Using 3 gpu`s
5 May 2015 11:25:30 UTC
Topic 198081
(moderation:
New user, its only running on my titan x tho and not the 780`s , i expect i need to add a setting file ?
05/05/2015 12:03:35 | | CUDA: NVIDIA GPU 0: GeForce GTX TITAN X (driver version 350.12, CUDA version 7.0, compute capability 5.2, 4096MB, 4096MB available, 6611 GFLOPS peak)
05/05/2015 12:03:35 | | CUDA: NVIDIA GPU 1 (not used): GeForce GTX 780 (driver version 350.12, CUDA version 7.0, compute capability 3.5, 3072MB, 2925MB available, 4636 GFLOPS peak)
05/05/2015 12:03:35 | | CUDA: NVIDIA GPU 2 (not used): GeForce GTX 780 (driver version 350.12, CUDA version 7.0, compute capability 3.5, 3072MB, 2925MB available, 4636 GFLOPS peak)
05/05/2015 12:03:35 | | OpenCL: NVIDIA GPU 0: GeForce GTX TITAN X (driver version 350.12, device version OpenCL 1.2 CUDA, 12288MB, 4096MB available, 6611 GFLOPS peak)
05/05/2015 12:03:35 | | OpenCL: NVIDIA GPU 1 (ignored by config): GeForce GTX 780 (driver version 350.12, device version OpenCL 1.2 CUDA, 3072MB, 2925MB available, 4636 GFLOPS peak)
05/05/2015 12:03:35 | | OpenCL: NVIDIA GPU 2 (ignored by config): GeForce GTX 780 (driver version 350.12, device version OpenCL 1.2 CUDA, 3072MB, 2925MB available, 4636 GFLOPS peak)
Using 3 gpu`s
)
Welcome and kudos on the nice hardware :)
Found this old thread on the Boinc forum which might help.
Thanks, it folds well LOL
)
Thanks, it folds well LOL
and sorted, now running all 3
05/05/2015 13:00:11 | | CUDA: NVIDIA GPU 0: GeForce GTX TITAN X (driver version 350.12, CUDA version 7.0, compute capability 5.2, 4096MB, 4096MB available, 6611 GFLOPS peak)
05/05/2015 13:00:11 | | CUDA: NVIDIA GPU 1: GeForce GTX 780 (driver version 350.12, CUDA version 7.0, compute capability 3.5, 3072MB, 2925MB available, 4636 GFLOPS peak)
05/05/2015 13:00:11 | | CUDA: NVIDIA GPU 2: GeForce GTX 780 (driver version 350.12, CUDA version 7.0, compute capability 3.5, 3072MB, 2925MB available, 4636 GFLOPS peak)
05/05/2015 13:00:11 | | OpenCL: NVIDIA GPU 0: GeForce GTX TITAN X (driver version 350.12, device version OpenCL 1.2 CUDA, 12288MB, 4096MB available, 6611 GFLOPS peak)
05/05/2015 13:00:11 | | OpenCL: NVIDIA GPU 1: GeForce GTX 780 (driver version 350.12, device version OpenCL 1.2 CUDA, 3072MB, 2925MB available, 4636 GFLOPS peak)
05/05/2015 13:00:11 | | OpenCL: NVIDIA GPU 2: GeForce GTX 780 (driver version 350.12, device version OpenCL 1.2 CUDA, 3072MB, 2925MB available, 4636 GFLOPS peak)
05/05/2015 13:00:11 | | Host name: Big1
05/05/2015 13:00:11 | | Processor: 16 GenuineIntel Intel(R) Core(TM) i7-5960X CPU @ 3.00GHz [Family 6 Model 63 Stepping 2]
05/05/2015 13:00:11 | | 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 fma cx16 sse4_1 sse4_2 movebe popcnt aes f16c rdrandsyscall nx lm avx avx2 vmx tm2 dca pbe fsgsbase bmi1 smep bmi2
05/05/2015 13:00:11 | | OS: Microsoft Windows 8.1: Core x64 Edition, (06.03.9600.00)
05/05/2015 13:00:11 | | Memory: 15.90 GB physical, 31.90 GB virtual
05/05/2015 13:00:11 | | Disk: 476.42 GB total, 297.98 GB free
Glad it works. One other
)
Glad it works.
One other thing you might want to consider is changing the number of CPU cores you use for non-GPU tasks. E@H GPU tasks work much faster if they get a CPU core to support them.
Your machine reports 16 CPU cores. Reserve at least 3 of them for GPU support.
In the BOINC manager go to menu "Tools - Computing preferences" and change the "On multiprocessor systems, use at most ..." value to something like 75%.
On such powerful hardware you might want to look into running more than one GPU task simultaneously per GPU. To change that setting you go to your E@H preference page on this website: https://einstein.phys.uwm.edu/prefs.php?subset=project. Change "GPU utilization factor of BRP apps" to 0.5 for 2 simultaneous tasks, 0.33 for 3 and so on. Changes done on this page only takes effect after the next download of tasks.
More GPU tasks will of course require more CPU cores to support them. So you might want to go back and change step 1 again :)
Agreed - on these powerful
)
Agreed - on these powerful cards you should definitely run 2 WUs concurrently, maybe even 3. If power consumption, heat and noise become too much simply lower the power target. The cards will take care of the rest and scale back clock speed and voltage to run more energy efficient.
I don't think you need to reserve as much CPU time as Logforme said, though. If you run several WUs concurrently on a GPU, each task matters less. And the new app version 1.52 needs significantly less CPU support than older ones. I suspect an overall CPU load around 80 - 90% (in task manager, not the setting in BOINC) will be fine.
MrS
Scanning for our furry friends since Jan 2002
thanks for the tips, lets see
)
thanks for the tips, lets see how well this can do now
Seems to be fine on 2 wu`s as
)
Seems to be fine on 2 wu`s as my rac shows :) onwards and upwards
RE: Seems to be fine on 2
)
Have you gotten the app 'gpu-z' yet? If not get it and it will tell you the actual load on each gpu in real time, that will help you figure out if you can run more workunits per gpu or not. This is where I get it but you can get it elsewhere too
http://www.guru3d.com/files-details/gpu-z-download-techpowerup.html
Good point, Mikey. As a rule
)
Good point, Mikey. As a rule of thumb: if you're under 95% GPU load, running another concurrent WU should give you a small throughput increase.
MrS
Scanning for our furry friends since Jan 2002