Only 1 GPU working

[c303a]

Since I upgraded to the latest Boinc only 1 gpu will work even though I have re-read the cc config file. I haven't had the problem before. I have reset and re-booted. No changes.

[John P. Myers]

What does your event log say?

[c303a]

5/2/2013 5:01:15 PM | | Re-reading cc_config.xml
5/2/2013 5:01:15 PM | | Not using a proxy
5/2/2013 5:01:15 PM | | Config: simulate 8 CPUs
5/2/2013 5:01:15 PM | | Config: use all coprocessors
5/2/2013 5:01:15 PM | | log flags: file_xfer, sched_ops, task


Everything seems normal. I haven't changed anything.

[John P. Myers]

Does the event log admit to seeing the GPU though or is nothing listed?

[c303a]

5/2/2013 6:33:33 PM | GPUGRID | Sending scheduler request: To fetch work.
5/2/2013 6:33:33 PM | GPUGRID | Requesting new tasks for NVIDIA
5/2/2013 6:33:37 PM | GPUGRID | Scheduler request completed: got 0 new tasks
5/2/2013 6:33:37 PM | GPUGRID | No tasks sent
5/2/2013 6:33:37 PM | GPUGRID | This computer has reached a limit on tasks in progress
5/2/2013 6:37:17 PM | | Re-reading cc_config.xml
5/2/2013 6:37:17 PM | | Not using a proxy
5/2/2013 6:37:17 PM | | Config: simulate 8 CPUs
5/2/2013 6:37:17 PM | | Config: use all coprocessors
5/2/2013 6:37:17 PM | | log flags: file_xfer, sched_ops, task

This is all I get but I do have 4 tasks loaded. MSI shows that both GPU's are active and the same with Device manager. Weird. Maybe I should throw out the config file and re do it.

[John P. Myers]

Which 2 GPUs are they?

And what i was getting at with my above questions was the info posted in the event log at startup. For example:

5/2/2013 2:52:55 PM | | NVIDIA GPU 0: GeForce GT 430 (driver version 314.7, CUDA version 5.0, compute capability 2.1, 512MB, 8380873MB available, 269 GFLOPS peak)
5/2/2013 2:52:55 PM | | ATI GPU 0: Cypress (CAL version 1.4.1741, 1024MB, 991MB available, 5440 GFLOPS peak)
5/2/2013 2:52:55 PM | | OpenCL: NVIDIA GPU 0: GeForce GT 430 (driver version 314.07, device version OpenCL 1.1 CUDA, 512MB, 8380873MB available)
5/2/2013 2:52:55 PM | | OpenCL: ATI GPU 0: Cypress (driver version 1016.4 (VM), device version OpenCL 1.2 AMD-APP (1016.4), 1024MB, 991MB available)

[Crazybob]

Also, what project are you trying to run? Some projects will use both GPU's to run 1 work unit. If you are running a project that uses a large amount of CPU processing, one GPU might sit idle waiting for a CPU to free up. Just a couple of things to think about.

[zombie67]

GPUGRID allows only 2 tasks per GPU. 1 in progress and 1 in the queue.

Is there a need for more?

GPUGRID pays a bonus for early returned tasks. A big queue would be a bad idea.

[Slicker]

The latest BOINC has issues with OpenCL apps when the GPUs are from different vendors. It used to be that one would get the device number by platform (1st AMD GPU, 2nd AMD GPU, 1st NVidia GPU, etc.). That worked great for CAL (AMD) and CUDA (NVidia). It sucks rocks with OpenCL since OpenCL works with both. The issue is that BOINC considers OpenCL a sub-type of CAL or CUDA when it should be a separate processor. Also, there doesn't seem to be any rhyme or reason to how OpenCL lists the processors. It __may__ not list them in the same order every time. To screw things up further, not all AMD GPUs can do CAL and not all can do OpenCL but most can do both. Keeping track of which is which and making sure BOINC can tell when the GPU is busy when one project uses CAL and another uses OpenCL is not easy (even if the BOINC design didn't suck).

So..... for best results, if you really want two GPUs in a box, get to of the same type and two of the same kind. Anything else may or may not work as expected. Until the GPU vendors come up with a way to check if "GPU is in use" for both OpenCL and CAL/CUDA drivers it really can't be fixed.

[c303a]

5/3/2013 9:21:38 AM | | Starting BOINC client version 7.0.64 for windows_x86_64
5/3/2013 9:21:38 AM | | log flags: file_xfer, sched_ops, task
5/3/2013 9:21:38 AM | | Libraries: libcurl/7.25.0 OpenSSL/1.0.1 zlib/1.2.6
5/3/2013 9:21:38 AM | | Data directory: D:\ProgramData\BOINC
5/3/2013 9:21:38 AM | | Running under account Mike
5/3/2013 9:21:38 AM | | Processor: 8 GenuineIntel Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz [Family 6 Model 42 Stepping 7]
5/3/2013 9:21:38 AM | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 cx16 sse4_1 sse4_2 popcnt aes syscall nx lm vmx tm2 pbe
5/3/2013 9:21:38 AM | | OS: Microsoft Windows 8: Professional x64 Edition, (06.02.9200.00)
5/3/2013 9:21:38 AM | | Memory: 15.97 GB physical, 31.97 GB virtual
5/3/2013 9:21:38 AM | | Disk: 111.66 GB total, 109.20 GB free
5/3/2013 9:21:38 AM | | Local time is UTC -5 hours
5/3/2013 9:21:38 AM | | CUDA: NVIDIA GPU 0: GeForce GTX 570 (driver version 314.22, CUDA version 5.0, compute capability 2.0, 1280MB, 1057MB available, 1530 GFLOPS peak)
5/3/2013 9:21:38 AM | | CUDA: NVIDIA GPU 1: GeForce GTX 570 (driver version 314.22, CUDA version 5.0, compute capability 2.0, 1280MB, 1057MB available, 1405 GFLOPS peak)
5/3/2013 9:21:38 AM | | OpenCL: NVIDIA GPU 0: GeForce GTX 570 (driver version 314.22, device version OpenCL 1.1 CUDA, 1280MB, 1057MB available, 1530 GFLOPS peak)
5/3/2013 9:21:38 AM | | OpenCL: NVIDIA GPU 1: GeForce GTX 570 (driver version 314.22, device version OpenCL 1.1 CUDA, 1280MB, 1057MB available, 1405 GFLOPS peak)
5/3/2013 9:21:38 AM | | Config: simulate 8 CPUs
5/3/2013 9:21:38 AM | | Config: use all coprocessors

It looks like it sees both GPU's. At the present time I am running Edges, Ibercivis (cpu version), GPU Grid, and WUProp. They are the same project I have been running. Should I just try to go back to the old version of BOINC?

[Fire$torm]

The latest BOINC has issues with OpenCL apps when the GPUs are from different vendors. It used to be that one would get the device number by platform (1st AMD GPU, 2nd AMD GPU, 1st NVidia GPU, etc.). That worked great for CAL (AMD) and CUDA (NVidia). It sucks rocks with OpenCL since OpenCL works with both. The issue is that BOINC considers OpenCL a sub-type of CAL or CUDA when it should be a separate processor. Also, there doesn't seem to be any rhyme or reason to how OpenCL lists the processors. It __may__ not list them in the same order every time. To screw things up further, not all AMD GPUs can do CAL and not all can do OpenCL but most can do both. Keeping track of which is which and making sure BOINC can tell when the GPU is busy when one project uses CAL and another uses OpenCL is not easy (even if the BOINC design didn't suck).

So..... for best results, if you really want two GPUs in a box, get to of the same type and two of the same kind. Anything else may or may not work as expected. Until the GPU vendors come up with a way to check if "GPU is in use" for both OpenCL and CAL/CUDA drivers it really can't be fixed.

Or do what I do, One ATI/AMD and one nVidia. Saves a lot of headaches on some projects. Although with one project, which one do not recall atm, would not let me run CUDA & ATI work at the same time. Anyhoot, that setup might work well for you.

[c303a]

I went ahead and aborted all the work units and loaded GPUGrid, WuProp, and RNA and it is running 2 gpu units again. It must have been something else that was taking up the resources. It seems to be ok now.