ABP2 CUDA applications

Gundolf Jahn
Gundolf Jahn
Joined: 1 Mar 05
Posts: 1,079
Credit: 341,280
RAC: 0

RE: Yes, after I had

Message 96344 in response to message 96343

Quote:
Yes, after I had changed my Einstein@home preferences to disallow use of the GPU and reset the project, it's still downloading CUDA units and using the GPU.


Could you post a link to such a CUDA task from your "Tasks for computer" list? That way we could check the server logs.

The one ABP2 task I checked was for CPU only.

Gruß,
Gundolf

Computer sind nicht alles im Leben. (Kleiner Scherz)

Cupojoe
Cupojoe
Joined: 24 Feb 05
Posts: 23
Credit: 33,068,058
RAC: 0

RE: Could you post a link

Message 96345 in response to message 96344

Quote:

Could you post a link to such a CUDA task from your "Tasks for computer" list? That way we could check the server logs.

The one ABP2 task I checked was for CPU only.

Here's one: [link]http://einsteinathome.org/task/157732021[/link]

This unit was downloaded about 24 hours after I changed my preferences and reset the project.

Thanks,

joe

Gundolf Jahn
Gundolf Jahn
Joined: 1 Mar 05
Posts: 1,079
Credit: 341,280
RAC: 0

Okay, the "don't use GPU"

Message 96346 in response to message 96345

Okay, the "don't use GPU" flag doesn't seem to work for resent (lost) results.

The task you linked to (p2030_54106_65364_0125_G67.88-02.05.C_3.dm_433_0) was one of the following resent batch:
2010-01-29 13:48:15.8193 [PID=19173] [debug] [HOST#2018446] MSG(high) Resent lost result h1_1140.40_S5R4__1409_S5R6a_0
2010-01-29 13:48:15.8193 [PID=19173] [debug] [HOST#2018446] MSG(high) Resent lost result h1_1140.40_S5R4__1408_S5R6a_0
2010-01-29 13:48:15.8193 [PID=19173] [debug] [HOST#2018446] MSG(high) Resent lost result h1_1140.40_S5R4__1407_S5R6a_0
2010-01-29 13:48:15.8193 [PID=19173] [debug] [HOST#2018446] MSG(high) Resent lost result h1_1140.40_S5R4__1406_S5R6a_0
2010-01-29 13:48:15.8193 [PID=19173] [debug] [HOST#2018446] MSG(high) Resent lost result h1_1140.40_S5R4__1405_S5R6a_0
2010-01-29 13:48:15.8194 [PID=19173] [debug] [HOST#2018446] MSG(high) Resent lost result h1_1140.40_S5R4__1404_S5R6a_0
2010-01-29 13:48:15.8194 [PID=19173] [debug] [HOST#2018446] MSG(high) Resent lost result p2030_54113_60211_0155_G49.93-00.02.C_1.dm_309_2
2010-01-29 13:48:15.8194 [PID=19173] [debug] [HOST#2018446] MSG(high) Resent lost result p2030_54106_65669_0128_G69.02-02.38.C_3.dm_32_1
2010-01-29 13:48:15.8194 [PID=19173] [debug] [HOST#2018446] MSG(high) Resent lost result p2030_54106_65364_0125_G67.88-02.05.C_3.dm_414_1
2010-01-29 13:48:15.8194 [PID=19173] [debug] [HOST#2018446] MSG(high) Resent lost result p2030_54106_65364_0125_G67.88-02.05.C_3.dm_427_1
2010-01-29 13:48:15.8194 [PID=19173] [debug] [HOST#2018446] MSG(high) Resent lost result p2030_54106_65364_0125_G67.88-02.05.C_3.dm_433_0
2010-01-29 13:48:15.8194 [PID=19173] [debug] [HOST#2018446] MSG(high) Resent lost result p2030_54106_65669_0128_G69.02-02.38.C_3.dm_52_1

Gruß,
Gundolf

Computer sind nicht alles im Leben. (Kleiner Scherz)

cristipurdel
cristipurdel
Joined: 19 Jul 07
Posts: 26
Credit: 11,991,887
RAC: 0

Any news on an updated ABP2

Any news on an updated ABP2 CUDA version with better GPU utilization, or on the ATI version?

Bernd Machenschalk
Bernd Machenschalk
Moderator
Administrator
Joined: 15 Oct 04
Posts: 4,027
Credit: 215,786,926
RAC: 79,539

RE: Any news on an updated

Message 96348 in response to message 96347

Quote:
Any news on an updated ABP2 CUDA version with better GPU utilization, or on the ATI version?

A better CUDA ABP2 App is being worked on, but we still have trouble with numerical differences between GPU and CPU calculations. Oliver will be on well deserved vacation for the next weeks, and I'll focus on preparing the next GW search, so there won't be much progress on CUDA in February.

I doubt that we'll ever make a native ATI App, we may at some point have OpenCL Apps that will run on ATis, too.

BM

BM

hoarfrost
hoarfrost
Joined: 9 Feb 05
Posts: 207
Credit: 53,487,228
RAC: 5,269

RE: I doubt that we'll ever

Message 96349 in response to message 96348

Quote:
I doubt that we'll ever make a native ATI App, we may at some point have OpenCL Apps that will run on ATis, too.


I am very disappointed... ATI cards cheaper and have a fantastic computation power.

Ver Greeneyes
Ver Greeneyes
Joined: 26 Mar 09
Posts: 140
Credit: 9,562,235
RAC: 0

RE: I am very

Message 96350 in response to message 96349

Quote:
I am very disappointed... ATI cards cheaper and have a fantastic computation power.


IIRC, ATI's poor support for CAL is largely to blame for this. Milkyway@Home only got it working through the dedication of one of their volunteers. However, OpenCL is vendor-agnostic, and should hopefully be fast on any graphics hardware - as long as the drivers are implemented right.

Duhem
Duhem
Joined: 2 May 07
Posts: 4
Credit: 2,845,511
RAC: 1,503

RE: Currently it looks

Quote:

Currently it looks like we'll finish the CUDA kernel for ABP2, the port this to an OpenCL version that would support both GPUs (and then some computing devices).
BM

Just registering my note of encouragement for ATI support. Currently the only
project I'm able to run on my ATI Radeon HD 4350 is the Collatz conjecture and
well; number theory, shnumber theory: I'd rather crunch on LIGO data. I don't
know how well the credits scale, but one Collatz process on my ATI is swamping
the collective rate of my (generally) 4 Einstein@home processes running on a
4 core Athlon II 2.8GHz. (I think Collatz work validates faster: I typically
have a lot of pending einstein credits. I suspect e@h may catch up)

Regards,
mjm

update: OpenCL is fine with me (probably a more efficient use of programming
effort) as long as ATI eventually gets covered. Btw, Milkyway@home tells my my
4350 board doesn't support double precision so I can't get their work for my
gpu. Why must there always be a problem?/)
doesn't support double precision

Fred J. Verster
Fred J. Verster
Joined: 27 Apr 08
Posts: 118
Credit: 22,451,438
RAC: 0

RE: RE: Currently it

Message 96352 in response to message 96351

Quote:
Quote:

Currently it looks like we'll finish the CUDA kernel for ABP2, the port this to an OpenCL version that would support both GPUs (and then some computing devices).
BM

Just registering my note of encouragement for ATI support. Currently the only
project I'm able to run on my ATI Radeon HD 4350 is the Collatz conjecture and
well; number theory, shnumber theory: I'd rather crunch on LIGO data. I don't
know how well the credits scale, but one Collatz process on my ATI is swamping
the collective rate of my (generally) 4 Einstein@home processes running on a
4 core Athlon II 2.8GHz. (I think Collatz work validates faster: I typically
have a lot of pending einstein credits. I suspect e@h may catch up)

Regards,
mjm

update: OpenCL is fine with me (probably a more efficient use of programming
effort) as long as ATI eventually gets covered. Btw, Milkyway@home tells my my
4350 board doesn't support double precision so I can't get their work for my
gpu. Why must there always be a problem?/)
doesn't support double precision

Hi, read this interresting, but long thread, have a problem, at the moment only on my Laptop, an HP Pavillion 8180nl (T2400CPU)2GB DDR2;WIN XP86.
Here's a piece of the BOINC log file:
8-2-2010 20:37:03 Reading preferences override file
8-2-2010 20:37:03 Preferences:
8-2-2010 20:37:03 max memory usage when active: 1023.02MB
8-2-2010 20:37:03 max memory usage when idle: 1841.44MB
8-2-2010 20:37:03 max disk usage: 16.91GB
8-2-2010 20:37:03 don't use GPU while active
8-2-2010 20:37:03 max download rate: 800000 bytes/sec
8-2-2010 20:37:03 max upload rate: 59996 bytes/sec
8-2-2010 20:37:03 (to change, visit the web site of an attached project,
8-2-2010 20:37:03 or click on Preferences)
8-2-2010 20:37:05 Docking [error] 1ebw1ajv_mod0014crossdockinghiv1_166_409078_0: negative FLOPs left -31998662328805.613000
8-2-2010 20:37:06 Docking [error] 1ebw1ajv_mod0014crossdockinghiv1_166_409078_0: negative FLOPs left -31998662328805.613000
8-2-2010 20:37:09 Docking [error] 1ebw1ajv_mod0014crossdockinghiv1_166_409078_0: negative FLOPs left -31998662328805.613000
8-2-2010 20:37:20 Docking task 1dif1ajv_mod0014crossdockinghiv1_1203_66232_0 suspended by user
8-2-2010 20:37:58 Docking [error] 1ebw1ajv_mod0014crossdockinghiv1_166_409078_0: negative FLOPs left -31998662328805.613000
8-2-2010 20:38:08 Docking [error] 1ebw1ajv_mod0014crossdockinghiv1_166_409078_0: negative FLOPs left -31998662328805.613000

Have stopt and restarted BOINC, REBOOTED. Nothing, anyone an idea?

And another thing is the (terrible)low use of CUDA, I noticed on a QX9650 (@3250MHz)+GTX9800+ , times to complete a ABP2 WU aren't that different. Also depends on CPU&GPU combination, the above one has little difference for Einstein, that is.
I've seen, this is already talked over, but no real steps to greatly enhance the GPU output, aren't planned(?).

Fred J. Verster
Fred J. Verster
Joined: 27 Apr 08
Posts: 118
Credit: 22,451,438
RAC: 0

RE: RE: RE: Currently

Message 96353 in response to message 96352

Quote:
Quote:
Quote:

Currently it looks like we'll finish the CUDA kernel for ABP2, the port this to an OpenCL version that would support both GPUs (and then some computing devices).
BM

Just registering my note of encouragement for ATI support. Currently the only
project I'm able to run on my ATI Radeon HD 4350 is the Collatz conjecture and
well; number theory, shnumber theory: I'd rather crunch on LIGO data. I don't
know how well the credits scale, but one Collatz process on my ATI is swamping
the collective rate of my (generally) 4 Einstein@home processes running on a
4 core Athlon II 2.8GHz. (I think Collatz work validates faster: I typically
have a lot of pending einstein credits. I suspect e@h may catch up)

Regards,
mjm

update: OpenCL is fine with me (probably a more efficient use of programming
effort) as long as ATI eventually gets covered. Btw, Milkyway@home tells my my
4350 board doesn't support double precision so I can't get their work for my
gpu. Why must there always be a problem?/)
doesn't support double precision

Hi, read this interresting, but long thread,
And another thing is the (terrible)low use of CUDA, I noticed on a QX9650 (@3250MHz)+GTX9800+ , times to complete a ABP2 WU aren't that different. Also depends on CPU&GPU combination, the above one has little difference for Einstein, that is.
I've seen, this is already talked over, but no real steps to greatly enhance the GPU output, aren't planned(?).

Sorry, for me mixing up Docking & Einstein .

Comment viewing options

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