V9 of modified SETI MB CUDA + opt AP package for full GPU+CPU utilization

[Raistmer]

I installed using "AstroPulse & CudaMB, 1+ GPUs, no Team (req SSE3)".  I do NOT have the <ncpus> in the cc_config file.  I am therefore running 4 AP work units on the CPU's and 1 CUDA on the video card.  This is exactly the configuration I want on my quads.

To be unable to monitor the progress of the CUDA-MB work in Boinc Manager is unsatisfactory.  You seem to feel that it's ok since it is crunching the data.  I with respect, humbly dissagree.

Your config differs from recommended for these packs. So you will see all "nice" side effects of incorrect BOINC scheduling for CUDA tasks. W/o cc_config time to time you will get idle GPU and idle core - it's worse than to have overloaded core time to time IMHO.

[Raistmer]

Been running this 64bit package for a few days now on my i7, and everything seems to go just fine. Seeing a nice speedup on the x64 AK also.

http://setiathome.berkeley.edu/show_host_detail.php?hostid=4496384

Fine. Will do x64 V10 then.

[VoidPilot]

thnx to everyone involved for all the effort on the optimised apps...love them.

I just want to check something out if I may.  I only crunch seti, no other projects.  I have boinc 6.6.9 running on two boxes with Raistmer's V9 package SSSE3.  Both boxes only run 4 threads, 3 on the cpu cores and 1 on the Nvidia graphics card.  Both boxes have cc_config at 5.  1 core in box is idle.  It was the same for the V8 and V7 packages even when I was running 6.6.4, 6.6.3.

Is it supposed to run this way or should there be 5 threads running..

There should be no idle cores.
Check place where you put cc_config. Look first post.

thankyou...i should have it more carefully...again thnx for all the great work...
will try V10 soon...

[|MatMan|]

I don't know if it was asked before: Can I use this mod to crunch on the GPU even with BOINC installed in protected (= service) mode?

If you don't use Vista yes.

thank you for your reply - I'm using Vista64 and BOINC as a service install... so that explains it

[ofp1]

Hi,  my  computer is only receiving Astropulse workunits, upgraded from V8a to v9 ssse3 package, it is crunching 5 AP v5.03 units simultaneous but the GPU is at 52°C, it used to be at 72°C when crunching Seti at Home Ehanced 6.03 Wu's, is this normal not to receive SAH units only extra Large AP that takes 11 Hrs to complete and claiming 1213 Credits? maybe it would be problems from SETI servers, is anyone having the same problem?

Here is a link of my Host
http://setiathome.berkeley.edu/show_host_detail.php?hostid=4711443

Thanks

[Raistmer]

Hi,  my  computer is only receiving Astropulse workunits, upgraded from V8a to v9 ssse3 package, it is crunching 5 AP v5.03 units simultaneous but the GPU is at 52°C, it used to be at 72°C when crunching Seti at Home Ehanced 6.03 Wu's, is this normal not to receive SAH units only extra Large AP that takes 11 Hrs to complete and claiming 1213 Credits? maybe it would be problems from SETI servers, is anyone having the same problem?

Here is a link of my Host
http://setiathome.berkeley.edu/show_host_detail.php?hostid=4711443

Thanks

Reciving tasks has no common with how app works. When you will get some work for CUDA MB try to suspend one of AP tasks to give CUDA MB chance to run (if BOINC will not do it by itself)

[chelski]

Hi - continuing from the topic from Seti NC forums:

Is it too much work for you to help compile a "Non-TEAM" package that does this:  Run AP on CPU cores and leaving the MBs to the CUDA client?

I run the V9 team version (SSSE3) but due to the shorter reporting deadline of the MB WUs, it defaults to run 3 MB WUs instead of 2AP + 1MU(onCUDA).  Tried suspending the MBs as you suggested but this starve the CUDA and run 3 APs instead.

Since the MB seems to be more efficient on CUDA and AP can only be run on CPU, my reasoning is that it should be the preferred configuration esp to clear off old R103 AP (Atropulse 5.00) WUs (unless when we have a situation where no AP WUs are available-then should run the TEAM config)

[Raistmer]

Hi - continuing from the topic from Seti NC forums:

Is it too much work for you to help compile a "Non-TEAM" package that does this:  Run AP on CPU cores and leaving the MBs to the CUDA client?

I run the V9 team version (SSSE3) but due to the shorter reporting deadline of the MB WUs, it defaults to run 3 MB WUs instead of 2AP + 1MU(onCUDA).  Tried suspending the MBs as you suggested but this starve the CUDA and run 3 APs instead.

Since the MB seems to be more efficient on CUDA and AP can only be run on CPU, my reasoning is that it should be the preferred configuration esp to clear off old R103 AP (Atropulse 5.00) WUs (unless when we have a situation where no AP WUs are available-then should run the TEAM config)

Use V9 multi-GPU pack.

[cyclo]

I'm completrly new to all of this. Just stumbled across this page as the default BIONC app didnt seem to be using my GPU (the temp didnt go up anyway). Could anyone tell me if i've done this right.. I have an E8500 + 9800GTX, i set the number of cores in cc_config to 3 and now BIONC is running 3 tasks. My GPU temp has risen to 70c so I can tell its using it. But instead of using Astropulse its now using setiathome_enhanced_6.0.8