poclbm using CPU 100% even though using GPU?

[allinvain]

Hmm if that's the case does anyone know of a windows catalyst driver package version for win 7 64 bit that does not exhibit this annoying 100% cpu usage? I'm currently  using 10.11 (why you may ask, well it's cause it actually works faster than 11.5 with phoenix and the phatk kernel).

[1714043602]

1714043602

[1714043602]

1714043602

[1714043602]

1714043602

[1714043602]

1714043602

[1714043602]

1714043602

[Kluge]

Just set CPU affinity for each miner to one core.

All miners can be shoved onto the same core. Through BIOS, AOD, or similar tool, you can downclock and downvolt individual cores in most modern setups. Shove all miners onto one core, downclock and downvolt that core as much as possible, and losses will be minimal. There is no effect on hash rate by doing this because (for the most part) the CPU usage does not seem related to the work the miner is doing. It does not solve the issue -- one core will still effectively be useless -- but I think this is preferable to the earlier suggestion of downclocking/volting all of the cores.

Edit: Meant to say entire CPU voltage can be lowered more after individually underclocking the core the miners use.

[supa]

My investigations shows that problem may be in driver version.

When I was using Ubuntu 10.10 and NVidia 260.19.36, there are no problems - poclbm was using only 0-5 %
Today I upgraded to Ubuntu 11.04 and fresh NVidia driver 270.41.19, and now poclbm (and phoenix too) are using up to 100%

When I install old NVidia driver 260.19.36 on Ubuntu 11.04, poclbm is digging and "top" shows low cpu usage (0-5%), but unfortunately X server cannot starting complaining about "ABI version mismatch". Now I am on Ubuntu 11.04 with fresh driver, and also seeking for better solution... Maybe I'll have to go back to Ubuntu 10.10.

EDIT: videocard is - nVidia Corporation GT215 [GeForce GT 240] (rev a2) ~20.000 khash/s

This.

I recently moved to Fedora 15 and upgraded drivers for my 8500GT.  poclbm was maxing at 100% CPU across two CPUs.

Threw the machine into run level 3, downloaded the nVidia driver mentioned by the user above (260.19.36), installed, reloaded the module - viola.  poclbm is taking < 1% CPU.

Very, very, very easy to recreate.  Use the versions mentioned above.

[DigDug3445]

But then U are limiting something from doing it's job, no?

I mean -there is a reason the CPU is on a 100%, no? Or actually the fault is that it's doing NOTHING while running on 100%???

I am limiting it but it does not seem to effect my hashrate as long as its set at 57% or so even if i drop it down too something like 20% i do not see any reduction in my hashrate with my two cards, i set it at 57% as at 58% it starts to use about 70w rather than 40w when at 57% mostly so it quieter although the powersaving is also good .
This is because i am still using my computer for search the web watching youtube etc when the two gfx cards are running and i the computer becomes sluggish if it was set real low but does not seem to effect the hashrate just the responsiveness of my system in general.

[Transisto]

My rig has a one core CPU, so that doesn't help.

CPU at 100% from poclbm does not affect hash rate but CPU at 100% from anything else almost stop it.

I can't use the afterburner UI when poclbm use 100%

[allinvain]

My rig has a one core CPU, so that doesn't help.

CPU at 100% from poclbm does not affect hash rate but CPU at 100% from anything else almost stop it.

I can't use the afterburner UI when poclbm use 100%

Yeah that sucks. For you the only option is to go the linux route or to get a cpu with dual cores (AMD Athlon II X2 250 may be a good one)

[SchizophrenicX]

So this have not been solve for ATI yet huh... Man. And I'm having problems for my windows to install correctly. damnit.

[unroar]

It doesn't seem like it should be that hard to debug python the to find where the cpu peg is happening.  Even if it's not poclbm's fault it would still be good to know on which opencl calls this happens.  I don't know opencl or bitcoin otherwise I'd dive in.

[supa]

I don't know opencl or bitcoin otherwise I'd dive in.

Sounds like the perfect opportunity to learn something new, eh?

[SchizophrenicX]

I realized that if I don't overclock the GPU (meaning I disable ATI Overdrive) then poclbm works normally. Idk if this is universal, could someone verify if this is repeatable (forgot the right word for bugs). If it is we might be getting closer to solving this problem.

[allinvain]

I realized that if I don't overclock the GPU (meaning I disable ATI Overdrive) then poclbm works normally. Idk if this is universal, could someone verify if this is repeatable (forgot the right word for bugs). If it is we might be getting closer to solving this problem.

Umm, that sounds odd. I don't overclock at all but I do have ATI Overdrive enabled just to set the fan profile manually, yet I still have high cpu usage. I also have the memory underclocked but not via ATI overdrive but MSI afterburner. Hmm, I guess this weekend I'm going to have to try disabling overdrive and see if this works. But maybe someone can beat me to it and report back their findings here?

[nased0]

My investigations shows that problem may be in driver version.

When I was using Ubuntu 10.10 and NVidia 260.19.36, there are no problems - poclbm was using only 0-5 %
Today I upgraded to Ubuntu 11.04 and fresh NVidia driver 270.41.19, and now poclbm (and phoenix too) are using up to 100%

When I install old NVidia driver 260.19.36 on Ubuntu 11.04, poclbm is digging and "top" shows low cpu usage (0-5%), but unfortunately X server cannot starting complaining about "ABI version mismatch". Now I am on Ubuntu 11.04 with fresh driver, and also seeking for better solution... Maybe I'll have to go back to Ubuntu 10.10.

EDIT: videocard is - nVidia Corporation GT215 [GeForce GT 240] (rev a2) ~20.000 khash/s

IMHO this problem is caused by the OpenCL implementation in the driver, I am testing the RPC miner (CUDA) under Windows 7 and it works fine, with very small CPU usage (< 1%) and 80% of the poclbm efficiency (43 vs 55 mhash/s).

I have tested also the OpenCL version of the RPC miner, and the tests were inconclusive, it used 100% of one of the cores for a few minutes and repeatedly generated 0 khash/s, then crashed because of the memory allocation error.

I have GTX 275 and C2Q 9450 @ 2,66 GHz.

[allinvain]

Well dammit why doesn't AMD fix this sh*t! Is it really so hard to at the very least allow us to turn off the cpu spin state flag off? I'm thinking a simple .ini or config file option would be all that is needed.

[jamesb]

When I was using Ubuntu 10.10 and NVidia 260.19.36, there are no problems - poclbm was using only 0-5 %
Today I upgraded to Ubuntu 11.04 and fresh NVidia driver 270.41.19, and now poclbm (and phoenix too) are using up to 100%

When I install old NVidia driver 260.19.36 on Ubuntu 11.04, poclbm is digging and "top" shows low cpu usage (0-5%), but unfortunately X server cannot starting complaining about "ABI version mismatch". Now I am on Ubuntu 11.04 with fresh driver, and also seeking for better solution... Maybe I'll have to go back to Ubuntu 10.10.

EDIT: videocard is - nVidia Corporation GT215 [GeForce GT 240] (rev a2) ~20.000 khash/s

I can confirm this problem too, new drivers 270... makes my cpu work more for nothing! I've downgraded to 260 and now it's fine. Thank you and about your ubuntu problem you should rmmod and modprobe your nvidia kernel module before restarting xorg, or reboot. It should fix your error.

IMHO this problem is caused by the OpenCL implementation in the driver, I am testing the RPC miner (CUDA) under Windows 7 and it works fine, with very small CPU usage (< 1%) and 80% of the poclbm efficiency (43 vs 55 mhash/s). [...]

I have GTX 275 and C2Q 9450 @ 2,66 GHz.

I've got a GTX 285 and the same CPU and got 60 Mhs on linux with poclbm (at max without Xorg). Do you think CUDA could change dramatically the performance? Something like > 65 Mhs?
EDIT: I've tested puddinpop's miner with CUDA and I'm getting less (45-55 Mhs and it has frozen my kernel!) than poclbm OpenCL at "low rate" (-f at 150 I get 56 Mhs) which allows me to use my computer without refreshing latency at all (if I whish I can set -f0 and mine at 60 Mhs without Xorg! How can it be better?).

[dishwara]

Check this link, It MUST solve CPU usage problem.

http://forum.bitcoin.org/index.php?topic=6188.msg168009#msg168009

[allinvain]

I already use 10.9 on two of my dedicated miners and it does not solve the high cpu usage problem. So just running 10 series does not work on windows. I'll look into that granola program. I hope though that I can trust it and it won't be some spyware/malware scam.

[Transisto]

does the phoenix miner exhibit this same problem ?

[Kluge]

does the phoenix miner exhibit this same problem ?

Yes, for at least some of us.

[allinvain]

Guys this is not a miner issue..it's just the opencl drivers...

[kiwiasian]

Guys...
Just set the CPU affinity...