I already have that in the root of my boinc directory to enable seti to use all 3 gpu's. do I need to add it to the root of the Einstein directory to enable multiple tasks on 1 gpu?
Hi, recently I 'rebuild' my WIN XP64; QX9650+GTX470+GTX480, host. (Btw, BOINC reports 2 GTX480's, looks like it takes the 'info'from the first (0) Device, though.
Had to install a bigger, 850Watt PSU, to get both cards running.
But the GPU-Load and Mem-Cntrl-Load, appear to be very low.
SETI MB tasks run with a higher load, but still possible to run 2-3 or even 4 MB tasks, whithout failiars.
Read through the last messages in this forum, this should also be possible with
Einstein CUDA tasks.
And if it's accepted, why not go for it, most important thing is, this should not lead to all kinds of errors.
Since the GPU-load, is low ~3 - 8% on the 480 and ~15%, for the 470, gonna change this to 1CPU+0.5GPU or even lower. It only improves efficiency and throughput so why not change that?
Also do it with MilkyWay and Collatz C. and SETI Bêta on ATI's HD5770 &
EAH4850 (Double-Precision) and works perfect.
I already have that in the root of my boinc directory to enable seti to use all 3 gpu's. do I need to add it to the root of the Einstein directory to enable multiple tasks on 1 gpu?
No, cc_config.xml belongs in the BOINC data directory and has nothing to do with running multiple CUDA tasks on one GPU.
Do you see any messages in the Messages tab after a restart that show the file being used? They usually start with "Config:"
Gruß,
Gundolf
[edit]And since you have three GPUs, why should it be an error to run six CUDA tasks when setting 0.5?[/edit]
Computer sind nicht alles im Leben. (Kleiner Scherz)
Sorry, that was what I meant.
I have 3 gpu's crunching 3 tasks per card in seti..
I have obviously not got the balance right in the app_info I posted as I am currently running 6 tasks concurrently, when it should be 2 tasks per gpu..
It does indeed, though looking at gpu usage, Im seeing the same 6-9% usage that I was when I ran 1 task per gpu, also completion times are suggesting that there is timeslicing going on rather than running tasks in parallel..
OK, Im very interested in
)
OK, Im very interested in improving gpu efficiency - I have a quad core cpu and 3 x fermi cards, (2 x 470 and 1 x 480)
I have created an app_info changing the and the values
einsteinbinary_ABP2
Arecibo Binary Pulsar Search (STSP)
einsteinbinary_ABP2_3.11_windows_intelx86__ABP2cuda23.exe
cudart32_23.dll
cufft32_23.dll
einsteinbinary_ABP2_3.03_graphics_windows_intelx86.exe
einsteinbinary_ABP2
311
windows_intelx86
0.5
0.5
ABP2cuda23
einsteinbinary_ABP2_3.11_windows_intelx86__ABP2cuda23.exe
cudart32_23.dll
cudart.dll
cufft32_23.dll
cufft.dll
einsteinbinary_ABP2_3.03_graphics_windows_intelx86.exe
graphics_app
CUDA
0.5
I must have done something wrong, as Im still seeing 6-9% gpu usage?
Hi! How many CUDA apps and
)
Hi!
How many CUDA apps and CPU apps do you see exectuting in parallel? A screenshot of the tasks list in BOINC manager might shed some light on this.
HB
I changed to 1 I
)
I changed to 1
I definately got something wrong as I only have 4 cpu's and Im only seeing sub 10% usage on the gpu's
Try this
)
Try this page:
http://boincfaq.mundayweb.com/index.php?language=1&view=471
and search cc_config section. You have to add new file cc_config.xml
I have tried it - BOINC starts to use both cards (GTX 460 & GT 240).
But GT 240 starts to generate errors.
Otherwise - it's normal to use 9-10% of GPU in Einstein@home project. Some changes, i.e. 2 CPU to 1 GPU association might help.
I already have that in the
)
I already have that in the root of my boinc directory to enable seti to use all 3 gpu's. do I need to add it to the root of the Einstein directory to enable multiple tasks on 1 gpu?
Hi, recently I 'rebuild' my
)
Hi, recently I 'rebuild' my WIN XP64; QX9650+GTX470+GTX480, host. (Btw, BOINC reports 2 GTX480's, looks like it takes the 'info'from the first (0) Device, though.
Had to install a bigger, 850Watt PSU, to get both cards running.
But the GPU-Load and Mem-Cntrl-Load, appear to be very low.
SETI MB tasks run with a higher load, but still possible to run 2-3 or even 4 MB tasks, whithout failiars.
Read through the last messages in this forum, this should also be possible with
Einstein CUDA tasks.
And if it's accepted, why not go for it, most important thing is, this should not lead to all kinds of errors.
Since the GPU-load, is low ~3 - 8% on the 480 and ~15%, for the 470, gonna change this to 1CPU+0.5GPU or even lower. It only improves efficiency and throughput so why not change that?
Also do it with MilkyWay and Collatz C. and SETI Bêta on ATI's HD5770 &
EAH4850 (Double-Precision) and works perfect.
RE: I already have that in
)
No, cc_config.xml belongs in the BOINC data directory and has nothing to do with running multiple CUDA tasks on one GPU.
Do you see any messages in the Messages tab after a restart that show the file being used? They usually start with "Config:"
Gruß,
Gundolf
[edit]And since you have three GPUs, why should it be an error to run six CUDA tasks when setting 0.5?[/edit]
Computer sind nicht alles im Leben. (Kleiner Scherz)
Sorry, that was what I
)
Sorry, that was what I meant.
I have 3 gpu's crunching 3 tasks per card in seti..
I have obviously not got the balance right in the app_info I posted as I am currently running 6 tasks concurrently, when it should be 2 tasks per gpu..
RE: I have obviously not
)
But 2 tasks per GPU with 3 GPUs does sum up to 6 tasks!
Gruß,
Gundolf
Computer sind nicht alles im Leben. (Kleiner Scherz)
It does indeed, though
)
It does indeed, though looking at gpu usage, Im seeing the same 6-9% usage that I was when I ran 1 task per gpu, also completion times are suggesting that there is timeslicing going on rather than running tasks in parallel..