parasitical running in conjunction with the SMP client. That is why I use the -smp flag designating a number of cores that is less than the total number of cores available. In a 4-core system, I use the -smp 3 flag, in an 8-core system I use the -smp 7 flag. One core is allocated for the GPU client(s) and isolated using an affinity app. You will see superior performance and minimal parasitism. The only time this method doesn't work is with -bigadv, but it's not what you're running.

 

[nimbyfaygo]

The GPU client is parasitical running in conjunction with the SMP client. That is why I use the -smp flag designating a number of cores that is less than the total number of cores available. In a 4-core system, I use the -smp 3 flag, in an 8-core system I use the -smp 7 flag. One core is allocated for the GPU client(s) and isolated using an affinity app. You will see superior performance and minimal parasitism. The only time this method doesn't work is with -bigadv, but it's not what you're running.

Got ya, I set the -smp 3 flag and verified that it's running on 3 cores.
I set the CPU affinity for 0,1,2 and the GPU client for 3. Task manager shows that it's working correctly.

I'll report back with results later.
This is good stuff. I know that my system is showing it's age, I consider this great practice getting everything from the overclocks, to the settings optimized. That way I can get the most out of what I have.

 

[nimbyfaygo]

I have been running the - smp 3 flag since yesterday, and I've seen positive results from it.

My TPF on the A3 core has increased by about 10%, but the gpu client is going faster. Overall I'm seeing about 500-800 ppd loss running -smp 3. But the upside is that the system has dropped its power usage a little, its running about 4c cooler, and daily usage isn't increasing the TPF at all.

I am going to be migrating my webserver to this machine. Traffic is pretty light, and I'm not worried about that, but all the different apps running is sure to use up some processor time.
So -smp 3 is going to work out nicely. Thanks APOLLO

 

[nimbyfaygo]

I have another annoyance. To recap, my Q6600 is running -smp 3. I'm also running mysql, apache, exchange and ftp.
I'm rather disappointed right now, my PPD on the A3 client has dropped to roughly 3500. Especially with the 6701's and 02's The server is taking up a lot more resources than I thought. I've also had to drop the OC to 3200 to keep the stability maximized.
It would seem that my mobo just doesn't like the power draw @ 3600, it will stay stable for days then I'll get the random blue screen.

To top it off, my gtx275 has starting giving me fits about the 1620 shader clock. Its started crashing the driver. I've kept an eye on temps, and it's holding steady at 78c
Well bumping the shaders down to 1560 has done the trick. But it's frustrating.

I want a new system, but my bank is being held hostage. I spent quite a bit working on my G5 case, and were moving next month. So of course extra spending is going to have to wait.

 

[APOLLO]

I have another annoyance. To recap, my Q6600 is running -smp 3. I'm also running mysql, apache, exchange and ftp.
I'm rather disappointed right now, my PPD on the A3 client has dropped to roughly 3500. Especially with the 6701's and 02's

There's really not much we can do about the P670x WUs. They are a major PITA and they've been numerous since last week. You can try running the -smp 4 flag again, but if your system is anything like mine it probably will get you similar or worse PPD. It still might be worth trying.

The server is taking up a lot more resources than I thought. I've also had to drop the OC to 3200 to keep the stability maximized.

If you want to maintain a higher OC, the only thing I would recommend is better cooling, especially if you're in a hot region or are experiencing a heat wave.

To top it off, my gtx275 has starting giving me fits about the 1620 shader clock. Its started crashing the driver. I've kept an eye on temps, and it's holding steady at 78c
Well bumping the shaders down to 1560 has done the trick. But it's frustrating.

Again, in this case I highly recommend superior cooling. It doesn't necessarily mean a new HSF, it could be as simple as removing the side panel of your case (this will help with CPU temps as well) or adding a fan blowing against the side of the video card. I have all my systems arranged this way and not only are the temps several degrees lower but the usable life of the card as well as other components will be extended.

Best of luck.