Checkpointing Fixed MAC OS X OpenCL Application
log in

Advanced search

Message boards : News : Checkpointing Fixed MAC OS X OpenCL Application

1 · 2 · 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: 2
Message 13289 - Posted: 12 Jan 2012, 23:00:03 UTC

Version 3.06 has been released for Mac OS X which fixes the checkpointing problem experienced with previous versions.

juice3
Send message
Joined: 6 Dec 11
Posts: 19
Credit: 971,466
RAC: 0
Message 13290 - Posted: 12 Jan 2012, 23:21:30 UTC

Do we need to reattach or reset?

juice3
Send message
Joined: 6 Dec 11
Posts: 19
Credit: 971,466
RAC: 0
Message 13291 - Posted: 12 Jan 2012, 23:27:25 UTC

For the long runs id like to see more credit though. 38,000+ second gpu runtime and 3k credit. ;(

Profile arkayn
Volunteer tester
Avatar
Send message
Joined: 30 Aug 09
Posts: 219
Credit: 676,877,192
RAC: 28,917
Message 13292 - Posted: 13 Jan 2012, 0:40:13 UTC

Are you going to release a OpenCL app for Windows as well, the HD7970's do not have CAL on them.

At Milkyway they have found that running the Nvidia OpenCL app works fine on it though.
____________

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: 2
Message 13300 - Posted: 13 Jan 2012, 16:39:45 UTC - in response to Message 13291.

For the long runs id like to see more credit though. 38,000+ second gpu runtime and 3k credit. ;(


Unfortunately, OpenCL doesn't perform nearly as well as CUDA or CAL so while the CUDA and ATI apps may be 10-20 times faster than a CPU, the OpenCL app is only 5-10 times faster. That's partly due to OpenCL being layered on top of CUDA or CAL and partly due to OpenCL being so generic that it can't make good use of the GPUs advanced features which can be taken advantage of when using CUDA or CAL. For example, Gipsel tweaked the assembly of the ATI app to improve speed on 5xxx and 6xxx cards by up to 20%. That isn't possible with OpenCL as the code is compiled at runtime. It is up to nVidia or AMD to improve their compilers as the developer can no longer tweak the code to improve performance.

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: 2
Message 13301 - Posted: 13 Jan 2012, 16:47:08 UTC - in response to Message 13292.

Are you going to release a OpenCL app for Windows as well, the HD7970's do not have CAL on them.

At Milkyway they have found that running the Nvidia OpenCL app works fine on it though.


I released the OS X app first because there was no AMD support for OS X. That being said, I actually created the app on Windows first because I prefer VS 2010 over XCode as an IDE. So yes, there is a Windows OpenCL version that runs on both nVidia and AMD. Performance is lousy compared to the Brook+/CAL app though. So, I'll have to come up with a way so that if the GPU supports CAL, the server will send it the ATI app and if not, will send the OpenCL app.

Nikolay
Send message
Joined: 10 Jan 12
Posts: 11
Credit: 174,679
RAC: 0
Message 13302 - Posted: 13 Jan 2012, 19:59:12 UTC
Last modified: 13 Jan 2012, 20:01:00 UTC

Unfortunately, OpenCL doesn't perform nearly as well as CUDA or CAL so while the CUDA and ATI apps may be 10-20 times faster than a CPU, the OpenCL app is only 5-10 times faster. That's partly due to OpenCL being layered on top of CUDA or CAL and partly due to OpenCL being so generic that it can't make good use of the GPUs advanced features which can be taken advantage of when using CUDA or CAL. For example, Gipsel tweaked the assembly of the ATI app to improve speed on 5xxx and 6xxx cards by up to 20%. That isn't possible with OpenCL as the code is compiled at runtime. It is up to nVidia or AMD to improve their compilers as the developer can no longer tweak the code to improve performance.


I hope that is only the temporary issue. The compilers would be improved and OpenCL
would compute with the same speed, or even faster than the older technologies! =)

I released the OS X app first because there was no AMD support for OS X. That being said, I actually created the app on Windows first because I prefer VS 2010 over XCode as an IDE. So yes, there is a Windows OpenCL version that runs on both nVidia and AMD. Performance is lousy compared to the Brook+/CAL app though. So, I'll have to come up with a way so that if the GPU supports CAL, the server will send it the ATI app and if not, will send the OpenCL app.


At the moment, your project is the ONLY project that I can use my GPU for computing it!

About two other projects that could use Mac AMD GPU:

1) Albert@Home is not working because of the bug :
http://albert.phys.uwm.edu/forum_thread.php?id=8854#111583

A fix is already available and should be part of the next release (i.e. 7.0.8).

Cheers,
Oliver (project administrator and developer)


2) Milkyway@Home does not work because my GPU (AMD HD 6750M) is single precision.

In fact, I entered your project only because of that.
You are awesome! Keep up the good work, please! ;-)

Profile cenit
Send message
Joined: 11 Sep 09
Posts: 14
Credit: 22,679,602
RAC: 0
Message 13305 - Posted: 14 Jan 2012, 11:25:46 UTC - in response to Message 13301.
Last modified: 14 Jan 2012, 11:26:08 UTC

Are you going to release a OpenCL app for Windows as well, the HD7970's do not have CAL on them.

At Milkyway they have found that running the Nvidia OpenCL app works fine on it though.


I released the OS X app first because there was no AMD support for OS X. That being said, I actually created the app on Windows first because I prefer VS 2010 over XCode as an IDE. So yes, there is a Windows OpenCL version that runs on both nVidia and AMD. Performance is lousy compared to the Brook+/CAL app though. So, I'll have to come up with a way so that if the GPU supports CAL, the server will send it the ATI app and if not, will send the OpenCL app.


well, it should be integrated into boinc. If an app is better on a host, newer boinc software releases should detect it (maybe after some test sending both types of app) and then send the best one!

Claggy
Send message
Joined: 27 Sep 09
Posts: 288
Credit: 14,320,498
RAC: 0
Message 13310 - Posted: 14 Jan 2012, 13:43:38 UTC - in response to Message 13302.


About two other projects that could use Mac AMD GPU:

1) Albert@Home is not working because of the bug :
http://albert.phys.uwm.edu/forum_thread.php?id=8854#111583

A fix is already available and should be part of the next release (i.e. 7.0.8).

Cheers,
Oliver (project administrator and developer)


Boinc 7.0.8 was released last night.

Claggy

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: 2
Message 13323 - Posted: 16 Jan 2012, 13:42:00 UTC - in response to Message 13305.

well, it should be integrated into boinc. If an app is better on a host, newer boinc software releases should detect it (maybe after some test sending both types of app) and then send the best one!


They tried that with 32-bit vs 64-bit applications. It didn't work so well. Several days after crunching with the 64-bit app which is more than double the speed of the 32-bit apps, users with 64-bit machines were still getting sent 32-bit apps. The fix was to tell the server to never ever send 32-bit anything to 64-bit hosts. I would expect any such logic about whether to use CAL vs OpenCL would result in the same problems.

mac intel
Send message
Joined: 21 Mar 11
Posts: 10
Credit: 10,140,700
RAC: 0
Message 13507 - Posted: 16 Feb 2012, 8:57:47 UTC - in response to Message 13323.
Last modified: 16 Feb 2012, 9:01:06 UTC

hello, 3.06 works well with 10.7, 7.0.15 for x86_64-apple-darwin and NVIDIA GTS450.

it is the first Collatz GPU app that works for me without errors. thx a lot

GPU load is about 55% is there a way to set higher load by myself?

also would i like to test alpha apps if posible

Profile anthropisches
Send message
Joined: 2 Nov 09
Posts: 2
Credit: 669,693,367
RAC: 0
Message 13534 - Posted: 22 Feb 2012, 17:19:05 UTC
Last modified: 22 Feb 2012, 17:45:26 UTC

What is my false?

I use an MacPro with GPU HD 6750m, BOINC 7.08.

All GPU work abort in a short time. :
http://boinc.thesonntags.com/collatz/results.php?hostid=92611

What can I do?

mathias from germany

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: 2
Message 13539 - Posted: 23 Feb 2012, 22:48:46 UTC - in response to Message 13507.

hello, 3.06 works well with 10.7, 7.0.15 for x86_64-apple-darwin and NVIDIA GTS450.

it is the first Collatz GPU app that works for me without errors. thx a lot

GPU load is about 55% is there a way to set higher load by myself?

also would i like to test alpha apps if posible


Take a look at the app_info.xml and readme in the windows 3.06 version available in the optimized apps (link on the main web page, and the Windows opt app name is collatz_3.06_windows_X86_64_opencl_opt_app.zip). The OS X app has the same options as the windows app. I just haven't taken the time to create an app_info for it yet (e.g. Just change the executable name and that should be the only difference.)

You might want to try "S1 T22 L8 N22" as the command line. If it doesn't crash the video driver, it will probably increase the GPU utilization considerably.

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: 2
Message 13540 - Posted: 23 Feb 2012, 22:49:52 UTC - in response to Message 13534.

What is my false?

I use an MacPro with GPU HD 6750m, BOINC 7.08.

All GPU work abort in a short time. :
http://boinc.thesonntags.com/collatz/results.php?hostid=92611

What can I do?

mathias from germany


I'm not really sure. We've been seeing a lot of error in OS X 11.3.

mac intel
Send message
Joined: 21 Mar 11
Posts: 10
Credit: 10,140,700
RAC: 0
Message 13541 - Posted: 24 Feb 2012, 8:37:36 UTC
Last modified: 24 Feb 2012, 9:13:34 UTC

hello, thx for reply.

<app_info>
<app>
<name>collatz</name>
<user_friendly_name>collatz</user_friendly_name>
</app>
<file_info>
<name>collatz_3.06_x86_64-apple-darwin__opencl_osx</name>
<executable/>
</file_info>
<app_version>
<app_name>collatz</app_name>
<version_num>209</version_num>
<flops>1.0e11</flops>
<api_version>6.1.0</api_version>
<coproc>
<type>ATI</type>
<count>1.0</count>
</coproc>
<plan_class>ati13ati</plan_class>
<avg_ncpus>0.02</avg_ncpus>
<max_ncpus>0.05</max_ncpus>
<cmdline>S1 T22 L8 N18</cmdline>
<file_ref>
<file_name>collatz_3.06_x86_64-apple-darwin__opencl_osx</file_name>
<main_program/>
</file_ref>
</app_version>
<app>
<name>mini_collatz</name>
<user_friendly_name>mini_collatz</user_friendly_name>
</app>
<file_info>
<name>mini_collatz_3.06_x86_64-apple-darwin__opencl_osx</name>
<executable/>
</file_info>
<app_version>
<app_name>mini_collatz</app_name>
<version_num>200</version_num>
<avg_ncpus>1.00</avg_ncpus>
<max_ncpus>1.00</max_ncpus>
<plan_class>sse</plan_class>
<file_ref>
<file_name>mini_collatz_3.06_x86_64-apple-darwin__opencl_osx</file_name>
<main_program/>
</file_ref>
</app_version>
</app_info>


but please, what do i have to insert here?
<type>ATI</type>
<plan_class>ati13ati</plan_class>

i´ve got an NVIDIA GTS450 GPU and want to run mini_collatz_3.06_x86_64-apple-darwin__opencl_osx or both ;)

and another question, won´t belong here but where do i have to sent my boinc alpha tester mail. sry, it is the first time for that i use this :)

@anthropisches: do you run 10.7 ?

mac intel
Send message
Joined: 21 Mar 11
Posts: 10
Credit: 10,140,700
RAC: 0
Message 13542 - Posted: 24 Feb 2012, 12:29:06 UTC - in response to Message 13541.
Last modified: 24 Feb 2012, 12:43:24 UTC

got it. now with S1 T22 L8 N22 98 % GPU load but i´ve got these message:
Collatz Conjecture | Found app_info.xml; using anonymous platform

and in boinc it is shown as:
Lokal: mini_collatz 3.06 (opencl_osx)

is this right or could i change it ?

here is my new app_info.xml:


<app_info>
<app>
<name>collatz</name>
<user_friendly_name>collatz</user_friendly_name>
</app>
<file_info>
<name>collatz_3.06_x86_64-apple-darwin__opencl_osx</name>
<executable/>
</file_info>
<app_version>
<app_name>collatz</app_name>
<version_num>306</version_num>
<flops>1.0e11</flops>
<api_version>6.1.0</api_version>
<coproc>
<type>NVIDIA</type>
<count>1.0</count>
</coproc>
<plan_class>opencl_osx</plan_class>
<avg_ncpus>0.02</avg_ncpus>
<max_ncpus>0.05</max_ncpus>
<cmdline>S1 T22 L8 N18</cmdline>
<file_ref>
<file_name>collatz_3.06_x86_64-apple-darwin__opencl_osx</file_name>
<main_program/>
</file_ref>
</app_version>
<app>
<name>mini_collatz</name>
<user_friendly_name>mini_collatz</user_friendly_name>
</app>
<file_info>
<name>mini_collatz_3.06_x86_64-apple-darwin__opencl_osx</name>
<executable/>
</file_info>
<app_version>
<app_name>mini_collatz</app_name>
<version_num>306</version_num>
<flops>1.0e11</flops>
<api_version>6.1.0</api_version>
<coproc>
<type>NVIDIA</type>
<count>1.0</count>
</coproc>
<plan_class>opencl_osx</plan_class>
<avg_ncpus>0.02</avg_ncpus>
<max_ncpus>0.05</max_ncpus>
<cmdline>S1 T22 L8 N18</cmdline>
<file_ref>
<file_name>mini_collatz_3.06_x86_64-apple-darwin__opencl_osx</file_name>
<main_program/>
</file_ref>
</app_version>
</app_info>


thx in advance

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: 2
Message 13543 - Posted: 24 Feb 2012, 15:55:10 UTC - in response to Message 13542.
Last modified: 24 Feb 2012, 15:55:31 UTC

got it. now with S1 T22 L8 N22 98 % GPU load but i´ve got these message:
Collatz Conjecture | Found app_info.xml; using anonymous platform

and in boinc it is shown as:
Lokal: mini_collatz 3.06 (opencl_osx)

is this right or could i change it ?



That is just BOINC's way of making sure you know that it is using an app_info file instead of running the application with default settings.

Congratulations on getting it up and running!

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: 2
Message 13544 - Posted: 24 Feb 2012, 16:35:34 UTC - in response to Message 13542.

I have added collatz_3.06_x86_64-apple-darwin__opencl_osx.zip as an optimized app using the above app_info as a guide. There are both nVidia and AMD versions of the app_info included. In addition, it should also support running mini_collatz workunits on the CPU.

Hint: Change your preferences to allow only Collatz but to accept work from other applications. That way, your GPU will run the longer Collatz workunits and since they are not available for the CPU, the CPU will run the shorter mini_collatz workunits via accepting work from other applications.

Profile anthropisches
Send message
Joined: 2 Nov 09
Posts: 2
Credit: 669,693,367
RAC: 0
Message 13549 - Posted: 25 Feb 2012, 18:33:38 UTC - in response to Message 13540.
Last modified: 25 Feb 2012, 18:34:05 UTC

What is my false?

I use an MacPro with GPU HD 6750m, BOINC 7.08.

All GPU work abort in a short time. :
http://boinc.thesonntags.com/collatz/results.php?hostid=92611

What can I do?

mathias from germany


I'm not really sure. We've been seeing a lot of error in OS X 11.3.

Since 2 hours it runs! I'm not sure, maybe you've done something, but now it works! Thank You?!
(I work with OS X 10.7.3)

anthropisches

mac intel
Send message
Joined: 21 Mar 11
Posts: 10
Credit: 10,140,700
RAC: 0
Message 13553 - Posted: 26 Feb 2012, 13:16:41 UTC - in response to Message 13544.
Last modified: 26 Feb 2012, 13:35:13 UTC

hello, thx for your support again.
now i use your xml because there is
mini_collatz_2.03_x86_64-apple-darwin inside too !

but got some questions:
could i configure that only one mini_collatz_2.03_x86_64-apple-darwin task is running (right now 2) or limit CPU usage to 50 % and could i edit that my GPU works with two collatz_3.06_x86_64-apple-darwin__opencl_osx Wu´s ?

are these edit´s right for OS X 10.6/10.7 and 1024 mb ram and important?
<platform>osx_intelx86</platform>
<gpu_ram>1073741824.00</gpu_ram>

could i use these config´s for any projekt ?
<avg_ncpus>0.020000</avg_ncpus>
<max_ncpus>0.050000</max_ncpus>
<cmdline>S1 T22 L8 N22</cmdline>

is there a sample config for cuda too ?

thx in advance

1 · 2 · Next
Post to thread

Message boards : News : Checkpointing Fixed MAC OS X OpenCL Application


Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.