Open Beta test: SoG for NVidia, Lunatics v0.45 - Beta6 (RC again)

Message boards : Number crunching : Open Beta test: SoG for NVidia, Lunatics v0.45 - Beta6 (RC again)
Message board moderation

To post messages, you must log in.

Previous · 1 . . . 12 · 13 · 14 · 15 · 16 · 17 · 18 . . . 32 · Next

AuthorMessage
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1810360 - Posted: 19 Aug 2016, 12:41:33 UTC - in response to Message 1810325.  

. . Thanks Richard, that clarifies things. Just need to spread the word to non-onedrive users like myself.
ID: 1810360 · Report as offensive
The_Matrix
Volunteer tester

Send message
Joined: 17 Nov 03
Posts: 414
Credit: 5,827,850
RAC: 0
Germany
Message 1810716 - Posted: 20 Aug 2016, 13:25:50 UTC
Last modified: 20 Aug 2016, 14:06:00 UTC

...but if you see big positive change over default please report.

So i read it in the documentation (old perhaps).

actual settings are -tune 2 1 1 64 i was wrong before(1,2,1,64)

i got a speed increase by one workunit (SoG/non Guppi) of ~300 seconds faster that before,

truly i overclocked the gpus memory by 200Mhz, but thats not much effecting the results too much, or too ?

Divers 356.45 , GTX 670 , Windows 7 , freeing up 2 cores or less has no effect.

Besides they´re vulkan drivers

Greetings

It´s the kernel ID i changed , what´s that ?
ID: 1810716 · Report as offensive
Profile Mike Special Project $75 donor
Volunteer tester
Avatar

Send message
Joined: 17 Feb 01
Posts: 33723
Credit: 79,922,639
RAC: 80
Germany
Message 1810744 - Posted: 20 Aug 2016, 15:30:25 UTC - in response to Message 1810716.  
Last modified: 20 Aug 2016, 15:41:00 UTC

...but if you see big positive change over default please report.

So i read it in the documentation (old perhaps).

actual settings are -tune 2 1 1 64 i was wrong before(1,2,1,64)

i got a speed increase by one workunit (SoG/non Guppi) of ~300 seconds faster that before,

truly i overclocked the gpus memory by 200Mhz, but thats not much effecting the results too much, or too ?

Divers 356.45 , GTX 670 , Windows 7 , freeing up 2 cores or less has no effect.

Besides they´re vulkan drivers

Greetings

It´s the kernel ID i changed , what´s that ?


This depends on memory timing and is always host dependent.
Tuning Kernel 2 sometimes gives speed up but not always.
Thats the reason i wrote testing required in the read me.
More important in this case is you now have work group size of 64 instead of 128.
Thats probably speed up in your case.
Would be most probably the same if you`d use 1 1 1 64.
This option was tested only and ATI GPU`s back then.

On my old HD 7970 i had tune option for both kernel 1 and kernel 2 for example.

On NV work group size of 256 or 512 should be faster.
AMD GPU`s only have wG of 256 whilst most modern NV GPU`s have 1024.


With each crime and every kindness we birth our future.
ID: 1810744 · Report as offensive
The_Matrix
Volunteer tester

Send message
Joined: 17 Nov 03
Posts: 414
Credit: 5,827,850
RAC: 0
Germany
Message 1810746 - Posted: 20 Aug 2016, 15:38:55 UTC - in response to Message 1810744.  
Last modified: 20 Aug 2016, 15:53:26 UTC

More important in this case is you now have kernel size of 64 instead of 128.


yeah, that why on 128 the percentage display got stuck for more than 10 seconds, i don´t wana wait on results ;D

Edit:

i´ll check that out...
ID: 1810746 · Report as offensive
robertmiles
Volunteer tester

Send message
Joined: 16 Jan 12
Posts: 213
Credit: 4,117,756
RAC: 6
United States
Message 1810809 - Posted: 20 Aug 2016, 20:57:26 UTC - in response to Message 1810360.  

. . Thanks Richard, that clarifies things. Just need to spread the word to non-onedrive users like myself.


I also don't use onedrive. I'm not sure if Windows Vista even supports onedrive.
ID: 1810809 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14542
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1810844 - Posted: 20 Aug 2016, 22:12:15 UTC - in response to Message 1810809.  

. . Thanks Richard, that clarifies things. Just need to spread the word to non-onedrive users like myself.

I also don't use onedrive. I'm not sure if Windows Vista even supports onedrive.

Provided it supports a web browser, I think it should be able to reach the download address.
ID: 1810844 · Report as offensive
robertmiles
Volunteer tester

Send message
Joined: 16 Jan 12
Posts: 213
Credit: 4,117,756
RAC: 6
United States
Message 1810866 - Posted: 20 Aug 2016, 23:36:31 UTC - in response to Message 1810844.  

. . Thanks Richard, that clarifies things. Just need to spread the word to non-onedrive users like myself.

I also don't use onedrive. I'm not sure if Windows Vista even supports onedrive.

Provided it supports a web browser, I think it should be able to reach the download address.


I tried it. It was able to reach this web page, which still has about half of each file name hidden (at least under my Mozilla Firefox browser).

https://onedrive.live.com/?authkey=%21AK7t%2D36P9nAec2Y&id=8D83BF774A4A86F5%21979&cid=8D83BF774A4A86F5
ID: 1810866 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13415
Credit: 208,696,464
RAC: 304
Australia
Message 1810867 - Posted: 20 Aug 2016, 23:45:32 UTC - in response to Message 1810866.  
Last modified: 20 Aug 2016, 23:45:49 UTC

I tried it. It was able to reach this web page, which still has about half of each file name hidden (at least under my Mozilla Firefox browser).

https://onedrive.live.com/?authkey=%21AK7t%2D36P9nAec2Y&id=8D83BF774A4A86F5%21979&cid=8D83BF774A4A86F5

If you do as Richard suggested in his earlier post and select the List view you'll be able to see the full file name for each file.


Grant
Darwin NT
ID: 1810867 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1813722 - Posted: 30 Aug 2016, 6:03:26 UTC
Last modified: 30 Aug 2016, 6:05:10 UTC

. . Hi Richard,

. . Now that WOW is over for another year I have run Beta(4) on my little donkey (3.0GHz C2D E7600 with nvidia GT730). I shut down BOINC first then ran it. As usual the install was seamless. The only thing that surprised me was that while I selected SoG r3500 and there is no SoG section in my app_config.xml it appears to be still following the config for CUDA50 and is still running 2 WU on GPU.

. . But just letting you know install was AOK.

. . Stephen
ID: 1813722 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14542
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1813746 - Posted: 30 Aug 2016, 9:18:30 UTC - in response to Message 1813722.  

. . Now that WOW is over for another year I have run Beta(4) on my little donkey (3.0GHz C2D E7600 with nvidia GT730). I shut down BOINC first then ran it. As usual the install was seamless. The only thing that surprised me was that while I selected SoG r3500 and there is no SoG section in my app_config.xml it appears to be still following the config for CUDA50 and is still running 2 WU on GPU.

It depends how you set up your app_config.xml file. I ended up using

<app_config>
   <app>
      <name>setiathome_v8</name>
      <gpu_versions>
          <gpu_usage>.50</gpu_usage>
          <cpu_usage>.04</cpu_usage>
      </gpu_versions>
   </app>

   <app_version>
       <app_name>setiathome_v8</app_name>
       <plan_class>opencl_nvidia_SoG</plan_class>
       <avg_ncpus>1.0</avg_ncpus>
       <ngpus>1.0</ngpus>
   </app_version>
</app_config>

Note that the first section sets up the device usage for all setiathome_v8 GPU versions. The second section sets up different (override) values for the opencl_nvidia_SoG <plan_class>. And I don't have an entry for cuda50 anywhere.

So, what happens when you switch applications in mid-flow? Previously cached tasks will still have the cuda50 labeling, so will use (in my case) the generic settings - even though the installer-generated app_info.xml file will direct BOINC to run the SoG application for these tasks. Newly fetched work will be labeled opencl_nvidia_SoG. and the bespoke setting in app_config will come into play when those tasks are reached. Another benefit of using a short cache when experimenting with new applications.
ID: 1813746 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1813752 - Posted: 30 Aug 2016, 10:47:59 UTC - in response to Message 1813746.  

. . Now that WOW is over for another year I have run Beta(4) on my little donkey (3.0GHz C2D E7600 with nvidia GT730). I shut down BOINC first then ran it. As usual the install was seamless. The only thing that surprised me was that while I selected SoG r3500 and there is no SoG section in my app_config.xml it appears to be still following the config for CUDA50 and is still running 2 WU on GPU.

It depends how you set up your app_config.xml file. I ended up using

<app_config>
   <app>
      <name>setiathome_v8</name>
      <gpu_versions>
          <gpu_usage>.50</gpu_usage>
          <cpu_usage>.04</cpu_usage>
      </gpu_versions>
   </app>

   <app_version>
       <app_name>setiathome_v8</app_name>
       <plan_class>opencl_nvidia_SoG</plan_class>
       <avg_ncpus>1.0</avg_ncpus>
       <ngpus>1.0</ngpus>
   </app_version>
</app_config>

Note that the first section sets up the device usage for all setiathome_v8 GPU versions. The second section sets up different (override) values for the opencl_nvidia_SoG <plan_class>. And I don't have an entry for cuda50 anywhere.

So, what happens when you switch applications in mid-flow? Previously cached tasks will still have the cuda50 labeling, so will use (in my case) the generic settings - even though the installer-generated app_info.xml file will direct BOINC to run the SoG application for these tasks. Newly fetched work will be labeled opencl_nvidia_SoG. and the bespoke setting in app_config will come into play when those tasks are reached. Another benefit of using a short cache when experimenting with new applications.



. . Yes, running the cache down might have been a very good idea. I still had a fairly full cache from WOW :(

. . I like that there is always something new to learn. I guess for now the quick solution is to modify the CUDA50 section to run singles. I will add the SoG section for when new work does arrive.

. . Actually I just thought, what if I edit client_state.xml to replace the cuda headers with SoG headers?

.
ID: 1813752 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14542
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1813755 - Posted: 30 Aug 2016, 11:01:06 UTC - in response to Message 1813752.  

Actually I just thought, what if I edit client_state.xml to replace the cuda headers with SoG headers?

Check that the <version_num> entries are right for that before you start, else it's a bigger switch.

For WOW, I set up a bespoke app_info.xml file with both cuda50 and SoG applications - just a single app_version for each, with the same version_num, because I wasn't needing to cater for all the various upgrade paths from stock and previous Lunatics versions.

Under those controlled conditions, all you need to do is replace the line

<plan_class>cuda50</plan_class>

with

<plan_class>opencl_nvidia_SoG</plan_class>

in the <result> section of each task, or vice-versa. That allowed my machines to process guppies with SoG, and Arecibo tasks with CUDA (getting a couple of CPUs back): I kept Notepad++ open through the fortnight, with the global 'Replace' dialog open, and a couple of times a day stopped BOINC to switch over the tasks I wanted to run with the alternate application.
ID: 1813755 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1813811 - Posted: 30 Aug 2016, 15:33:17 UTC - in response to Message 1813755.  

Actually I just thought, what if I edit client_state.xml to replace the cuda headers with SoG headers?

Check that the <version_num> entries are right for that before you start, else it's a bigger switch.

For WOW, I set up a bespoke app_info.xml file with both cuda50 and SoG applications - just a single app_version for each, with the same version_num, because I wasn't needing to cater for all the various upgrade paths from stock and previous Lunatics versions.

Under those controlled conditions, all you need to do is replace the line

<plan_class>cuda50</plan_class>

with

<plan_class>opencl_nvidia_SoG</plan_class>

in the <result> section of each task, or vice-versa. That allowed my machines to process guppies with SoG, and Arecibo tasks with CUDA (getting a couple of CPUs back): I kept Notepad++ open through the fortnight, with the global 'Replace' dialog open, and a couple of times a day stopped BOINC to switch over the tasks I wanted to run with the alternate application.



. . That is very clever. I like lateral thinking. I will have to check that SoG r3500 still like MB 812 tasks.
ID: 1813811 · Report as offensive
Stephen "Heretic" Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 20 Sep 12
Posts: 5557
Credit: 192,787,363
RAC: 628
Australia
Message 1814133 - Posted: 31 Aug 2016, 12:22:10 UTC - in response to Message 1813755.  

Actually I just thought, what if I edit client_state.xml to replace the cuda headers with SoG headers?

Check that the <version_num> entries are right for that before you start, else it's a bigger switch.

For WOW, I set up a bespoke app_info.xml file with both cuda50 and SoG applications - just a single app_version for each, with the same version_num, because I wasn't needing to cater for all the various upgrade paths from stock and previous Lunatics versions.

Under those controlled conditions, all you need to do is replace the line

<plan_class>cuda50</plan_class>

with

<plan_class>opencl_nvidia_SoG</plan_class>

in the <result> section of each task, or vice-versa. That allowed my machines to process guppies with SoG, and Arecibo tasks with CUDA (getting a couple of CPUs back): I kept Notepad++ open through the fortnight, with the global 'Replace' dialog open, and a couple of times a day stopped BOINC to switch over the tasks I wanted to run with the alternate application.



. . I moved some guppi tasks from the CPU queue to the GPU queue as SoG and worked AOK.

.
ID: 1814133 · Report as offensive
Profile SongBird
Volunteer tester

Send message
Joined: 23 Oct 01
Posts: 104
Credit: 164,826,157
RAC: 297
Bulgaria
Message 1814277 - Posted: 1 Sep 2016, 7:41:33 UTC

Hi,

May I ask what does "Signal on GPU" mean? We've been testing signals on GPUs for years, haven't we? What changed?

Regards,
Georgi
ID: 1814277 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14542
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1814286 - Posted: 1 Sep 2016, 8:10:12 UTC - in response to Message 1814277.  

May I ask what does "Signal on GPU" mean? We've been testing signals on GPUs for years, haven't we? What changed?

Good question. Raistmer will be able to explain it better, but my understanding is:

Each separate signal - as you say - has been found by mathematical processing on the GPU for years. But up until now, every potential candidate signal has been passed back over the bus to the CPU and main memory for indexing, comparing with previous candidates, sorting, and finally deciding what to report. My understanding is that many more potential candidates are considered before the search identifies the central peak of the signal - and all the bus transfers are slow and costly compared with the internal speed of the GPU.

So, SoG does more of the logical housekeeping directly on the GPU, and only passes the fully-identified signals back to the CPU.

It's ironic that the CPU is thus doing less actual scientific work during task processing than with the older cuda applications, but because of the different development tools used, shows up as being busier. You win some, you lose some.
ID: 1814286 · Report as offensive
AMDave
Volunteer tester

Send message
Joined: 9 Mar 01
Posts: 234
Credit: 11,671,730
RAC: 0
United States
Message 1814338 - Posted: 1 Sep 2016, 14:11:03 UTC - in response to Message 1814286.  

May I ask what does "Signal on GPU" mean? We've been testing signals on GPUs for years, haven't we? What changed?

Good question. Raistmer will be able to explain it better, but my understanding is:

Each separate signal - as you say - has been found by mathematical processing on the GPU for years. But up until now, every potential candidate signal has been passed back over the bus to the CPU and main memory for indexing, comparing with previous candidates, sorting, and finally deciding what to report. My understanding is that many more potential candidates are considered before the search identifies the central peak of the signal - and all the bus transfers are slow and costly compared with the internal speed of the GPU.

So, SoG does more of the logical housekeeping directly on the GPU, and only passes the fully-identified signals back to the CPU.

It's ironic that the CPU is thus doing less actual scientific work during task processing than with the older cuda applications, but because of the different development tools used, shows up as being busier. You win some, you lose some.

Plus:

ID: 1814338 · Report as offensive
EdwardPF
Volunteer tester

Send message
Joined: 26 Jul 99
Posts: 389
Credit: 236,772,605
RAC: 374
United States
Message 1814761 - Posted: 3 Sep 2016, 3:28:38 UTC

Boy this is embarrassing (again).

I am running boinc 6.12.34 on windows-7 12GB mem core-7 (gen 1) Nvidia 770.

I would like to help here (if possible) but get lost in the associated threads RE: this.

I downloaded "Lunatics_Win64_v0.45_Beta4-for-SoG_setup.exe" and ran it. It ran just fine. Now what do I do to actually start helping (If I can).

I am running BOINC 6.12.34 now because it fits my needs just fine. is all this compatible?

A thread need to be started every once in a while to help us who are late in jumping in. Otherwise all the back-and-forth in the thread only helps us in getting lost ... perhaps a thread for technical updates and a re-introduction thread for each new beta re-start.

(this make any sense?)

Ed F

P.s. My computers are hidden
ID: 1814761 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13415
Credit: 208,696,464
RAC: 304
Australia
Message 1814762 - Posted: 3 Sep 2016, 3:42:53 UTC - in response to Message 1814761.  

I downloaded "Lunatics_Win64_v0.45_Beta4-for-SoG_setup.exe" and ran it. It ran just fine. Now what do I do to actually start helping (If I can).

That's it.
You ran the installer, it installed & kept all your existing work.
Everything's good.


If you start having issues with the new application, or want to get more out of it, then the "Seti applications for Nvidia GPU improvement" thread or one of the several support/optimisation threads would be the place to start.
Grant
Darwin NT
ID: 1814762 · Report as offensive
robertmiles
Volunteer tester

Send message
Joined: 16 Jan 12
Posts: 213
Credit: 4,117,756
RAC: 6
United States
Message 1814770 - Posted: 3 Sep 2016, 5:25:36 UTC - in response to Message 1814761.  

Boy this is embarrassing (again).

I am running boinc 6.12.34 on windows-7 12GB mem core-7 (gen 1) Nvidia 770.

I would like to help here (if possible) but get lost in the associated threads RE: this.

I downloaded "Lunatics_Win64_v0.45_Beta4-for-SoG_setup.exe" and ran it. It ran just fine. Now what do I do to actually start helping (If I can).

I am running BOINC 6.12.34 now because it fits my needs just fine. is all this compatible?

A thread need to be started every once in a while to help us who are late in jumping in. Otherwise all the back-and-forth in the thread only helps us in getting lost ... perhaps a thread for technical updates and a re-introduction thread for each new beta re-start.

(this make any sense?)

Ed F

P.s. My computers are hidden


I wasn't quite as late starting. You might want to mention whether has any program able to show the temperature of the GPU - one of mine was overheating.
ID: 1814770 · Report as offensive
Previous · 1 . . . 12 · 13 · 14 · 15 · 16 · 17 · 18 . . . 32 · Next

Message boards : Number crunching : Open Beta test: SoG for NVidia, Lunatics v0.45 - Beta6 (RC again)


 
©2023 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.