Open beta for OpenCL AstroPulse for Intel GPUs started

Message boards : Number crunching : Open beta for OpenCL AstroPulse for Intel GPUs started
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · 6 · Next

AuthorMessage
Profile Dr Grey

Send message
Joined: 27 May 99
Posts: 154
Credit: 104,147,344
RAC: 21
United Kingdom
Message 1364750 - Posted: 5 May 2013, 11:48:10 UTC

Just to correlate with Holmis' stalling issue, I'm getting the same behaviour too. Closing Chrome stops the calculation and the iGPU downclocks. Snoozing and unsnoozing GPU restarts the work again.
ID: 1364750 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1364989 - Posted: 5 May 2013, 23:23:56 UTC - in response to Message 1364750.  

Just to correlate with Holmis' stalling issue, I'm getting the same behaviour too. Closing Chrome stops the calculation and the iGPU downclocks. Snoozing and unsnoozing GPU restarts the work again.

Please do report on Intel OpenCL forum about this issue. Maybe they finally could fix this in new driver.

SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1364989 · Report as offensive
Harri Liljeroos
Avatar

Send message
Joined: 29 May 99
Posts: 3988
Credit: 85,281,665
RAC: 126
Finland
Message 1369914 - Posted: 20 May 2013, 6:59:18 UTC

Hi,
I have set up my Intel HDGraphics 4000 for OpenCL and installed Raistmer's application. I noticed that I needed to plug in my monitor to that card to make Boinc or GPU-Z to recognize its OpenCL capability, installing the driver and SDK were not enough. Normally the monitor is hooked to Nvidia GTX 650 Ti.

Now Boinc is requesting work for intel_gpu but hasn't received any, the response is "no work avaliable for Seti Enhanced or AstroPulse 5.05 or v6". For CPU and CUDA I do get AstroPulse WUs and they run OK. This is in Windows 7 64 bit with Boinc 7.0.64.

Here's the host http://setiathome.berkeley.edu/show_host_detail.php?hostid=6969332
ID: 1369914 · Report as offensive
Profile Wiggo
Avatar

Send message
Joined: 24 Jan 00
Posts: 34744
Credit: 261,360,520
RAC: 489
Australia
Message 1369918 - Posted: 20 May 2013, 7:24:45 UTC - in response to Message 1364750.  

Just to correlate with Holmis' stalling issue, I'm getting the same behaviour too. Closing Chrome stops the calculation and the iGPU downclocks. Snoozing and unsnoozing GPU restarts the work again.

Have you disabled hardware acceleration on your browser?

Cheers.
ID: 1369918 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34253
Credit: 79,922,639
RAC: 80
Germany
Message 1369919 - Posted: 20 May 2013, 7:28:04 UTC

Since intel GPU has no plan class you might get some marked as CPU.
Just for Info.



With each crime and every kindness we birth our future.
ID: 1369919 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3720
Credit: 9,385,827
RAC: 0
Bulgaria
Message 1369943 - Posted: 20 May 2013, 9:09:59 UTC - in response to Message 1369919.  

Since intel GPU has no plan class you might get some marked as CPU.
Just for Info.

But in BOINC Manager I think they will be shown as:
astropulse_v6 6.04 (opencl_intel_100)


I decide this by looking what is in AP6_win_x86_SSE2_OpenCL_Intel.aistub

<app_name>astropulse_v6</app_name>
<version_num>604</version_num>
<plan_class>opencl_intel_100</plan_class>


 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 
ID: 1369943 · Report as offensive
Profile SongBird
Volunteer tester

Send message
Joined: 23 Oct 01
Posts: 104
Credit: 164,826,157
RAC: 297
Bulgaria
Message 1370601 - Posted: 22 May 2013, 15:21:53 UTC
Last modified: 22 May 2013, 15:39:51 UTC

I just added this to my app_info.xml
<app>
<name>astropulse_v6</name>
</app>
<file_info>
<name>AP6_win_x86_SSE2_OpenCL_Intel_r1764.exe</name>
<executable/>
</file_info>
<file_info>
<name>libfftw3f-3.dll</name>
<executable/>
</file_info>
<file_info>
<name>ap_cmdline_win_x86_SSE2_OpenCL_Intel.txt</name>
</file_info>
<app_version>
<app_name>astropulse_v6</app_name>
<version_num>604</version_num>
<platform>windows_intelx86</platform>

<avg_ncpus>0.04</avg_ncpus>
<max_ncpus>0.2</max_ncpus>
<plan_class>opencl_intel_100</plan_class>
<cmdline></cmdline>
<coproc>
<type>intel_gpu</type>
<count>1</count>
</coproc>
<file_ref>
<file_name>AP6_win_x86_SSE2_OpenCL_Intel_r1764.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>libfftw3f-3.dll</file_name>
</file_ref>
<file_ref>
<file_name>ap_cmdline_win_x86_SSE2_OpenCL_Intel.txt</file_name>
<open_name>ap_cmdline.txt</open_name>
</file_ref>
</app_version>
<app_version>
<app_name>astropulse_v6</app_name>
<version_num>604</version_num>
<platform>windows_intelx86</platform>
<file_ref>
<file_name>AP6_win_x86_SSE2_OpenCL_Intel_r1764.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>libfftw3f-3.dll</file_name>
</file_ref>
<file_ref>
<file_name>ap_cmdline_win_x86_SSE2_OpenCL_Intel.txt</file_name>
<open_name>ap_cmdline.txt</open_name>
</file_ref>
</app_version>


And started receiving AP tasks.
Next, from event log:
5/22/2013 6:03:18 PM | SETI@home | Starting task ap_02jn10aa_B1_P0_00195_20130522_19949.wu_1 using astropulse_v6 version 604 (opencl_intel_100) in slot 1



Which I gathered to mean that I now crunch on the Intel GPU. This is confirmed by the Tasks tabsheet, which says "Running (0.04 CPUs + 1 intel_gpu GPU)".

The problem is that I have a process AP6_win_x86_SSE2_OpenCL_Intel_r1764.exe which uses a CPU core at 100%. So I'm guessing it works on the CPU, and not on the GPU.

Can someone help me with this? I want to use the GPU only.

Here is my full App_info.xml

[edit]The host is:
GenuineIntel
Intel(R) Core(TM) i7-3612QM CPU @ 2.10GHz [Family 6 Model 58 Stepping 9]
Number of processors 8
Coprocessors NVIDIA GeForce GT 630M (1023MB) driver: 314.07, INTEL Intel(R) HD Graphics 4000 (1624MB)
ID: 1370601 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1370616 - Posted: 22 May 2013, 15:52:10 UTC - in response to Message 1370601.  
Last modified: 22 May 2013, 15:53:09 UTC

it uses CPU for synching.
Write about issue on Intel forums.
And try to run with full loaded CPU - perhaps CPU usage will be much less (but execution time will increase considerably too).

Also, check blanking % of task. High blanking uses CPU (this problem not driver-connected so no sense to write on Intel forums if it's the case).
SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1370616 · Report as offensive
Profile SongBird
Volunteer tester

Send message
Joined: 23 Oct 01
Posts: 104
Credit: 164,826,157
RAC: 297
Bulgaria
Message 1370624 - Posted: 22 May 2013, 16:05:53 UTC - in response to Message 1370616.  

GPU-Z reports over 90% usage, so the GPU is doing something. Furthermore the task is at 25% after ~1hour, with an estimated remaining at 11hours. 25% per hour and 11 hours estimation amount to about 7 hours difference which I take it might be some artifact suggesting that the GPU is indeed working and producing stuff.

As for the blanking, I do not know how to do that :/

Do you think it is worth it spending a CPU core to GPU app or I should just load the CPU core with CPU app and forget about the Intel GPU? I'm having trouble comparing the two...

ID: 1370624 · Report as offensive
Tom*

Send message
Joined: 12 Aug 11
Posts: 127
Credit: 20,769,223
RAC: 9
United States
Message 1370676 - Posted: 22 May 2013, 18:02:55 UTC
Last modified: 22 May 2013, 18:08:18 UTC

I use a rough calc by looking at the runtime vs cputime for my AP tasks.
NVIDIA show almost the same time so I allocate 1 cpu core per AP task
runtime 2,430.97 cputime 2,392.00

My HD7950 only shows 1/3 cputime vs runtime so I allocate .33 CPU per AP
Need to make sure you are looking at almost no blanking when doing this check.
runtime 1,431.28 cputime 479.95

Check the STDERR OUTPUT for

single pulses: 1
repetitive pulses: 3
percent blanked: 2.39


If percent blanked is over 25% use another tasks stderr

PS do this check with nothing else running except one AP no CPU tasks or loading
ID: 1370676 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 34253
Credit: 79,922,639
RAC: 80
Germany
Message 1370726 - Posted: 22 May 2013, 20:56:01 UTC

This is definitely not for general use.
Not everyone can babysit each task.



With each crime and every kindness we birth our future.
ID: 1370726 · Report as offensive
Tom*

Send message
Joined: 12 Aug 11
Posts: 127
Credit: 20,769,223
RAC: 9
United States
Message 1370752 - Posted: 22 May 2013, 22:55:50 UTC
Last modified: 22 May 2013, 23:02:53 UTC

Its a one time set and forget,unless of course the Intel OpenCL GPU varies its usage of the CPU, with same blanking.

No babysitter required
ID: 1370752 · Report as offensive
zombie67 [MM]
Volunteer tester
Avatar

Send message
Joined: 22 Apr 04
Posts: 758
Credit: 27,771,894
RAC: 0
United States
Message 1370754 - Posted: 22 May 2013, 22:58:06 UTC

What is to babysit? Once you get the app_info.xml working*, it is all automatic.

*When will the server code be updated so that we can select the new third GPU type?
Dublin, California
Team: SETI.USA
ID: 1370754 · Report as offensive
Profile SongBird
Volunteer tester

Send message
Joined: 23 Oct 01
Posts: 104
Credit: 164,826,157
RAC: 297
Bulgaria
Message 1370897 - Posted: 23 May 2013, 11:12:45 UTC
Last modified: 23 May 2013, 11:16:07 UTC

Quick report on my new AP OpenCL endeavor. For now I've completed 4 task. One is validated, the other three are Validation inconclusive. I don't actually know what that means, but - you know what they say about three being a trend - I decided to post here anyway. Do you think inconclusive validation might have something to do with the app being beta?

[edit]One CPU core continues to be fully occupied and as far as I can tell there is no blanking involved. Can someone tell me where on the Intel forums should I post about this and how do I define the problem? A link to a relevant thread would be welcome.
ID: 1370897 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1370926 - Posted: 23 May 2013, 14:16:35 UTC - in response to Message 1370897.  
Last modified: 23 May 2013, 14:18:20 UTC

1)Inconclusives:
yes, it's known issue. I see this on own host too. Precision of Intel OpenCL AP less than for Ati and NV AP. Sometimes such inconclusive finishes as invalid result (yes, even so). What causes such loss of precision not identified still. I hope with more mature OpenCL driver this will go away w/o additional efforts, but will see.

2)Increased CPU usage:
I would say this forum is most appropriate place: http://software.intel.com/en-us/forums/intel-opencl-sdk unless you have direct support contract with Intel. Then fill report to your Intel's representative.
Describe issue as you see it, but keep in mind that they know about own driver and hardware but dont' know about what is SETI (in general) and what is AstroPulse and so on.
What about experiment I proposed earlier?

I repeat conditions one more time:
find almost zero blanking task (2 such tasks). Allow full CPU load (by BOINC) while first of task runs through completion.
Completely disable (or at least free one full core) CPU work while processing next Intel GPU AP task. Then compare runtimes and CPU times of these 2 similar tasks running in such different conditions.
What I expect you will see:
on fully loaded CPU elapsed time will be bigger (maybe much bigger), but CPU time will be considerably lower than for task executed on idle CPU.

If so, describe this situation in intel forum (ideal driver should not use spin-waiting loop for synching that consumes too much CPU).
If you don't see described outcome post here what result you will get.
SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1370926 · Report as offensive
Harri Liljeroos
Avatar

Send message
Joined: 29 May 99
Posts: 3988
Credit: 85,281,665
RAC: 126
Finland
Message 1371016 - Posted: 23 May 2013, 17:53:42 UTC

I briefly looked at SongBird's app_info and noticed that you don't have AstroPulse application for your Cuda card. Your app_info seems to be similar than mine for the AstroPulse Intel OpenCL (mine is from aistub file) except for the last <app_version> section. I did add that part to my app_info.xml to see if that makes any difference. Well, it didn't. Unfortunately I can not get any AstroPulse WUs for the Intel HDGraphics. Maybe this is because I have the Cuda application for AstroPulse running as well? This all I get as a response for scheduler request:

SETI@home	23.5.2013 20:17:22	Sending scheduler request: To fetch work.	
SETI@home	23.5.2013 20:17:22	Requesting new tasks for intel_gpu	
SETI@home	23.5.2013 20:17:26	Scheduler request completed: got 0 new tasks	
SETI@home	23.5.2013 20:17:26	No tasks sent	
SETI@home	23.5.2013 20:17:26	No tasks are available for SETI@home Enhanced	
SETI@home	23.5.2013 20:17:26	No tasks are available for Astropulse v505	
SETI@home	23.5.2013 20:17:26	No tasks are available for AstroPulse v6	
SETI@home	23.5.2013 20:17:26	Tasks for AMD/ATI GPU are available, but your preferences are set to not accept them	
SETI@home	23.5.2013 20:17:26	Project has no tasks available	


ID: 1371016 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1371084 - Posted: 23 May 2013, 22:03:48 UTC - in response to Message 1371016.  

try to decieve BOINC telling it that Intel GPU AP is CPU app.
SETI BOINC server has no intel GPU app so we need same trick as when there was no ATi support from BOINC..

SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1371084 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1371090 - Posted: 23 May 2013, 22:17:36 UTC

And maybe this (taken from BOINC dev mail list) can be relevant to this issue too:

There was a bug in the scheduler.
I checked in a possible fix.
-- David

On 23-May-2013 7:50 AM, Jon Sonntag wrote:
> I'm having trouble with OpenCL Intel GPU work requests. The problem is no
> work ever gets sent.
> If the Intel GPU is requesting 108000 seconds of work would the
> work_req_seconds show as 0? Is that a bug or is the work_req_seconds now
> deprecated? The log on the server shows:
>
>
> 013-05-23 09:34:21.4934 [PID=27500] [send] CPU: req 0.00 sec, 0.00
> instances; est delay 0.00
> 013-05-23 09:34:21.4934 [PID=27500] [send] Intel GPU: req 108000.00 sec,
> 1.00 instances; est delay 0.00
> 013-05-23 09:34:21.4934 [PID=27500] [send] work_req_seconds: 0.00 secs
> 013-05-23 09:34:21.4934 [PID=27500] [send] available disk 95.47 GB,
> work_buf_min 86400
> 013-05-23 09:34:21.4934 [PID=27500] [send] active_frac 0.999834 on_frac
> 0.999692
> 013-05-23 09:34:21.4934 [PID=27500] [send] CPU features: fpu vme de pse
> tsc msr pae mce cx8 apic sep mtrr pge mca c$
> 013-05-23 09:34:21.4941 [PID=27500] Sending reply to [HOST#123978]: 0
> results, delay req 181.80
> 013-05-23 09:34:21.4945 [PID=27500] Scheduler ran 0.005 seconds
>
>
> Jon Sonntag

SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1371090 · Report as offensive
Claggy
Volunteer tester

Send message
Joined: 5 Jul 99
Posts: 4654
Credit: 47,537,079
RAC: 4
United Kingdom
Message 1371091 - Posted: 23 May 2013, 22:21:45 UTC - in response to Message 1371084.  

try to decieve BOINC telling it that Intel GPU AP is CPU app.
SETI BOINC server has no intel GPU app so we need same trick as when there was no ATi support from BOINC..

David might have fixed the scheduler Bug today, Thanks Jon Sonntag for asking again, it didn't get fixed last month when i asked.

Changeset 64ccb6a in boinc-v2

Claggy
ID: 1371091 · Report as offensive
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1371093 - Posted: 23 May 2013, 22:24:31 UTC - in response to Message 1371091.  

Yes, I just posted citation earlier :)

SETI apps news
We're not gonna fight them. We're gonna transcend them.
ID: 1371093 · Report as offensive
Previous · 1 · 2 · 3 · 4 · 5 · 6 · Next

Message boards : Number crunching : Open beta for OpenCL AstroPulse for Intel GPUs started


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.