Intel GPU App ready?


log in

Advanced search

Message boards : Number crunching : Intel GPU App ready?

Previous · 1 · 2
Author Message
Richard HaselgroveProject donor
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8458
Credit: 48,598,056
RAC: 79,550
United Kingdom
Message 1471674 - Posted: 1 Feb 2014, 17:53:54 UTC - in response to Message 1471669.

But it repeated again:

2/1/2014 9:32:56 PM | SETI@home | [sched_op] Starting scheduler request
2/1/2014 9:32:56 PM | SETI@home | Sending scheduler request: To fetch work.
2/1/2014 9:32:56 PM | SETI@home | Reporting 43 completed tasks
2/1/2014 9:32:56 PM | SETI@home | Requesting new tasks for CPU and NVIDIA and intel_gpu
2/1/2014 9:32:56 PM | SETI@home | [sched_op] CPU work request: 1284986.07 seconds; 0.00 devices
2/1/2014 9:32:56 PM | SETI@home | [sched_op] NVIDIA work request: 569324.72 seconds; 0.00 devices
2/1/2014 9:32:56 PM | SETI@home | [sched_op] intel_gpu work request: 518400.00 seconds; 1.00 devices
2/1/2014 9:33:00 PM | SETI@home | Scheduler request completed: got 43 new tasks
2/1/2014 9:33:00 PM | SETI@home | [sched_op] Server version 703
2/1/2014 9:33:00 PM | SETI@home | Project requested delay of 303 seconds
2/1/2014 9:33:00 PM | SETI@home | [sched_op] estimated total CPU task duration: 212585 seconds
2/1/2014 9:33:00 PM | SETI@home | [sched_op] estimated total NVIDIA task duration: 110451 seconds
2/1/2014 9:33:00 PM | SETI@home | [sched_op] estimated total intel_gpu task duration: 0 seconds

You swapped out 43 completed CPU and NVidia tasks, and got exactly the same mixture in return.

Did you see my suggestion of temporarily disabling NV in your preferences, so that next time you report NV, you get Intel to top up your GPU quota?

And your cache is still more than you need, to an absurd degree - you're still asking for six days. Even with 100 NVidia tasks cached, your client still wanted another 5 days of work for them to reach your target cache. You have two cards, so your cache request is still 2.5 days too high to fill both device classes. Cut it down to 2 days or less - at least while testing.

Ulrich Metzner
Volunteer tester
Avatar
Send message
Joined: 3 Jul 02
Posts: 976
Credit: 8,402,301
RAC: 8,730
Germany
Message 1471681 - Posted: 1 Feb 2014, 19:12:34 UTC

Here is my part of the app_info.xml, which is working as expected.
Maybe this can help:

<app_version>
<app_name>setiathome_v7</app_name>
<version_num>703</version_num>
<platform>windows_intelx86</platform>
<avg_ncpus>0.04</avg_ncpus>
<max_ncpus>0.2</max_ncpus>
<plan_class>intel_opencl_sah</plan_class>
<cmdline></cmdline>
<coproc>
<type>intel_gpu</type>
<count>1</count>
</coproc>
<file_ref>
<file_name>setiathome_7.03_windows_intelx86__opencl_intel_gpu_sah.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>libfftw3f-3.dll</file_name>
</file_ref>
<file_ref>
<file_name>mb_cmdline-7.03-opencl_intel_gpu_sah.txt</file_name>
<open_name>mb_cmdline.txt</open_name>
</file_ref>
<file_ref>
<file_name>MultiBeam_Kernels_r2061.cl</file_name>
</file_ref>
</app_version>
<app_version>
<app_name>setiathome_v7</app_name>
<version_num>703</version_num>
<platform>windows_intelx86</platform>
<avg_ncpus>0.04</avg_ncpus>
<max_ncpus>0.2</max_ncpus>
<plan_class>opencl_intel_sah</plan_class>
<cmdline></cmdline>
<coproc>
<type>intel_gpu</type>
<count>1</count>
</coproc>
<file_ref>
<file_name>setiathome_7.03_windows_intelx86__opencl_intel_gpu_sah.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>libfftw3f-3.dll</file_name>
</file_ref>
<file_ref>
<file_name>mb_cmdline-7.03-opencl_intel_gpu_sah.txt</file_name>
<open_name>mb_cmdline.txt</open_name>
</file_ref>
<file_ref>
<file_name>MultiBeam_Kernels_r2061.cl</file_name>
</file_ref>
</app_version>


Please keep in mind, i'm using the name convention from seti BETA project, that's why the executable has another name. This was made long before intel was on seti main project. I also see different plan classes?

Hope, this helps, Raistmer, i very much appreciated your work!
____________
Aloha, Uli

Profile Raistmer
Volunteer developer
Volunteer tester
Avatar
Send message
Joined: 16 Jun 01
Posts: 3390
Credit: 46,312,879
RAC: 9,594
Russia
Message 1471683 - Posted: 1 Feb 2014, 19:23:50 UTC - in response to Message 1471681.



Hope, this helps, Raistmer, i very much appreciated your work!


Thanks, finally I got 2 tasks under opencl_intel_gpu_sah.
So it was all GPU-wide limit indeed just as Richard said.
Able to test next app version now.
____________

Profile Rick A. Sponholz
Volunteer tester
Avatar
Send message
Joined: 13 Jun 99
Posts: 43
Credit: 19,741,597
RAC: 7,938
United States
Message 1471719 - Posted: 1 Feb 2014, 21:56:29 UTC - in response to Message 1470671.

Hi there,

I am using the lunatics optimised apps at the moment (mostly due to an incompatibility between the standard nvidia apps and my GTX660).

Just wondering if anyone can suggest an app_info.xml snippet to enable work fetch of Intel GPU work units on an anonymous platform.
I expect I will have to download the application manually too.

I have added <use_all_gpus>1</use_all_gpus> to my cc_config.xml file and the BOINC client shows that it has detected the Intel GPU
OpenCL CPU: Intel(R) Core(TM) i7-3770K CPU @ 3.50GHz (OpenCL driver vendor: Intel(R) Corporation, driver version 3.0.1.10878, device version OpenCL 1.2 (Build 76413))
My project settings allow all GPU types.

I will be interested to see if the project favours one type of GPU for processing or if it keeps work queued for all available GPUs.

Regards,
Daniel


I have the same situation of using lunatic's optimized app for seti. Does anyone know how/what to add to the app_info file to get seti to run on my intel gpu's?
Thanks in advance,
Rick
____________

Profile HAL9000
Volunteer tester
Avatar
Send message
Joined: 11 Sep 99
Posts: 4041
Credit: 110,998,207
RAC: 147,249
United States
Message 1472053 - Posted: 2 Feb 2014, 20:09:39 UTC - in response to Message 1471683.
Last modified: 2 Feb 2014, 20:10:20 UTC

Hope, this helps, Raistmer, i very much appreciated your work!

Thanks, finally I got 2 tasks under opencl_intel_gpu_sah.
So it was all GPU-wide limit indeed just as Richard said.
Able to test next app version now.

Having 0 GPU tasks and adding Intel GPU app to this host:

02-Feb-2014 14:29:03 [SETI@home] Sending scheduler request: To fetch work.
02-Feb-2014 14:29:03 [SETI@home] Requesting new tasks for CPU and intel_gpu
02-Feb-2014 14:29:04 [SETI@home] update requested by user
02-Feb-2014 14:29:07 [SETI@home] Scheduler request completed: got 100 new tasks
02-Feb-2014 14:29:07 [SETI@home] No tasks are available for the applications you have selected
02-Feb-2014 14:29:07 [SETI@home] No tasks are available for AstroPulse v6
02-Feb-2014 14:29:07 [SETI@home] Your preferences allow tasks from applications other than those selected
02-Feb-2014 14:29:07 [SETI@home] Sending tasks from other applications

Scheduler seemed very happy to see me.
____________
SETI@home classic workunits: 93,865 CPU time: 863,447 hours

Join the BP6/VP6 User Group today!

Richard HaselgroveProject donor
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8458
Credit: 48,598,056
RAC: 79,550
United Kingdom
Message 1472093 - Posted: 2 Feb 2014, 22:21:47 UTC - in response to Message 1471719.

Does anyone know how/what to add to the app_info file to get seti to run on my intel gpu's?
Thanks in advance,
Rick

Ah, that's the other thing I was going to do this weekend, wasn't it?

The plan was to wait for AP tasks to be split, and then do both apps together. Well, AP has been split, and - nada.

Checking the front page, the announcement says "We've released SETI@home 7.03 for Intel GPUs that support OpenCL 1.2". Yup, that's MB only - no AP release. I take it as axiomatic that Eric knows what he's doing, so there will be a reason why not: possibly validation rate problems in Beta, or simply too many volunteers chasing too few tasks. Whatever, I'm not going to jump the gun.

But this app_info for MB is what you need. Assuming that you run a 64-bit version of Windows, this will preserve cached intel_gpu tasks issued as stock. (For 32-bit OSs, remove the <platform> line - but I'm assuming few volunteers will run 32-bit on such new processors)

<app_info>
<app>
<name>setiathome_v7</name>
</app>
<file_info>
<name>setiathome_7.03_windows_intelx86__opencl_intel_gpu_sah.exe</name>
<executable/>
</file_info>
<file_info>
<name>MultiBeam_Kernels_r2061_windows_intel_gpu.cl</name>
<executable/>
</file_info>
<file_info>
<name>libfftw3f-3.dll</name>
<executable/>
</file_info>
<file_info>
<name>mb_cmdline-7.03-opencl_intel_gpu_sah.txt</name>
</file_info>
<app_version>
<app_name>setiathome_v7</app_name>
<version_num>703</version_num>
<platform>windows_intelx86</platform>
<coproc>
<type>intel_gpu</type>
<count>1</count>
</coproc>
<plan_class>opencl_intel_gpu_sah</plan_class>
<avg_ncpus>0.04</avg_ncpus>
<max_ncpus>0.04</max_ncpus>
<file_ref>
<file_name>setiathome_7.03_windows_intelx86__opencl_intel_gpu_sah.exe</file_name>
<main_program/>
</file_ref>
<file_ref>
<file_name>MultiBeam_Kernels_r2061_windows_intel_gpu.cl</file_name>
<open_name>MultiBeam_Kernels_r2061.cl</open_name>
</file_ref>
<file_ref>
<file_name>libfftw3f-3.dll</file_name>
</file_ref>
<file_ref>
<file_name>mb_cmdline-7.03-opencl_intel_gpu_sah.txt</file_name>
<open_name>mb_cmdline.txt</open_name>
</file_ref>
</app_version>
</app_info>


You must also - as always - supply the necessary files manually. Here are the download links:

http://boinc2.ssl.berkeley.edu/sah/download_fanout/setiathome_7.03_windows_intelx86__opencl_intel_gpu_sah.exe (931 KB)
http://boinc2.ssl.berkeley.edu/sah/download_fanout/libfftw3f-3.dll (2,235 KB)
http://boinc2.ssl.berkeley.edu/sah/download_fanout/MultiBeam_Kernels_r2061_windows_intel_gpu.cl (114 KB)
http://boinc2.ssl.berkeley.edu/sah/download_fanout/mb_cmdline-7.03-opencl_intel_gpu_sah.txt (0 bytes)

The last one is an empty file - a placeholder for future options. Browsers won't allow a manual download, so just create an empty text file called "mb_cmdline-7.03-opencl_intel_gpu_sah.txt". We'll get somebody at Lunatics to put together notes on what cmd options are available for this specific build (the online readme file is wrong), but the app runs fine with the file empty.

Important notes

1) Check that you have a compatible Intel chip before wasting any time. You need one of

3rd Generation Intel® Core™ Processors with:
Intel® HD Graphics 4000/2500

4th Generation Intel® Core™ Processors with:
Intel® HD Graphics 4200/4400/4600/5000
Intel® IrisTM Graphics 5100
Intel® IrisTM Pro graphics 5200

2) Use an Intel GPU driver which supports OpenCL 1.2: the recommended package (assuming a 64-bit OS, as before) is Win64_153117 - available in both .EXE and .ZIP form from the Intel Download Center. This driver (9.18.10.3257) is compatible with both SETI and Einstein applications: there is a 32-bit version available too, if you persevere with the Intel search tools.

3) Connect the display output from your Intel GPU to a suitable load - a spare monitor port or dummy plug.

4) Use a current version of BOINC for correct Intel GPU detection - v7.2.33 (recommended) or v7.2.39 (testing) are best. Check http://boinc.berkeley.edu/download_all.php.

5) If you have been Beta testing an earlier version of the Intel OpenCL application under app_info, check that you have been using the same <plan_class>. If you need to transition, place this <app_version> higher in the combined app_info than your existing entries, and verify that newly fetched work is marked as 'opencl_intel_gpu_sah'. This will help you achieve painless upgrades with future releases of the Lunatics installer.

Profile Rick A. Sponholz
Volunteer tester
Avatar
Send message
Joined: 13 Jun 99
Posts: 43
Credit: 19,741,597
RAC: 7,938
United States
Message 1473294 - Posted: 6 Feb 2014, 14:47:36 UTC - in response to Message 1472053.
Last modified: 6 Feb 2014, 14:50:37 UTC

Richard,
After several days of trying to merge the app_info file you posted with the app_info file generated by Lunatics' installer, I still need more help. How do I merge these two files, so seti will also work on the intel gpu? seti beta is working fine on both of my intel gpu's. BTW, the machines also have two GTX690's running other projects and seti on them. Thanks again, Rick
____________

Richard HaselgroveProject donor
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8458
Credit: 48,598,056
RAC: 79,550
United Kingdom
Message 1473312 - Posted: 6 Feb 2014, 15:11:01 UTC - in response to Message 1473294.

Richard,
After several days of trying to merge the app_info file you posted with the app_info file generated by Lunatics' installer, I still need more help. How do I merge these two files, so seti will also work on the intel gpu? seti beta is working fine on both of my intel gpu's. BTW, the machines also have two GTX690's running other projects and seti on them. Thanks again, Rick

The process is very much as I described to Bil in message 1471242:

You will need to 'splice' the two app_info files together - link them together, one after the other, but take out the

</app_info>
<app_info>

pair in the middle of the file, to hide the join. You just want one <app_info> right at the start, and one </app_info> at the very end.

I'd suggest that you copy the Lunatics generated file to a temporary folder so you can work on it.

Open the file for editing with whatever simple text editor you usually use. The Windows-supplied 'Notepad' is ideal. Then, paste the Intel GPU file at the very end, after everything else. Finally, take out the two lines above to hide the join, and save the combined file. That's really all you need to do.

Check it all by eye, put it back into the project directory (replacing the existing file), and re-start BOINC.

Ulrich Metzner
Volunteer tester
Avatar
Send message
Joined: 3 Jul 02
Posts: 976
Credit: 8,402,301
RAC: 8,730
Germany
Message 1473324 - Posted: 6 Feb 2014, 15:31:50 UTC - in response to Message 1473312.

(...)
Check it all by eye, put it back into the project directory (replacing the existing file), and re-start BOINC.

...and ensure, it's saved with the name "app_info.xml" and NOT "app_info.xml.txt", as often accidentally done by users having file extensions hidden in Windows. ;)
____________
Aloha, Uli

Previous · 1 · 2

Message boards : Number crunching : Intel GPU App ready?

Copyright © 2014 University of California