Lunatics Windows Installer v0.40 release notes

Message boards : Number crunching : Lunatics Windows Installer v0.40 release notes
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 · 8 . . . 14 · Next

AuthorMessage
LadyL
Volunteer tester
Avatar

Send message
Joined: 14 Sep 11
Posts: 1679
Credit: 5,230,097
RAC: 0
Message 1210938 - Posted: 28 Mar 2012, 8:34:06 UTC - in response to Message 1210859.  

This app works as I crunched 5 last night

ap_6.00_win_x86_SSE3_OpenCL_NV_r540.exe

AstroPulse_Kernels_r540.cl

The name of the application is v6 and call for 601 in the app_info.xml

1 hour 15 minutes approx / task

http://setiathome.berkeley.edu/results.php?userid=8612202&offset=0&show_names=0&state=0&appid=12


Michael Miles



ouch, is that still online? [assorted swearwords]
Where did you get that from? It's a AP 6.00 it doesn't contain the code the validator needs to detect outliers!

Please replace ASAP with this.
Haven't gotten around to write an appropriate app_info entry but

<cmdline>-instances_per_device 1 -unroll 4 -ffa_block 2048 -ffa_block_fetch 1024 -sbs 128</cmdline>


should be a good if low starting point.


I'm not the Pope. I don't speak Ex Cathedra!
ID: 1210938 · Report as offensive
Profile Michael W.F. Miles
Avatar

Send message
Joined: 24 Mar 07
Posts: 268
Credit: 34,410,870
RAC: 0
Canada
Message 1210946 - Posted: 28 Mar 2012, 8:57:29 UTC

Roger that
Thank you
ID: 1210946 · Report as offensive
Profile BilBg
Volunteer tester
Avatar

Send message
Joined: 27 May 07
Posts: 3720
Credit: 9,385,827
RAC: 0
Bulgaria
Message 1211081 - Posted: 28 Mar 2012, 15:31:39 UTC - in response to Message 1210310.  
Last modified: 28 Mar 2012, 15:54:15 UTC


Small discrepancy in different ReadMe files

Here:
http://lunatics.kwsn.net/index.php?module=Downloads;sa=dlview;id=328
http://lunatics.kwsn.net/downloads/Lunatics%20ReadMev0.40.txt

"You will usually find that the SSSE3x application is fastest on the Intel
processors which support SSSE3x, except on dual-core CPUs with especially fast
memory subsystems."

Somehow the last part was deleted (Or not added initially? "Lunatics ReadMe.txt" found back in Lunatics_Win??_v0.38_setup.exe have the same omission).

In other places (ReleaseNotes.txt, first post here) the sentence is:

"You will usually find that the SSSE3x application is fastest on the Intel
processors which support SSSE3x, except on dual-core CPUs with especially fast
memory subsystems, where SSE4.1 may be faster."


 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 
ID: 1211081 · Report as offensive
S@NL - John van Gorsel
Volunteer tester
Avatar

Send message
Joined: 5 Jul 99
Posts: 193
Credit: 139,673,078
RAC: 0
Netherlands
Message 1211085 - Posted: 28 Mar 2012, 15:55:51 UTC
Last modified: 28 Mar 2012, 15:56:59 UTC

I installed v0.40 on this pc and initially it went well, but after about 20 Cuda tasks I noticed a sharp drop in processing rate.
MSI Afterburner tells me the GPU load of only 2% with an occasional peak to 90%. I run 2 tasks at the same time on this 560Ti and the normal GPU load is 95-97%.


The result files (stderr output) do not show any information that indicates a problem.

What can I do to solve this?


Seti@Netherlands website
ID: 1211085 · Report as offensive
S@NL - John van Gorsel
Volunteer tester
Avatar

Send message
Joined: 5 Jul 99
Posts: 193
Credit: 139,673,078
RAC: 0
Netherlands
Message 1211120 - Posted: 28 Mar 2012, 17:20:49 UTC
Last modified: 28 Mar 2012, 17:21:20 UTC

I went back to Lunatics_x38g_win32_cuda32.exe (only exchanged the Cuda executable and modified app_info accordingly) and it runs ok now.

Looked at most of the tasks finished today on this host and around 10-15% of these tasks took 10x the usual time. The usual time for a 2.7 AR task is 3 minutes, now some of them took 1800-3200 seconds.
ID: 1211120 · Report as offensive
LadyL
Volunteer tester
Avatar

Send message
Joined: 14 Sep 11
Posts: 1679
Credit: 5,230,097
RAC: 0
Message 1211128 - Posted: 28 Mar 2012, 17:31:36 UTC - in response to Message 1211120.  
Last modified: 28 Mar 2012, 17:35:53 UTC

I went back to Lunatics_x38g_win32_cuda32.exe (only exchanged the Cuda executable and modified app_info accordingly) and it runs ok now.

Looked at most of the tasks finished today on this host and around 10-15% of these tasks took 10x the usual time. The usual time for a 2.7 AR task is 3 minutes, now some of them took 1800-3200 seconds.


host/task links please. we didn't change anything in that ares from v0.39 but yours is iirc the second report of sudden speed loss.

driver version?

edit sorry, you gave that info already. I'm too tired...

560Ti? They are known to cause trouble with later x-branch, when they start getting pushed. jason has a post somewhere that lists about a dozen things you can check...
I'm not the Pope. I don't speak Ex Cathedra!
ID: 1211128 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14644
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1211132 - Posted: 28 Mar 2012, 17:36:22 UTC - in response to Message 1211128.  

I went back to Lunatics_x38g_win32_cuda32.exe (only exchanged the Cuda executable and modified app_info accordingly) and it runs ok now.

Looked at most of the tasks finished today on this host and around 10-15% of these tasks took 10x the usual time. The usual time for a 2.7 AR task is 3 minutes, now some of them took 1800-3200 seconds.

host/task links please. we didn't change anything in that ares from v0.39 but yours is iirc the second report of sudden speed loss.

driver version?

There's a link in the first post - host ID 5907499, driver 275.33

I'm trying to remember the history of x41g and 560Ti cards - I may need to do some digging. I think we had some early reports that some early samples of the 560Ti weren't being supplied with enough voltage to support the increased optimisation (equals workload) of x41g.
ID: 1211132 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1211138 - Posted: 28 Mar 2012, 17:43:19 UTC - in response to Message 1211132.  
Last modified: 28 Mar 2012, 17:49:14 UTC

I went back to Lunatics_x38g_win32_cuda32.exe (only exchanged the Cuda executable and modified app_info accordingly) and it runs ok now.

Looked at most of the tasks finished today on this host and around 10-15% of these tasks took 10x the usual time. The usual time for a 2.7 AR task is 3 minutes, now some of them took 1800-3200 seconds.

host/task links please. we didn't change anything in that ares from v0.39 but yours is iirc the second report of sudden speed loss.

driver version?

There's a link in the first post - host ID 5907499, driver 275.33

I'm trying to remember the history of x41g and 560Ti cards - I may need to do some digging. I think we had some early reports that some early samples of the 560Ti weren't being supplied with enough voltage to support the increased optimisation (equals workload) of x41g.


Looked through and spotted some long running shorties. There are major things to check with 560ti:
- sufficient core voltage such that it doesn't downclock (couldn't see evidence of downclock online, but possible)
- Running a solid state drive ? Make really sure its firmware is up to date... then check again. ( Yes this causes problems with Cuda apps due to broken SSDs hogging DMA, that was a real toughy to find )
- Update driver using the "clean Install" Advanced Option
- [Check temperatures] x41g pushes harder.
- running other projects with non-threadsafe Cuda apps ?
- If you run into those long runners, check the file sizes on those tasks... Could be a reappearance of those freaked out tasks, or corruption during download of some sort

Jason
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1211138 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14644
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1211141 - Posted: 28 Mar 2012, 17:51:51 UTC - in response to Message 1211138.  

- If you run into those long runners, check the file sizes on those tasks... Could be a reappearance of those freaked out tasks, or corruption during download of some sort

Jason

The mega-WUs only caused an extended delay at startup, while the coded receiver data was re-assembled using the CPU. Once the task got into its stride, the CUDA part ran at normal speed. Any other workunit corruption would have caused an error, not a slowdown - like an MD5 error, of a filesize mismatch during download, I'd have thought.
ID: 1211141 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1211142 - Posted: 28 Mar 2012, 17:54:47 UTC - in response to Message 1211141.  
Last modified: 28 Mar 2012, 18:03:27 UTC

- If you run into those long runners, check the file sizes on those tasks... Could be a reappearance of those freaked out tasks, or corruption during download of some sort

Jason

The mega-WUs only caused an extended delay at startup, while the coded receiver data was re-assembled using the CPU. Once the task got into its stride, the CUDA part ran at normal speed. Any other workunit corruption would have caused an error, not a slowdown - like an MD5 error, of a filesize mismatch during download, I'd have thought.


Could be. Just trying to list as many options to cover as I can recall. The 560ti were also originally sold with some crackpot drivers ( 266.66) that messed up a hidden Cuda kernel compute cache if they were ever installed, though doesn't look like the problem here.

I'm wagering on the other project, or SSD issues until we hear more back from the OP

[Edit:] While I'm asleep, If the OP eliminates those possiilities & still has trouble, I'd like yourself or other Lunatics crew to trial the x41u Cuda 4.1 build on the same machine if possible with the OP (would need attention to driver version etc). I'm in the process of cosmetic cleanups, mostly Kepler related, and intend to release after this AP 6.01 transition dust has settled anyway.

x41g should work just as well as x38g on that card really, but x41u Cuda 4.1 is in a different league on 560ti.

Jason
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1211142 · Report as offensive
S@NL - John van Gorsel
Volunteer tester
Avatar

Send message
Joined: 5 Jul 99
Posts: 193
Credit: 139,673,078
RAC: 0
Netherlands
Message 1211157 - Posted: 28 Mar 2012, 18:30:59 UTC - in response to Message 1211138.  
Last modified: 28 Mar 2012, 18:37:20 UTC



Looked through and spotted some long running shorties. There are major things to check with 560ti:
- sufficient core voltage such that it doesn't downclock (couldn't see evidence of downclock online, but possible)
- Running a solid state drive ? Make really sure its firmware is up to date... then check again. ( Yes this causes problems with Cuda apps due to broken SSDs hogging DMA, that was a real toughy to find )
- Update driver using the "clean Install" Advanced Option
- [Check temperatures] x41g pushes harder.
- running other projects with non-threadsafe Cuda apps ?
- If you run into those long runners, check the file sizes on those tasks... Could be a reappearance of those freaked out tasks, or corruption during download of some sort

Jason


1. This pc was running 24-7 in this configuration since September 25 2011. The 560Ti never showed any sign of downclocking. MSI Afterburner confirmed that nothing happened with the clockrates.
2. No SSD
3. Will update the driver and let you know the result.
4. As you can see in the Afterburner screenshot, the temperature was 60°C. Now with x38g running, the temperature is 70°C at 63% fan speed
5. No other Cuda apps were running

One other thing I should mention: prior to installing v0.40 I updated Boinc from 6.10.58 to the current 6.12.34.
ID: 1211157 · Report as offensive
Profile jason_gee
Volunteer developer
Volunteer tester
Avatar

Send message
Joined: 24 Nov 06
Posts: 7489
Credit: 91,093,184
RAC: 0
Australia
Message 1211159 - Posted: 28 Mar 2012, 18:38:17 UTC - in response to Message 1211157.  

1. This pc was running 24-7 in this configuration since September 25 2011. The 560Ti never showed any sign of downclocking. MSI Afterburner confirmed that nothing happened with the clockrates.
2. No SSD
3. Will update the driver and let you know the result.
4. As you can see in the Afterburner screenshot, the temperature was 60°C. Now with x38g running, the temperature is 70°C at 63% fan speed
5. No other Cuda apps were running

One other thing I should mention: prior to installing v0.40 I updated Boinc from 5.10.58 to the current 6.12.34.


Hmm, definitely weird. Well we'll find out one way or another. With those things eliminated, if after updating driver you still get weirdness, we will get you testing x41u. Can't very well push out an updated release next week if there is some unknown latent problem lurking.

Thanks for the cross-checking.

Jason
"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.
ID: 1211159 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14644
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1211173 - Posted: 28 Mar 2012, 19:11:39 UTC

If either John van Gorsel or Victor is game to try out the experimental x41u version, please PM me. I have them here ready to email (tell me your email address), or if you prefer I can find a hosting site for you to download from.
ID: 1211173 · Report as offensive
Profile zoom3+1=4
Volunteer tester
Avatar

Send message
Joined: 30 Nov 03
Posts: 65689
Credit: 55,293,173
RAC: 49
United States
Message 1211178 - Posted: 28 Mar 2012, 19:23:01 UTC - in response to Message 1211173.  

ID: 1211178 · Report as offensive
Dave

Send message
Joined: 29 Mar 02
Posts: 778
Credit: 25,001,396
RAC: 0
United Kingdom
Message 1211202 - Posted: 28 Mar 2012, 20:19:12 UTC

So just to confirm, when I get to this, which may be a week from now, i7-2 will be fastest with SSE3 correct? Or should I try SSSE4.x as it's a i7-2?

ID: 1211202 · Report as offensive
Josef W. Segur
Volunteer developer
Volunteer tester

Send message
Joined: 30 Oct 99
Posts: 4504
Credit: 1,414,761
RAC: 0
United States
Message 1211215 - Posted: 28 Mar 2012, 20:39:40 UTC - in response to Message 1211202.  

So just to confirm, when I get to this, which may be a week from now, i7-2 will be fastest with SSE3 correct? Or should I try SSSE4.x as it's a i7-2?


Any i7 should like the SSE3 better. Both the SSSE3x and SSE4.1 builds are really meant for Intel Core 2 architecture and are suboptimal for later Intel developments.
                                                                   Joe
ID: 1211215 · Report as offensive
S@NL - John van Gorsel
Volunteer tester
Avatar

Send message
Joined: 5 Jul 99
Posts: 193
Credit: 139,673,078
RAC: 0
Netherlands
Message 1211216 - Posted: 28 Mar 2012, 20:40:03 UTC - in response to Message 1211159.  


Hmm, definitely weird. Well we'll find out one way or another. With those things eliminated, if after updating driver you still get weirdness, we will get you testing x41u. Can't very well push out an updated release next week if there is some unknown latent problem lurking.

Jason


Some additional info:
Did a clean install of the latest nVidia driver (296.10), and reinstalled v0.40, and only changed the <count> tag to 0.5. I immediately noticed the intermittent load on the GPU, and both instances of the exe in the TaskManager confirmed this. Screenshot of Afterburner:



Just to be sure, I installed v0.40 again and left the <count> tag at 1. When I looked at the Afterburner graph I remembered having seen the same thing on my Linux pc's after I installed x41g a couple of weeks ago. All I needed to do was free up one core (set the "Use at most ...% of the processors" to 75% or 90%). In the graph below you can see where I set the processor use to 75%:



The Linux pc's were running GTX260 cards so it seem unrelated to the GTX560Ti. I have one other pc running a GTX580 and x38g (3 tasks simultanously, GPU load is a straight line at 96%). I can install x41g on that pc and see what happens.

Also no problem to test the x41u on the Q9660/GTX560Ti. This pc is only used for testing and running Boinc. I will send a PM to Richard.
ID: 1211216 · Report as offensive
JLConawayII

Send message
Joined: 2 Apr 02
Posts: 188
Credit: 2,840,460
RAC: 0
United States
Message 1211218 - Posted: 28 Mar 2012, 20:45:06 UTC
Last modified: 28 Mar 2012, 21:16:32 UTC

That looks identical to my ATI utilization when it's trying to run a VLAR.

edit: And, apparently, some normally marked WU's as well. Except CPU load has no effect on mine, it stays low.

http://setiathome.berkeley.edu/result.php?resultid=2373271846 This one, for instance, showed low utilization, where this one http://setiathome.berkeley.edu/result.php?resultid=2373271874 ran at 89% utilization and completed 5x faster
ID: 1211218 · Report as offensive
Profile Michael W.F. Miles
Avatar

Send message
Joined: 24 Mar 07
Posts: 268
Credit: 34,410,870
RAC: 0
Canada
Message 1211228 - Posted: 28 Mar 2012, 21:13:25 UTC

This app works as I crunched 5 last night

ap_6.00_win_x86_SSE3_OpenCL_NV_r540.exe

AstroPulse_Kernels_r540.cl

The name of the application is v6 and call for 601 in the app_info.xml

1 hour 15 minutes approx / task

http://setiathome.berkeley.edu/results.php?userid=8612202&offset=0&show_names=0&state=0&appid=12


Michael Miles





This app validated
http://setiathome.berkeley.edu/result.php?resultid=2370788573

http://setiathome.berkeley.edu/result.php?resultid=2371083839

These where the 540 app

Curious if the validation code was not in then it should not have validated?

I am running 555 app now

Thanks for the app
ID: 1211228 · Report as offensive
Richard Haselgrove Project Donor
Volunteer tester

Send message
Joined: 4 Jul 99
Posts: 14644
Credit: 200,643,578
RAC: 874
United Kingdom
Message 1211234 - Posted: 28 Mar 2012, 21:24:04 UTC - in response to Message 1211228.  

v6.00 with r540 should validate against v6.01 with r555.

The difference between them is not in task validation, but another part of the code which helps to keep the project running smoothly, with sensible runtime estimates.

That bit affects us all, so it would be a great help if people could ensure that they are running at least r555 as in the installer, not any earlier beta version - as Michael is now doing. Thank you.
ID: 1211234 · Report as offensive
Previous · 1 · 2 · 3 · 4 · 5 · 6 · 7 · 8 . . . 14 · Next

Message boards : Number crunching : Lunatics Windows Installer v0.40 release notes


 
©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.