OpenCL NV MultiBeam v8 SoG edition for Windows

Message boards : Number crunching : OpenCL NV MultiBeam v8 SoG edition for Windows
Message board moderation

To post messages, you must log in.

1 · 2 · 3 · 4 . . . 18 · Next

AuthorMessage
Profile Raistmer
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 16 Jun 01
Posts: 6325
Credit: 106,370,077
RAC: 121
Russia
Message 1762525 - Posted: 5 Feb 2016, 16:40:41 UTC

Here http://lunatics.kwsn.info/index.php?action=downloads;sa=view;down=487 new MultiBeam build for nVidia GPUs available as standalone pack.

Feel free to use it and post comments here.
ID: 1762525 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1762547 - Posted: 5 Feb 2016, 17:22:22 UTC - in response to Message 1762525.  

Please note that this pack uses values which do not conform to the likely pattern of future stock deployment:

<version_num>802</version_num>
<plan_class>cuda</plan_class>

At Beta, Eric used "8.07 (opencl_nvidia_SoG)"

At Main, I would expect the version number to be 800, and the plan class to be the same as Beta. If you deploy the pack in its current form, be ready to make adjustments when the next installer comes out.
ID: 1762547 · 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 1762548 - Posted: 5 Feb 2016, 17:29:30 UTC - in response to Message 1762547.  

Please note that this pack uses values which do not conform to the likely pattern of future stock deployment:

<version_num>802</version_num>
<plan_class>cuda</plan_class>

At Beta, Eric used "8.07 (opencl_nvidia_SoG)"

At Main, I would expect the version number to be 800, and the plan class to be the same as Beta. If you deploy the pack in its current form, be ready to make adjustments when the next installer comes out.


It was in our inner RC more then week already to note this before release and not after it.
Well, till next release.
ID: 1762548 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1762552 - Posted: 5 Feb 2016, 17:53:35 UTC

This is what happens when the developer makes private releases without thinking about the context within which the developed application will be used.
ID: 1762552 · 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 1762555 - Posted: 5 Feb 2016, 17:58:54 UTC - in response to Message 1762552.  

This is what happens when the developer makes private releases without thinking about the context within which the developed application will be used.

Provided app_info snippet allows copy/paste w/o any change. But if I would receive adequate testing/feedback before release I could make it easier for next version of automatization tools like Lunatics installer to accommodate to it.
Well, let it be the biggest issue of all :)
ID: 1762555 · 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 1762559 - Posted: 5 Feb 2016, 18:03:35 UTC - in response to Message 1762557.  

Thanks, I'll update aistub in next commit.
ID: 1762559 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1762562 - Posted: 5 Feb 2016, 18:05:45 UTC - in response to Message 1762557.  

I look forward to it as well.

500 validated on 1 machine, 2400 validated on another on beta.

Any timeline as to when these will be released on Main? (one can hope)
ID: 1762562 · 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 1762568 - Posted: 5 Feb 2016, 18:11:08 UTC - in response to Message 1762562.  
Last modified: 5 Feb 2016, 18:11:56 UTC


Any timeline as to when these will be released on Main? (one can hope)


Release as stock will require unattended deployment and stealth execution (versus "user knows what he's doing" on anonymous platform. So before such release issues with low-end GPUs should be solved (at least driver restarts). So, if you have low-end GPU and experience lags/driver restart with this build in its default config, and if you was able to find tuning that allow to solve this issues please post your command line here or in corresponding thread on beta: http://setiweb.ssl.berkeley.edu/beta/forum_thread.php?id=2288
ID: 1762568 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1762578 - Posted: 5 Feb 2016, 18:30:58 UTC

@ Tutankhamon,

Thanks for posting that. We can't be certain that 800 / opencl_nvidia_SoG will be the ultimate deployment values - they depend on Eric, and how much water has flowed under the bridge before the scheduler issues have been resolved to his satisfaction, for official Berkeley release.

For full public releases, we normally prefer to add the .cl file info and ref lines into app_info as well: in this case, they would be

    <file_info>
        <name>MultiBeam_Kernels_r3366.cl</name>
    </file_info>

        <file_ref>
            <file_name>MultiBeam_Kernels_r3366.cl</file_name>
        </file_ref>

Before the generation of aistub flies was automated, that was one additional step to forget during early-stage testing, and so they were often omitted during pre-release testing.

@ Raistmer,

It's a good thing you put that disclaimer about 'experienced users only' and 'may cause driver restarts on low-end cards' into your initial release post.

It's because I knew that you were still working with Eric to work round such deployment issues that I allowed myself a week's holiday from development matters after the installer release.
ID: 1762578 · 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 1762586 - Posted: 5 Feb 2016, 18:36:56 UTC - in response to Message 1762578.  

'may cause driver restarts on low-end cards' into your initial release post.

http://lunatics.kwsn.info/index.php?action=downloads;sa=view;down=487
347+ drivers required.
For low-end cards tuning required to avoid lags or driver restarts.
Also some tuning recommended to maximize throughout on top devices.
ID: 1762586 · 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 1762648 - Posted: 5 Feb 2016, 21:35:32 UTC - in response to Message 1762647.  


-sbs 256

Did you reconsider omitting this one?
ID: 1762648 · 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 1762662 - Posted: 5 Feb 2016, 22:03:41 UTC - in response to Message 1762653.  


-sbs 256

Did you reconsider omitting this one?

Well, I will try it here at first. Then maybe delete it. I think I need to do many more WU's with -sbs 256, before I can say for sure, whether or not it is faster or slower than the default setting.

yes, cause the influence of this option is AR-dependent.
ID: 1762662 · 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 1762663 - Posted: 5 Feb 2016, 22:07:32 UTC - in response to Message 1762653.  


-sbs 256

Did you reconsider omitting this one?

Well, I will try it here at first. Then maybe delete it. I think I need to do many more WU's with -sbs 256, before I can say for sure, whether or not it is faster or slower than the default setting.


Also try -sbs 192 and 384.

Should be faster on your GPU.


With each crime and every kindness we birth our future.
ID: 1762663 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1762667 - Posted: 5 Feb 2016, 22:28:09 UTC - in response to Message 1762666.  

Tut are you getting work for that OpenCL SOG?
ID: 1762667 · 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 1762671 - Posted: 5 Feb 2016, 22:59:44 UTC - in response to Message 1762667.  

Tut are you getting work for that OpenCL SOG?


Maybe you need to change plan claas to opencl_nvidia_sah until Eric has released plan class for SoG version.


With each crime and every kindness we birth our future.
ID: 1762671 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14649
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1762679 - Posted: 5 Feb 2016, 23:17:33 UTC - in response to Message 1762671.  

Tut are you getting work for that OpenCL SOG?

Maybe you need to change plan claas to opencl_nvidia_sah until Eric has released plan class for SoG version.

Plan Class names used under Anonymous Platform don't have to match the plan classes used for stock distributions - I've made up plan classes including my initials and the word 'test' before now, and they worked just fine.

But they should include the keyword for the type of scheduling anticipated - OpenCL in this case (for BOINC versions >= 7.0.40). All mine did, so I can't speak for what happens if you leave it out. It'll be in a (debug) log if you fall foul of something, and need to look it up.

This is the other way round, but error messages might look something like this:

11/15/2012 8:53:52 AM | | App version needs opencl but GPU doesn't support it
ID: 1762679 · 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 1762683 - Posted: 5 Feb 2016, 23:35:37 UTC - in response to Message 1762681.  

Very high CPU usage for WU's other than High AR's. Almost a full core, for AR's other than VHAR's where the CPU usage is 8-10% only.

Since the WU's I tested this with on BETA, was all above 2.something in AR, the low CPU usage was what surprised me most. However here on main, with mostly lower AR's the high CPU usage really shows.

Thanks Dog, that we do not get VLAR's for CPU here, or even this GTX980 would come to a screeching halt :-)

EDIT: But SoG is fast, scaringly fast. Geeze....

ATi OpenCL build handles VLAR quite easely. Worth to try with OpenCL NV also.
That's the disadvantage of beta - subset of ARs, subset of devices...

Pulses and Triplets still processed by old way - and synhing uses lot of CPU as before (again, NV-specific).
ID: 1762683 · 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 1762684 - Posted: 5 Feb 2016, 23:37:28 UTC - in response to Message 1762681.  

Very high CPU usage for WU's other than High AR's. Almost a full core, for AR's other than VHAR's where the CPU usage is 8-10% only.

Since the WU's I tested this with on BETA, was all above 2.something in AR, the low CPU usage was what surprised me most. However here on main, with mostly lower AR's the high CPU usage really shows.

Thanks Dog, that we do not get VLAR's for CPU here, or even this GTX980 would come to a screeching halt :-)

EDIT: But SoG is fast, scaringly fast. So I can live with high CPU usage, just dropping a CPU core from CPU crunching.

Geeze....


You can try _use_sleep or -use_sleep_ex 5 to reduce CPU usage.
But i suggst to use this only running multiple instances.


With each crime and every kindness we birth our future.
ID: 1762684 · 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 1762686 - Posted: 5 Feb 2016, 23:53:05 UTC - in response to Message 1762685.  

because by using -use_sleep, this app will not be any faster than CUDA50.

Would be interesting to check this BTW.
Sleep() implemented mostly in PulseFind area. And VHAR has small amount of PulseFind so -use_sleep impact there would be quite small and CPU savings with midrange AR could be substantional.
From other side, balancing overall host performance depends on GPU vs CPU work share. For fast GPUs most of host RAC should come from GPU part and CPU part could be neglectible.
ID: 1762686 · 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 1762687 - Posted: 5 Feb 2016, 23:54:27 UTC - in response to Message 1762685.  
Last modified: 5 Feb 2016, 23:55:28 UTC

Very high CPU usage for WU's other than High AR's. Almost a full core, for AR's other than VHAR's where the CPU usage is 8-10% only.

Since the WU's I tested this with on BETA, was all above 2.something in AR, the low CPU usage was what surprised me most. However here on main, with mostly lower AR's the high CPU usage really shows.

Thanks Dog, that we do not get VLAR's for CPU here, or even this GTX980 would come to a screeching halt :-)

EDIT: But SoG is fast, scaringly fast. So I can live with high CPU usage, just dropping a CPU core from CPU crunching.

Geeze....


You can try _use_sleep or -use_sleep_ex 5 to reduce CPU usage.
But i suggst to use this only running multiple instances.

Well, running 3 at a time is indeed multiple instances. However, I'll wait and see if I can live with this, because by using -use_sleep, this app will not be any faster than CUDA50.


Thats why i suggested -use_sleep_ex 5.
Shouldn`t be much slower running 3 instances but reduces CPU usage at least a little bit.

Running benches atm.


With each crime and every kindness we birth our future.
ID: 1762687 · Report as offensive
1 · 2 · 3 · 4 . . . 18 · Next

Message boards : Number crunching : OpenCL NV MultiBeam v8 SoG edition for Windows


 
©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.