Server Upgrade on Thursday, April 18th
log in

Advanced search

Message boards : News : Server Upgrade on Thursday, April 18th

1 · 2 · 3 · Next
Author Message
Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 16250 - Posted: 16 Apr 2013, 21:35:09 UTC

On Thursday morning, April 18th, the Collatz server will be taken down to upgrade it to the latest BOINC server software. This upgrade should enable Collatz to support OpenCL applications on Intel embedded GPUs.

zombie67 [MM]
Volunteer tester
Avatar
Send message
Joined: 3 Jul 09
Posts: 156
Credit: 612,751,377
RAC: 293
Message 16251 - Posted: 16 Apr 2013, 23:23:06 UTC
Last modified: 16 Apr 2013, 23:24:32 UTC

I can't wait!

Will there be an app too? Or will that come later?
____________
Dublin, California
Team: SETI.USA

Profile mikey
Avatar
Send message
Joined: 11 Aug 09
Posts: 3242
Credit: 1,695,004,707
RAC: 5,340,085
Message 16253 - Posted: 17 Apr 2013, 10:52:06 UTC - in response to Message 16250.

On Thursday morning, April 18th, the Collatz server will be taken down to upgrade it to the latest BOINC server software. This upgrade should enable Collatz to support OpenCL applications on Intel embedded GPUs.


WOO HOO!!! That means my laptop might just work here!

Dirk Broer
Send message
Joined: 20 Aug 10
Posts: 33
Credit: 176,957,861
RAC: 578,680
Message 16255 - Posted: 17 Apr 2013, 11:47:47 UTC
Last modified: 17 Apr 2013, 11:48:01 UTC

You might have to write the Intel OpenCL application yourself then...able to support does not have to mean they have an application ready for you.
____________

zombie67 [MM]
Volunteer tester
Avatar
Send message
Joined: 3 Jul 09
Posts: 156
Credit: 612,751,377
RAC: 293
Message 16256 - Posted: 17 Apr 2013, 13:59:25 UTC - in response to Message 16255.

You might have to write the Intel OpenCL application yourself then...able to support does not have to mean they have an application ready for you.

I know. That's why I asked the question.

In the past, Slicker said that he would write one, but not until the server supported the Intel platform. Otherwise there was no point.
____________
Dublin, California
Team: SETI.USA

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 16264 - Posted: 18 Apr 2013, 19:31:55 UTC - in response to Message 16256.

You might have to write the Intel OpenCL application yourself then...able to support does not have to mean they have an application ready for you.

I know. That's why I asked the question.

In the past, Slicker said that he would write one, but not until the server supported the Intel platform. Otherwise there was no point.


In theory, the same executable should work. So, I just copied the OpenCL for nVidia as if it was OpenCL for Intel. To give it a try, go into the preferences, check solo_collatz, and wait for BOINC to do an update. Pleaselet me know if it runs OK or if it errors out.

zombie67 [MM]
Volunteer tester
Avatar
Send message
Joined: 3 Jul 09
Posts: 156
Credit: 612,751,377
RAC: 293
Message 16267 - Posted: 18 Apr 2013, 21:06:31 UTC - in response to Message 16264.
Last modified: 18 Apr 2013, 21:15:11 UTC

I'll try it out now.

But which device selection controls tasks being assigned to the GPU? There are still only the three choices, CPU/ATI GPU/nVidia GPU. I thought there would be a new 4th selection for Intel GPU?

Edit: With only CPU selected, and only solo_collatz selected, and "accept work from other applications?" set to NO, I got one task each for the CPU and for the Intel GPU. How does one set it so that you get tasks only for the Intel GPU?

The CPU task is running in panic mode, with an estimated time of 16 days. But based on the first 5 minutes, it looks like it will really take only about 27 hours.

The GPU task errored out immediately:

http://boinc.thesonntags.com/collatz/result.php?resultid=139019569

Stderr output

<core_client_version>7.0.62</core_client_version>
<![CDATA[
<message>
- exit code -54 (0xffffffca)
</message>
<stderr_txt>
Collatz Conjecture v4.04 x86_64 for OpenCL 1.0 or later
Based on the AMD Brook+ kernels by Gipsel
Invalid work group size enqueue kernelZeroBuffer at line 650
clEnqueueNDRangeKernel(queue, kernelZeroBuffer, 1, NULL,65536,0,0,NULL,NULL);
14:09:09 (988): called boinc_finish

</stderr_txt>
]]>
____________
Dublin, California
Team: SETI.USA

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 16269 - Posted: 18 Apr 2013, 21:49:18 UTC - in response to Message 16267.

I'll try it out now.

But which device selection controls tasks being assigned to the GPU? There are still only the three choices, CPU/ATI GPU/nVidia GPU. I thought there would be a new 4th selection for Intel GPU?

Edit: With only CPU selected, and only solo_collatz selected, and "accept work from other applications?" set to NO, I got one task each for the CPU and for the Intel GPU. How does one set it so that you get tasks only for the Intel GPU?

The CPU task is running in panic mode, with an estimated time of 16 days. But based on the first 5 minutes, it looks like it will really take only about 27 hours.

The GPU task errored out immediately:

http://boinc.thesonntags.com/collatz/result.php?resultid=139019569

Stderr output

<core_client_version>7.0.62</core_client_version>
<![CDATA[
<message>
- exit code -54 (0xffffffca)
</message>
<stderr_txt>
Collatz Conjecture v4.04 x86_64 for OpenCL 1.0 or later
Based on the AMD Brook+ kernels by Gipsel
Invalid work group size enqueue kernelZeroBuffer at line 650
clEnqueueNDRangeKernel(queue, kernelZeroBuffer, 1, NULL,65536,0,0,NULL,NULL);
14:09:09 (988): called boinc_finish

</stderr_txt>
]]>


Doh! I should have used the AMD OpenCL app instead of the NVidia one for Intel GPUs. The difference is that the AMD app defaults to 64 work groups whereas the NVidia defaults to 512. The temporary solution is to edit the .config file in the project folder and add the following two lines:

verbose=1
threads=64

If that works, I'll go ahead and create a new app. As far as AMD, vs NVidia vs Intel choices, I agree that there should be an option. If I can figure out how to add it I will do so as well as send the code to the BOINC developers.

zombie67 [MM]
Volunteer tester
Avatar
Send message
Joined: 3 Jul 09
Posts: 156
Credit: 612,751,377
RAC: 293
Message 16270 - Posted: 18 Apr 2013, 23:59:56 UTC - in response to Message 16269.
Last modified: 19 Apr 2013, 0:11:35 UTC

Doh! I should have used the AMD OpenCL app instead of the NVidia one for Intel GPUs. The difference is that the AMD app defaults to 64 work groups whereas the NVidia defaults to 512. The temporary solution is to edit the .config file in the project folder and add the following two lines:

verbose=1
threads=64

If that works, I'll go ahead and create a new app. As far as AMD, vs NVidia vs Intel choices, I agree that there should be an option. If I can figure out how to add it I will do so as well as send the code to the BOINC developers.


Hmm. There is no .config file. I thought one was created automatically?

Edit: I tried creating a collatz.config file manually, with those two lines in it. But still getting failures:

http://boinc.thesonntags.com/collatz/result.php?resultid=139025682

Stderr output

<core_client_version>7.0.62</core_client_version>
<![CDATA[
<message>
- exit code -54 (0xffffffca)
</message>
<stderr_txt>
Collatz Conjecture v4.04 x86_64 for OpenCL 1.0 or later
Based on the AMD Brook+ kernels by Gipsel
Invalid work group size enqueue kernelZeroBuffer at line 650
clEnqueueNDRangeKernel(queue, kernelZeroBuffer, 1, NULL,65536,0,0,NULL,NULL);
17:07:23 (5652): called boinc_finish

</stderr_txt>
]]>

P.S.: How does the collatz.config file work, when there are both CPU and GPU tasks? Are there two files? Or...?
____________
Dublin, California
Team: SETI.USA

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 16272 - Posted: 19 Apr 2013, 2:49:36 UTC - in response to Message 16270.

Doh! I should have used the AMD OpenCL app instead of the NVidia one for Intel GPUs. The difference is that the AMD app defaults to 64 work groups whereas the NVidia defaults to 512. The temporary solution is to edit the .config file in the project folder and add the following two lines:

verbose=1
threads=64

If that works, I'll go ahead and create a new app. As far as AMD, vs NVidia vs Intel choices, I agree that there should be an option. If I can figure out how to add it I will do so as well as send the code to the BOINC developers.


Hmm. There is no .config file. I thought one was created automatically?

Edit: I tried creating a collatz.config file manually, with those two lines in it. But still getting failures:

http://boinc.thesonntags.com/collatz/result.php?resultid=139025682

Stderr output

<core_client_version>7.0.62</core_client_version>
<![CDATA[
<message>
- exit code -54 (0xffffffca)
</message>
<stderr_txt>
Collatz Conjecture v4.04 x86_64 for OpenCL 1.0 or later
Based on the AMD Brook+ kernels by Gipsel
Invalid work group size enqueue kernelZeroBuffer at line 650
clEnqueueNDRangeKernel(queue, kernelZeroBuffer, 1, NULL,65536,0,0,NULL,NULL);
17:07:23 (5652): called boinc_finish

</stderr_txt>
]]>

P.S.: How does the collatz.config file work, when there are both CPU and GPU tasks? Are there two files? Or...?


I would expect to be named <EXE_NAME_WITHOUT_EXTENSION>.config and located in the project folder. When the task runs, it should be copied/renamed to collatz.config in the slot folder.

I created a 4.05 version which contains additional defaults. Thanks for posting the error. The 6th parameter shouldn't have been 0. It should use 64 at a minimum or use the default for the device as returned by the OpenCL driver.

BOINC does not yet support OpenCL CPU apps. That is potentially slated for the 7.1.x client. It won't be in the next 7.0.x release for sure. There are separate config files though since the executable names are different.

mexico
Send message
Joined: 12 May 10
Posts: 4
Credit: 5,400,466
RAC: 0
Message 16274 - Posted: 19 Apr 2013, 4:12:17 UTC

After the server upgrade, I am not able to get any GPU work to my AMD ATI Radeon HD 5x00 series (Cedar) (512MB) driver: 1.4.635. I am using the optimized ATI GPU application version 2.09 by Gipsel (Win64, CAL 1.4)

19-Apr-13 06:59:30 | Collatz Conjecture | [sched_op] Starting scheduler request
19-Apr-13 06:59:30 | Collatz Conjecture | Sending scheduler request: To fetch work.
19-Apr-13 06:59:30 | Collatz Conjecture | Requesting new tasks for ATI
19-Apr-13 06:59:30 | Collatz Conjecture | [sched_op] CPU work request: 0.00 seconds; 0.00 devices
19-Apr-13 06:59:30 | Collatz Conjecture | [sched_op] ATI work request: 2159.14 seconds; 0.49 devices
19-Apr-13 06:59:38 | Collatz Conjecture | Scheduler request completed: got 0 new tasks
19-Apr-13 06:59:38 | Collatz Conjecture | [sched_op] Server version 701
19-Apr-13 06:59:38 | Collatz Conjecture | No tasks sent
19-Apr-13 06:59:38 | Collatz Conjecture | No tasks are available for mini_collatz
19-Apr-13 06:59:38 | Collatz Conjecture | Message from server: Your app_info.xml file doesn't have a usable version of solo_collatz.

That's the host.

Profile Sabroe_SMC
Send message
Joined: 12 Jul 09
Posts: 10
Credit: 2,068,456,111
RAC: 744
Message 16275 - Posted: 19 Apr 2013, 9:35:26 UTC - in response to Message 16274.

Sure, the server status shows zero at normal and mini collatz WUs since the server are back.
I hope there will be work for our GPU afap
____________
Should we shag now, or shag later?

mexico
Send message
Joined: 12 May 10
Posts: 4
Credit: 5,400,466
RAC: 0
Message 16277 - Posted: 19 Apr 2013, 12:21:09 UTC - in response to Message 16275.

Sure, the server status shows zero at normal and mini collatz WUs since the server are back.
I hope there will be work for our GPU afap


Mhm, you are right. :) Looks like it's sending only solo_collatz WUs. I don't remember reading somewhere about this happening after the server upgrade and that's why I posted here. :)

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 16278 - Posted: 19 Apr 2013, 13:04:00 UTC

It turns out that there was a bug in the work generator that's been there since day one. I have now fixed it which should eliminate any of the "Project has no work" messages. I also sent the fix the BOINC developers so that they can add it (or at least a comment explaining the issue) into the sample work generator so that when people like me use the sample as a starting template for their own code, it won't be screwed up if there are multiple apps (collatz, mini, and solo).

Profile Slicker
Volunteer moderator
Project administrator
Project developer
Project tester
Project scientist
Avatar
Send message
Joined: 11 Jun 09
Posts: 2525
Credit: 740,580,099
RAC: 1
Message 16279 - Posted: 19 Apr 2013, 13:06:23 UTC - in response to Message 16277.

Sure, the server status shows zero at normal and mini collatz WUs since the server are back.
I hope there will be work for our GPU afap


Mhm, you are right. :) Looks like it's sending only solo_collatz WUs. I don't remember reading somewhere about this happening after the server upgrade and that's why I posted here. :)


Once things settle down, I'll probably be adding a mini-solo app for CPUs tasks as the solo are the same large size as the collatz app.

Profile Sabroe_SMC
Send message
Joined: 12 Jul 09
Posts: 10
Credit: 2,068,456,111
RAC: 744
Message 16280 - Posted: 19 Apr 2013, 13:33:06 UTC - in response to Message 16278.

It turns out that there was a bug in the work generator that's been there since day one. I have now fixed it which should eliminate any of the "Project has no work" messages. I also sent the fix the BOINC developers so that they can add it (or at least a comment explaining the issue) into the sample work generator so that when people like me use the sample as a starting template for their own code, it won't be screwed up if there are multiple apps (collatz, mini, and solo).

Good work Slicker Thumbs Up
____________
Should we shag now, or shag later?

zombie67 [MM]
Volunteer tester
Avatar
Send message
Joined: 3 Jul 09
Posts: 156
Credit: 612,751,377
RAC: 293
Message 16281 - Posted: 19 Apr 2013, 13:50:24 UTC - in response to Message 16272.
Last modified: 19 Apr 2013, 13:52:45 UTC

I would expect to be named <EXE_NAME_WITHOUT_EXTENSION>.config and located in the project folder. When the task runs, it should be copied/renamed to collatz.config in the slot folder.

Ah! I understand now. I added these to the solo_collatz_4.04_windows_x86_64__opencl_intel_gpu.config file.

verbose=1
threads=64

Edit: I there is a "Use Intel GPU" setting now. Thanks!

Edit2: I just got work and it's running. I'll report back when complete.
____________
Dublin, California
Team: SETI.USA

zombie67 [MM]
Volunteer tester
Avatar
Send message
Joined: 3 Jul 09
Posts: 156
Credit: 612,751,377
RAC: 293
Message 16282 - Posted: 19 Apr 2013, 15:14:27 UTC
Last modified: 19 Apr 2013, 15:25:32 UTC

24% complete after 80 minutes.

Observation: The task shows 0.02C + 1 intel_gpu GPU

In reality, the task is consuming almost full thread, according to the Task Manager. It fluctuates from 11.9-12.5% (8 thread machine, so 12.5% = a full thread).

Edit: Even with 8 other CPU tasks running, GPU-Z shows 99% load on the GPU. So reserving more of the CPU by the GPU app would not improve GPU performance anyway. Just FYI.
____________
Dublin, California
Team: SETI.USA

Profile arkayn
Volunteer tester
Avatar
Send message
Joined: 30 Aug 09
Posts: 219
Credit: 676,877,192
RAC: 17,625
Message 16283 - Posted: 19 Apr 2013, 15:26:10 UTC

It still likes to send the CUDA 23 app to newer generation cards, but I just switched my preferences to only send the Solo Collatz work now.
____________

Profile x3mEn
Send message
Joined: 14 Jul 10
Posts: 6
Credit: 10,264,715
RAC: 0
Message 16285 - Posted: 19 Apr 2013, 17:19:44 UTC

19/04/2013 18:02:10 | | Processor: 4 GenuineIntel Intel(R) Core(TM) i5-3570K CPU @ 3.40GHz [Family 6 Model 58 Stepping 9]
19/04/2013 18:02:10 | | OpenCL: Intel GPU 0: Intel(R) HD Graphics 4000 (driver version 9.17.10.2932, device version OpenCL 1.1, 728MB, 728MB available, 45 GFLOPS peak)
19/04/2013 20:01:56 | Collatz Conjecture | Sending scheduler request: Requested by user.
19/04/2013 20:01:56 | Collatz Conjecture | Requesting new tasks for intel_gpu
19/04/2013 20:01:58 | Collatz Conjecture | Scheduler request completed: got 0 new tasks

1 · 2 · 3 · Next
Post to thread

Message boards : News : Server Upgrade on Thursday, April 18th


Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.