Upcoming Release of SMP2 Cores

[El_Fiendo]

Originally Posted by sswilson

I'm using Process Lasso. I've got the SMP set to 80% in the SMP config which loads 3 cores to 100% and leaves the 4'th at 80%. Affinity is set for core 3 on all of the GPU clients (as well as pretty well all other non-MS apps), priority is left to IDLE for the SMP and set to ABOVE NORMAL for the GPU clients.

Now that I'm thinking about it.... the only thing I havn't done is turn off the "LOCK CPU CORES" switch in the GPU clients.... that might actually be the problem if they're trying to use a certain core and Process Lasso is fighting to keep them on core 3.........

Have you told Process Lasso to 'more strictly enforce priorities'? Also, I never set affinities to my GPUs and CPUs as I find it more often causes a points decrease than increase. Perhaps removing affinities might be something to try, and only have it rely on priorities. A third thing to try is to set the priority of the GPU clients to 'high' and see if that solves the issue if the above don't work.

My setup (as I can remember it) is 0 affinities, SMP / VM to idle, GPUs to high, set Process Lasso to more strictly enforce priorities (its one of the right most menus, should be found easily), GPUs not locked to CPU cores (if that's the default).

Hopefully there is something there that can help.

I can't remember for certain, but I seem to recall the recomendation of not locking the client to CPU cores via the config. Can't remember why that was though, so it could be as you suggest. At any rate, I always leave that option default (which I believe is off).

[sswilson]

If nothing else works, I can always setup a 4 core VMware with XP inside which will make assigning priorities a lot less demanding.... just set it for the VMware instance and not worry about what's going on inside the box.

But yeah... I'd forgotten about that "aggressively manage priorities" setting in process lasso. I'll have another look at it.

edit: Just a quick observation... it seems to be the fluctuating gpu usage WUs that are greatly affected.... I suspect the 100% usage WUs keep their priority, but the fluctuating ones might be losing PPD because it takes time between each cycle for it to grab the 100% priority back.

[Dead Things]

Do you have the GPU clients themselves set to "slightly higher" priority? Just a shot in the dark Steve, but something similar happened to me once. Even though I had the GPUs set to a higher priority, they were tanking until I changed them to "slightly higher" priority in the client config. Taskman reported them to be operating at the exact same priority as before, but this worked for me.

[Perineum]

Is there a way to make my current VM's run these to get the bonus? I'm not touching MPI crap. Every single time I have touched it has been met with refusal to work, and I don't have any sort of difficulties in following instructions.

[Zero82z]

Originally Posted by Perineum

Is there a way to make my current VM's run these to get the bonus? I'm not touching MPI crap. Every single time I have touched it has been met with refusal to work, and I don't have any sort of difficulties in following instructions.

Upgrade to the 6.29 executable and set the -advmethods flag to run these on your Linux SMP clients. However, the A3 core doesn't actually make use of MPI, although you do have to set it up to run the client in the first place. All you have to do is go through the initial setup process for MPI, and then you don't have to worry about it any more.

[martin_metal_88]

do this SMP update will be able to run 8 core? I have seen some telling to add -smp 8 and -bigadvs flag but how can we add 2 flag? my SMP2 is now running at 15% of CPU ( 1 core ) and can't figure why. Can some one copy past me the complete shortuc to seehow to add these flag correctly.

Thx

[MarkOne]

ah ah Martin it's why I catch on you so easily , your big is not working. I,m using the Linux client. so I just add " -bigadv -smp 8 " must me the same with your client. just add it after the quote " in your short cut . an other one will confirm it

[Dead Things]

For now, -bigadv still runs on the A2 core, which means it's exclusive to the Linux client.

[MarkOne]

Martin just drop this smp thing, go with the linux client in VMware. so easy to use, and with your 860 at 3.5 ghz you will get 20k a day ... don't let me overtake you in 4 days . fight my friend ...

[El_Fiendo]

I'm noticing some issues where Core 14 specifically doesn't play well with Core A3. Sorry to sswilson, I thought my clients were fine. Everything Core 11 runs without throttling, but as soon as I get a Core 14 (or at least any that I've witnessed) all my GPUs slow down to 200 PPD (GSOs) and 400 PPD (GTX 260). Shutting down all clients and restarting the GPUs only does nothing, only deleting the work folder and queue.dat and getting a new work unit. It could be possible its a batch of bad work units as well. So far I've seen the issue with a couple different Core 14 work units.

Process Lasso is up to date, running strict management of priorities with Core 14 set to high priority. It has the same settings as Core 11, which all appear to be running fine with the SMP2 client open. I can't figure it out, but I'm going to shut down the SMP2 console and see if I can get a 'bad' Core 14 without SMP2 running. Has anyone else seen similar?