New BRP CUDA Apps 1.07 / 1.08

Vadim Gusev
Vadim Gusev
Joined: 19 Nov 07
Posts: 6
Credit: 8,551,818
RAC: 0

RE: If running Linux and

Quote:

If running Linux and you have the beta driver 270 installed, you should get an App version 1.08 plan class "BRP3cuda32nv270".

If running Linux and not have installed the 270 driver, you should get an App version 1.08 plan class "BRP3cuda32fullCPU".


I've got 270.18 drivers, however I get fullCPU Apps.
When BOINC starts I get that nvidia card related string:
Вт. 22 февр. 2011 01:09:40 NVIDIA GPU 0: GeForce GTX 460 (driver version unknown, CUDA version 4000, compute capability 2.1, 1023MB, 717 GFLOPS peak)
Seems like BOINC cant recognize driver's version properly.

M. Schmitt
M. Schmitt
Joined: 27 Jun 05
Posts: 478
Credit: 15,872,262
RAC: 0

Same here. nvidia-smi

Same here.

nvidia-smi -q

==============NVSMI LOG==============

Timestamp : Tue Feb 22 05:06:14 2011

Driver Version : 270.18
[...]

But when BOINC starts the log says:

Tue 22 Feb 2011 04:47:56 AM CET NVIDIA GPU 0: GeForce GTX 460 (driver version unknown, CUDA version 4000, compute capability 2.1, 1024MB, 681 GFLOPS peak)

astrocrab
astrocrab
Joined: 28 Jan 08
Posts: 208
Credit: 429,202,534
RAC: 0

same here. have to download

same here. have to download 270 app manually

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 3,956
Credit: 203,605,825
RAC: 29,841

That's bad. Anyone knows

That's bad. Anyone knows what's wrong with BOINC there, or how the project can work around that?

BM

BM

M. Schmitt
M. Schmitt
Joined: 27 Jun 05
Posts: 478
Credit: 15,872,262
RAC: 0

Maybe BOINC is using the old

Maybe BOINC is using the old query from bikemans post: http://einsteinathome.org/node/195607&nowrap=true#110130

'nvidia-smi -a' is no valid query any more. Instead 'nvidia-smi -q' is used. If this is true, a small script would do the job. Maybe even a symlink can be used. Can't try it atm.
But someone should ask the devs first or look in the source.

Jord
Joined: 26 Jan 05
Posts: 2,952
Credit: 5,710,412
RAC: 335

According to the changes,

According to the changes, there's:

Quote:

* Added support for the following GPUs -- GeForce GTX 560 Ti
* Added NV-CONTROL event notification for NV_CTRL_FRAMELOCK_SYNC_READY status changes.
* Added a new X configuration option "Interactive", which defaults to enabled, but can be disabled to allow long-running GPU compute programs to run concurrently with X.
* Fixed a bug in the VDPAU presentation queue that could cause VDPAU "display preemption" when rendering to tiny or zero-sized windows or pixmaps.
* Fixed a bug in VDPAU which prevented use of the overlay presentation queue following an application exiting without gracefully destroying its VDPAU presentation queue.


So, has anyone tried to disable the "Interactive" X config option yet and then restart BOINC?

Vadim Gusev
Vadim Gusev
Joined: 19 Nov 07
Posts: 6
Credit: 8,551,818
RAC: 0

RE: So, has anyone tried to

Quote:
So, has anyone tried to disable the "Interactive" X config option yet and then restart BOINC?


What exactly is it supposed to change? I've done listed above and noticed no changes at all in either X responsibility or computing speed.

fruehwf
fruehwf
Joined: 22 Feb 10
Posts: 1
Credit: 2,833,037
RAC: 0

22.02.2011

22.02.2011 16:44:41 Einstein@Home Message from server: No work sent
22.02.2011 16:44:41 Einstein@Home Message from server: Opted out of BRP3 CPU
22.02.2011 16:44:41 Einstein@Home Message from server: Your NVIDIA GPU has insufficient memory (need 300MB)

NVIDIA Quadro NVS 160M (241MB) driver: 26658
:-(

Will there be a change sometimes?

br

franz

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 3,956
Credit: 203,605,825
RAC: 29,841

RE: Will there be a change

Quote:
Will there be a change sometimes?

The memory requirements might be changed (again) at some point.

  • * On our reference machine (Linux x86, Tesla C1060) we found the GPU memory footprint of the BRP3 CUDA App (with PMPS data) to be below 200MB, so initially we set it to that.
    * Then we got some reports / complaints from participants that on their machines the BRP3 App took more than 300MB of GPU memory, leading to the BOINC Client postponing computation over and over again until the tasks finally errored out. So we raised the memory requirement to 300MB
    * The new applications do write the amount of GPU memory occupied before and after memory allocation to stderr. This should give us an idea of how much GPU memory a BRP3 task actually takes on a given machine (probably just depending on the graphics card).
    * After having collected enough data, we'll scan through it and may modify the memory requirements again according to our findings.
    * I doubt, however, that less than 256MB of total GPU memory will ever be enough to get a BRP3 CUDA task.

BM

BM

astrocrab
astrocrab
Joined: 28 Jan 08
Posts: 208
Credit: 429,202,534
RAC: 0

RE: * Added a new X

Quote:
* Added a new X configuration option "Interactive", which defaults to enabled, but can be disabled to allow long-running GPU compute programs to run concurrently with X.

what does this option do?

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.