KnaBench -select GPU

[Caprio]

Hello,

How to configure KnaBench to select a specific GPU if you have multiple installed? The EVGA 670 FTW just arrived and I would like to give it a go.

[Claggy]

Put in a -device 1 or 2 or 3 cmd parameter into the BenchCfg.txt along with the app file name,
you can do multiple lines, with -device 0 on one line, -device 1 on another, etc, the Bench program will then repeat the Wu on each device.

Claggy

[Caprio]

Thanks. I also noticed that Boinc does not use all GPU by default. When I had GTX560ti in both slots they worked. But the following setup does not use all GPU by default. Will the <use all GPU> option in cc_config.xml work?

27/09/2012 20:00:46 |  | Starting BOINC client version 7.0.28 for windows_x86_64
27/09/2012 20:00:46 |  | log flags: file_xfer, sched_ops, task, cpu_sched, dcf_debug, sched_op_debug
27/09/2012 20:00:46 |  | log flags: work_fetch_debug
27/09/2012 20:00:46 |  | Libraries: libcurl/7.25.0 OpenSSL/1.0.1 zlib/1.2.6
27/09/2012 20:00:46 |  | Data directory: C:\ProgramData\BOINC
27/09/2012 20:00:46 |  | Running under account Mark
27/09/2012 20:00:46 |  | Processor: 4 GenuineIntel Intel(R) Core(TM)2 Quad CPU    Q6600  @ 2.40GHz [Family 6 Model 15 Stepping 11]
27/09/2012 20:00:46 |  | Processor: 4.00 MB cache
27/09/2012 20:00:46 |  | 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 syscall nx lm vmx tm2 pbe
27/09/2012 20:00:46 |  | OS: Microsoft Windows Vista: Home Premium x64 Edition, Service Pack 2, (06.00.6002.00)
27/09/2012 20:00:46 |  | Memory: 8.00 GB physical, 7.75 GB virtual
27/09/2012 20:00:46 |  | Disk: 74.53 GB total, 23.36 GB free
27/09/2012 20:00:46 |  | Local time is UTC +3 hours
27/09/2012 20:00:46 |  | NVIDIA GPU 0: GeForce GTX 680 (driver version 306.23, CUDA version 5.0, compute capability 3.0, 2048MB, 8382318MB available, 3252 GFLOPS peak)
27/09/2012 20:00:46 |  | NVIDIA GPU 1 (not used): GeForce GTX 670 (driver version 306.23, CUDA version 5.0, compute capability 3.0, 2048MB, 1961MB available, 2915 GFLOPS peak)
27/09/2012 20:00:46 |  | OpenCL: NVIDIA GPU 0: GeForce GTX 680 (driver version 306.23, device version OpenCL 1.1 CUDA, 2048MB, 8382318MB available)
27/09/2012 20:00:46 |  | OpenCL: NVIDIA GPU 1 (not used): GeForce GTX 670 (driver version 306.23, device version OpenCL 1.1 CUDA, 2048MB, 1961MB available)
27/09/2012 20:00:46 | SETI@home | Found app_info.xml; using anonymous platform
27/09/2012 20:00:46 | Milkyway@home | URL http://milkyway.cs.rpi.edu/milkyway/; Computer ID 344575; resource share 100
27/09/2012 20:00:46 | SETI@home | URL http://setiathome.berkeley.edu/; Computer ID 3299266; resource share 100
27/09/2012 20:00:46 | SETI@home | General prefs: from SETI@home (last modified 10-Nov-2011 21:23:36)
27/09/2012 20:00:46 | SETI@home | Computer location: home
27/09/2012 20:00:46 | SETI@home | General prefs: no separate prefs for home; using your defaults
27/09/2012 20:00:46 |  | Reading preferences override file
27/09/2012 20:00:46 |  | Preferences:
27/09/2012 20:00:46 |  | max memory usage when active: 4094.79MB
27/09/2012 20:00:46 |  | max memory usage when idle: 7370.62MB
27/09/2012 20:00:46 |  | max disk usage: 10.00GB
27/09/2012 20:00:46 |  | (to change preferences, visit the web site of an attached project, or select Preferences in the Manager)
27/09/2012 20:00:46 |  | [work_fetch] Request work fetch: Prefs update
27/09/2012 20:00:46 |  | [work_fetch] Request work fetch: Startup
27/09/2012 20:00:46 |  | Not using a proxy
27/09/2012 20:00:49 | SETI@home | Started download of 27my12aa.31498.9069.9.10.206
27/09/2012 20:00:49 | SETI@home | Started download of 27my12aa.31498.9069.9.10.181
27/09/2012 20:00:49 | SETI@home | Started download of 27my12aa.31498.9069.9.10.184
27/09/2012 20:00:49 | SETI@home | Started download of 27my12ab.7750.99358.6.10.119
27/09/2012 20:00:49 | SETI@home | Started download of 27my12ab.7750.99358.6.10.124
27/09/2012 20:00:49 | SETI@home | Started download of 27my12ab.7750.99358.6.10.113
27/09/2012 20:00:49 | SETI@home | Restarting task 28jn12ab.31494.50634.13.10.119.vlar_1 using setiathome_enhanced version 603 in slot 5
27/09/2012 20:00:49 | SETI@home | Restarting task 10ja10zz.28877.484.13.10.48_0 using setiathome_enhanced version 610 (cuda_fermi) in slot 6
27/09/2012 20:00:49 | SETI@home | Restarting task 26my12ad.2787.14795.16.10.124_0 using setiathome_enhanced version 603 in slot 11
27/09/2012 20:00:49 | SETI@home | Restarting task 10ja10zz.28877.484.13.10.51_0 using setiathome_enhanced version 610 (cuda_fermi) in slot 12
27/09/2012 20:00:49 | SETI@home | Restarting task 10ja10zz.28877.484.13.10.37_0 using setiathome_enhanced version 610 (cuda_fermi) in slot 1
27/09/2012 20:00:49 | SETI@home | Restarting task 26my12ag.11897.11268.5.10.229_1 using setiathome_enhanced version 603 in slot 4
27/09/2012 20:00:49 | SETI@home | Restarting task 26my12ag.11897.11268.5.10.244_0 using setiathome_enhanced version 603 in slot 8
27/09/2012 20:00:49 |  | [work_fetch] work fetch start
27/09/2012 20:00:49 |  | [work_fetch] ------- start work fetch state -------
27/09/2012 20:00:49 |  | [work_fetch] target work buffer: 604800.00 + 43200.00 sec
27/09/2012 20:00:49 | Milkyway@home | [work_fetch] REC 0.000 priority -0.000000 (no new tasks)
27/09/2012 20:00:49 | SETI@home | [work_fetch] REC 106219.560 priority -1.018286
27/09/2012 20:00:49 |  | [work_fetch] CPU: shortfall 217551.42 nidle 0.00 saturated 592577.65 busy 0.00
27/09/2012 20:00:49 | Milkyway@home | [work_fetch] CPU: fetch share 0.000 rsc backoff (dt 0.00, inc 0.00)
27/09/2012 20:00:49 | SETI@home | [work_fetch] CPU: fetch share 0.000 rsc backoff (dt 0.00, inc 0.00)
27/09/2012 20:00:49 |  | [work_fetch] NVIDIA: shortfall 306590.95 nidle 0.00 saturated

[arkayn]

Thanks. I also noticed that Boinc does not use all GPU by default. When I had GTX560ti in both slots they worked. But the following setup does not use all GPU by default. Will the <use all GPU> option in cc_config.xml work?

Yes, that is what you need.

[Claggy]

Thanks. I also noticed that Boinc does not use all GPU by default. When I had GTX560ti in both slots they worked. But the following setup does not use all GPU by default. Will the <use all GPU> option in cc_config.xml work?

27/09/2012 20:00:46 |  | Starting BOINC client version 7.0.28 for windows_x86_64
27/09/2012 20:00:46 |  | log flags: file_xfer, sched_ops, task, cpu_sched, dcf_debug, sched_op_debug
27/09/2012 20:00:46 |  | log flags: work_fetch_debug
27/09/2012 20:00:46 |  | Libraries: libcurl/7.25.0 OpenSSL/1.0.1 zlib/1.2.6
27/09/2012 20:00:46 |  | Data directory: C:\ProgramData\BOINC
27/09/2012 20:00:46 |  | Running under account Mark
27/09/2012 20:00:46 |  | Processor: 4 GenuineIntel Intel(R) Core(TM)2 Quad CPU    Q6600  @ 2.40GHz [Family 6 Model 15 Stepping 11]
27/09/2012 20:00:46 |  | Processor: 4.00 MB cache
27/09/2012 20:00:46 |  | 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 syscall nx lm vmx tm2 pbe
27/09/2012 20:00:46 |  | OS: Microsoft Windows Vista: Home Premium x64 Edition, Service Pack 2, (06.00.6002.00)
27/09/2012 20:00:46 |  | Memory: 8.00 GB physical, 7.75 GB virtual
27/09/2012 20:00:46 |  | Disk: 74.53 GB total, 23.36 GB free
27/09/2012 20:00:46 |  | Local time is UTC +3 hours
27/09/2012 20:00:46 |  | NVIDIA GPU 0: GeForce GTX 680 (driver version 306.23, CUDA version 5.0, compute capability 3.0, 2048MB, 8382318MB available, 3252 GFLOPS peak)
27/09/2012 20:00:46 |  | NVIDIA GPU 1 (not used): GeForce GTX 670 (driver version 306.23, CUDA version 5.0, compute capability 3.0, 2048MB, 1961MB available, 2915 GFLOPS peak)
27/09/2012 20:00:46 |  | OpenCL: NVIDIA GPU 0: GeForce GTX 680 (driver version 306.23, device version OpenCL 1.1 CUDA, 2048MB, 8382318MB available)
27/09/2012 20:00:46 |  | OpenCL: NVIDIA GPU 1 (not used): GeForce GTX 670 (driver version 306.23, device version OpenCL 1.1 CUDA, 2048MB, 1961MB available)

Boinc 7.0.28 has a Wacky GPU memory reporting Bug, upgrade to Boinc 7.0.36 for a partial fix, it also has an increased internal GPU flops value (by times 10), which means existing GPU work will often get a maximum time exceeded error, so eithier empty GPU cache before hand,
or use Boinc Rescheduler to protect against -177 errors,
(Boinc uses the compute version, memory and GFLOPS peak values (amounst others) to determine if the GPUs are the same)

Claggy

[Caprio]

Boinc 7.0.28 has a Wacky GPU memory reporting Bug, upgrade to Boinc 7.0.36 for a partial fix, it also has an increased internal GPU flops value (by times 10), which means existing GPU work will often get a maximum time exceeded error, so eithier empty GPU cache before hand,
or use Boinc Rescheduler to protect against -177 errors,
(Boinc uses the compute version, memory and GFLOPS peak values (amounst others) to determine if the GPUs are the same)

Claggy

Hahaa.. I think I prefer the memory reporting Bug over ghe maximum time exceeded error bug. Thanks for informing anyway.

Reading wiki entry for the use all CPu explains why less capable GPU's are not used by default.

Nice in a way if you have multiple heavy cruncher GPU and only one less capable CPU for driving the display. A custom profile of how many Cores on CPU and which GPU you could "Snooze" would be nice feature.

[Claggy]

Hahaa.. I think I prefer the memory reporting Bug over ghe maximum time exceeded error bug. Thanks for informing anyway.

It's only with existing GPU Wu's, New Wu's get revised rsc values based on the new higher flops value,

Claggy