CUDA Processing


log in

Advanced search

Message boards : Number crunching : CUDA Processing

1 · 2 · Next
Author Message
strue87
Send message
Joined: 20 Mar 13
Posts: 13
Credit: 476,694
RAC: 0
United States
Message 1349921 - Posted: 23 Mar 2013, 21:10:43 UTC

I have a laptop with a Core-I7-3630QM and an Nvidia GTX 660M and I've been running SETI on it for a couple of days and the program is downloading SETI@home Enhanced 6.10 (cuda Fermi) tasks that process for about 18 seconds and then lists them complete with a status of Computation error.

Is their something wrong with my system or should I turn off the GPU computing?

Any help is appreciated.

Profile betregerProject donor
Avatar
Send message
Joined: 29 Jun 99
Posts: 2490
Credit: 5,216,220
RAC: 8,247
United States
Message 1349954 - Posted: 23 Mar 2013, 22:43:02 UTC

I have been looking at a 650ti, is there any plan to make these lower cards usable?
____________

Profile trader
Volunteer tester
Send message
Joined: 25 Jun 00
Posts: 126
Credit: 4,968,173
RAC: 0
United States
Message 1349955 - Posted: 23 Mar 2013, 22:50:41 UTC - in response to Message 1349954.

I have been looking at a 650ti, is there any plan to make these lower cards usable?


as far as i know you can use the 650 but you can get a 660 for under $200

Richard HaselgroveProject donor
Volunteer tester
Send message
Joined: 4 Jul 99
Posts: 8631
Credit: 51,469,942
RAC: 49,303
United Kingdom
Message 1349961 - Posted: 23 Mar 2013, 22:56:13 UTC - in response to Message 1349954.

I have been looking at a 650ti, is there any plan to make these lower cards usable?

You can use them now, but you have to set the CUDA_GRID_SIZE_COMPAT environment variable described in NVidia driver problems which cause computation errors.

Either that, or run an optimised application.

Profile arkaynProject donor
Volunteer tester
Avatar
Send message
Joined: 14 May 99
Posts: 3688
Credit: 48,718,551
RAC: 6,716
United States
Message 1349963 - Posted: 23 Mar 2013, 23:00:20 UTC - in response to Message 1349954.

I have been looking at a 650ti, is there any plan to make these lower cards usable?


The 650Ti works pretty good, I originally had it in with my 670, but ended up swapping it with the 660 to give my gaming computer a boost.
http://setiathome.berkeley.edu/show_host_detail.php?hostid=5338281
____________

Profile betregerProject donor
Avatar
Send message
Joined: 29 Jun 99
Posts: 2490
Credit: 5,216,220
RAC: 8,247
United States
Message 1349965 - Posted: 23 Mar 2013, 23:01:45 UTC - in response to Message 1349961.

Thanks, now I have to figure out where that is hidden in Windows 8.
____________

Profile trader
Volunteer tester
Send message
Joined: 25 Jun 00
Posts: 126
Credit: 4,968,173
RAC: 0
United States
Message 1349967 - Posted: 23 Mar 2013, 23:11:02 UTC - in response to Message 1349965.

Thanks, now I have to figure out where that is hidden in Windows 8.



goto control panel
select view by large icons (if you have not already)
open system
click on advanced system settings
click on the avanced tab
click Environment variable button
click edit in the system variable section

add the variable name and its value

strue87
Send message
Joined: 20 Mar 13
Posts: 13
Credit: 476,694
RAC: 0
United States
Message 1350026 - Posted: 24 Mar 2013, 5:19:52 UTC - in response to Message 1349923.
Last modified: 24 Mar 2013, 5:24:31 UTC

I followed your thread and entered the CUDA_GRID_SIZE_COMPAT value to 1 and rebooted the system and now the seti screensaver never runs just the BOINC screensaver that says it is loading. I'm not getting any CUDA errors now but it's no longer processing any CUDA blocks at all.

I played around with the screensaver settings and tried the fix #1 by turning off my power savings settings but nothing seems to work as far as getting my GPU processing wu's again.

This Lenovo laptop has 2 video cards, the Nvidia GTX 660M for gaming and the Intel Graphics HD 4000 for everday use, so it may be that the Nvidia card is just no longer running for SETI now with that variable set.

ClaggyProject donor
Volunteer tester
Send message
Joined: 5 Jul 99
Posts: 4139
Credit: 33,500,761
RAC: 23,193
United Kingdom
Message 1350069 - Posted: 24 Mar 2013, 10:13:07 UTC - in response to Message 1350026.

I followed your thread and entered the CUDA_GRID_SIZE_COMPAT value to 1 and rebooted the system and now the seti screensaver never runs just the BOINC screensaver that says it is loading. I'm not getting any CUDA errors now but it's no longer processing any CUDA blocks at all.

I played around with the screensaver settings and tried the fix #1 by turning off my power savings settings but nothing seems to work as far as getting my GPU processing wu's again.

This Lenovo laptop has 2 video cards, the Nvidia GTX 660M for gaming and the Intel Graphics HD 4000 for everday use, so it may be that the Nvidia card is just no longer running for SETI now with that variable set.

Are your CPU Wu's progressing? you haven't reported any since yesterday, Only the CPU app has a screensaver, if there are no Seti CPU Wu's running there will be no Seti Screensaver,
Check you haven't eithier suspended the project, or Suspended Activity in the Activity menu.

Claggy

Profile ivan
Volunteer tester
Avatar
Send message
Joined: 5 Mar 01
Posts: 622
Credit: 143,069,954
RAC: 146,782
United Kingdom
Message 1350112 - Posted: 24 Mar 2013, 12:08:53 UTC - in response to Message 1350040.

This Lenovo laptop has 2 video cards, the Nvidia GTX 660M for gaming and the Intel Graphics HD 4000 for everday use, so it may be that the Nvidia card is just no longer running for SETI now with that variable set.

I am not certain, but I am guessing that the Intel is an onboard GPU and you may have to disable it to get the 660m to work properly for crunching.

It should be OK in Windows. I have a similar set-up with my Dell laptop and it finds the Nvidia card quite easily, for both seti and my digital hologram reconstructions. My problem is when I dual-boot into Linux, then I can only use the onboard Intel graphics.
____________

strue87
Send message
Joined: 20 Mar 13
Posts: 13
Credit: 476,694
RAC: 0
United States
Message 1350200 - Posted: 24 Mar 2013, 16:56:10 UTC - in response to Message 1350069.
Last modified: 24 Mar 2013, 17:02:32 UTC

I followed your thread and entered the CUDA_GRID_SIZE_COMPAT value to 1 and rebooted the system and now the seti screensaver never runs just the BOINC screensaver that says it is loading. I'm not getting any CUDA errors now but it's no longer processing any CUDA blocks at all.

I played around with the screensaver settings and tried the fix #1 by turning off my power savings settings but nothing seems to work as far as getting my GPU processing wu's again.

This Lenovo laptop has 2 video cards, the Nvidia GTX 660M for gaming and the Intel Graphics HD 4000 for everday use, so it may be that the Nvidia card is just no longer running for SETI now with that variable set.

Are your CPU Wu's progressing? you haven't reported any since yesterday, Only the CPU app has a screensaver, if there are no Seti CPU Wu's running there will be no Seti Screensaver,
Check you haven't eithier suspended the project, or Suspended Activity in the Activity menu.

Claggy


All of my CPU wu's have been processed but when I try to update it defers communication for 24 hours. I went to the BOINC website and they have a message posted saying their servers will be down from 3/22-3/24. I'm assuming thats why my computer cant get any work and hasn't uploaded what it's finished.

I also tried downloading Rosetta@home to give my computer some work to do but it wasn't able to download that information from BOINC either.

Profile Floyd
Avatar
Send message
Joined: 19 May 11
Posts: 524
Credit: 1,870,625
RAC: 0
United States
Message 1350203 - Posted: 24 Mar 2013, 17:08:43 UTC - in response to Message 1350200.
Last modified: 24 Mar 2013, 17:15:29 UTC

All of my CPU wu's have been processed but when I try to update it defers communication for 24 hours. I went to the BOINC website and they have a message posted saying their servers will be down from 3/22-3/24. I'm assuming thats why my computer cant get any work and hasn't uploaded what it's finished.

I also tried downloading Rosetta@home to give my computer some work to do but it wasn't able to download that information from BOINC either.

The Boinc web site is NOT where the WU's report to...
The Seti@Home site is where Boinc reports the Wu's that are Seti's WU's to.

The Boinc program on your computer does all the work and then reports to the correct project site.

You may have a network / internet problem.

You have to add Rosetta to the Boinc projects on your computer , like you did seti.
On the top of the boinc manager page click ( tools ) then (add project) then find rosetta in the drop down menu that will open and click rosetta.

____________

Profile WilliamProject donor
Volunteer tester
Avatar
Send message
Joined: 14 Feb 13
Posts: 1610
Credit: 9,469,907
RAC: 44
Message 1350215 - Posted: 24 Mar 2013, 17:51:51 UTC - in response to Message 1350203.
Last modified: 24 Mar 2013, 17:53:11 UTC

All of my CPU wu's have been processed but when I try to update it defers communication for 24 hours. I went to the BOINC website and they have a message posted saying their servers will be down from 3/22-3/24. I'm assuming thats why my computer cant get any work and hasn't uploaded what it's finished.

I also tried downloading Rosetta@home to give my computer some work to do but it wasn't able to download that information from BOINC either.


The Boinc web site is NOT where the WU's report to...
The Seti@Home site is where Boinc reports the Wu's that are Seti's WU's to.
The Boinc program on your computer does all the work and then reports to the correct project site.

You may have a network / internet problem.

You have to add Rosetta to the Boinc projects on your computer , like you did seti.
On the top of the boinc manager page click ( tools ) then (add project) then find rosetta in the drop down menu that will open and click rosetta.

Floyd, please don't use that much bold. It really discouraged me from reading your post. If you highlight everything, the bits you actually want to highlight get lost... it also helps if you'd actually quote the text you are quoting.

see - much better ;)

strue87 - your host contacted the server at 17:35 UTC [i.e. a bit after your last post] and reported a bunch of failed GPU tasks - those are probably the ones you mentioned in your initial post.

We may need to talk you through setting a few debug flags to see what the problem is, but try updating the project again first.

NB Rosetta is CPU only and doesn't allow setting resource share to 0 (to use as true backup project)

As Floyd stated, you don't need the boinc site to be online to communicate with any project. You need that specific project to be online.
____________
A person who won't read has no advantage over one who can't read. (Mark Twain)

strue87
Send message
Joined: 20 Mar 13
Posts: 13
Credit: 476,694
RAC: 0
United States
Message 1350220 - Posted: 24 Mar 2013, 17:58:40 UTC - in response to Message 1350203.

All of my CPU wu's have been processed but when I try to update it defers communication for 24 hours. I went to the BOINC website and they have a message posted saying their servers will be down from 3/22-3/24. I'm assuming thats why my computer cant get any work and hasn't uploaded what it's finished.

I also tried downloading Rosetta@home to give my computer some work to do but it wasn't able to download that information from BOINC either.

The Boinc web site is NOT where the WU's report to...
The Seti@Home site is where Boinc reports the Wu's that are Seti's WU's to.

The Boinc program on your computer does all the work and then reports to the correct project site.

You may have a network / internet problem.

You have to add Rosetta to the Boinc projects on your computer , like you did seti.
On the top of the boinc manager page click ( tools ) then (add project) then find rosetta in the drop down menu that will open and click rosetta.


I did try to add Rosetta through the BOINC manager but on the notices page it had a message saying it was unable to connect. I was up on the web the whole time so the only thing I could think that might have caused it was the 2 changes I tried with the environment variable or the computer sleep fix that was posted.

I saw on another posting that the BOINC 7.0.58 version was a release candidate and had a lot of these fixes in it so I just downloaded that one and installed it and I finally got BOINC communicating again and it's processing CUDA now and says it's using .341 CPUs + 1 NVIDA GPU and is going through each wu in about 5 minutes each.

ClaggyProject donor
Volunteer tester
Send message
Joined: 5 Jul 99
Posts: 4139
Credit: 33,500,761
RAC: 23,193
United Kingdom
Message 1350237 - Posted: 24 Mar 2013, 18:49:34 UTC - in response to Message 1350200.
Last modified: 24 Mar 2013, 18:51:21 UTC

I went to the BOINC website and they have a message posted saying their servers will be down from 3/22-3/24.


That was the 22nd to the 24th February, it's now the 24th March:

BOINC servers down 22-24 February
The BOINC web site will be down from Friday 22 Feb 3 PM (Pacific time) to Sunday 24 Feb noon.21 Feb 2013, 8:29:29 UTC


Claggy

strue87
Send message
Joined: 20 Mar 13
Posts: 13
Credit: 476,694
RAC: 0
United States
Message 1350310 - Posted: 24 Mar 2013, 23:26:24 UTC

I dont know what the problem is but when I set the CUDA_GRID_SIZE_COMPAT value to 1 my BOINC Manager stops communicating with SETI and wont download or upload any work units.

A friend gave me an optimized version called Lunatics_Win64_v0.40_setup and it seems to be working but much more quickly than the estimated times even after running the CPU Benchmark program and my 8 CPU cores are now processing the CUDA_Fermi workunits along with my GPU. The only problem is that BOINC is only sending me 1 or 2 work units at a time and I'm processing 9 at a time counting my GPU so I am about to run out of work again.

I have tried running the Update button to get the schedular to run but it always defers it for 5 minutes and never gets around to running. I also set my BOINC preferences to have 7 days of work on hand but still it wont download more than 1 or 2 wu's at a time. I now only have 5 wu's waiting to start but I will let it keep running and see if it downloads any more.

Profile WilliamProject donor
Volunteer tester
Avatar
Send message
Joined: 14 Feb 13
Posts: 1610
Credit: 9,469,907
RAC: 44
Message 1350326 - Posted: 25 Mar 2013, 0:05:56 UTC - in response to Message 1350310.

I dont know what the problem is but when I set the CUDA_GRID_SIZE_COMPAT value to 1 my BOINC Manager stops communicating with SETI and wont download or upload any work units.

Unless you set a couple of debug flags and post the proceeds, we don't know and can't help either. But I see you've found a way around the problem.
Not much point in troubleshooting you now, and if there's a smoking gun lurking somewhere, we'll get rid of it with V7.

A friend gave me an optimized version called Lunatics_Win64_v0.40_setup and it seems to be working but much more quickly than the estimated times even after running the CPU Benchmark program and my 8 CPU cores are now processing the CUDA_Fermi workunits along with my GPU. The only problem is that BOINC is only sending me 1 or 2 work units at a time and I'm processing 9 at a time counting my GPU so I am about to run out of work again.

Your friend should have known better than to redistribute.
Yes, AKv8b2 is a lot faster and so is x41g.
After you've validated your first 11 units and get APR-driven estimates you should be fine.

I have tried running the Update button to get the schedular to run but it always defers it for 5 minutes and never gets around to running. I also set my BOINC preferences to have 7 days of work on hand but still it wont download more than 1 or 2 wu's at a time. I now only have 5 wu's waiting to start but I will let it keep running and see if it downloads any more.

Server won't let you update in shorter intervalls than 5 minutes.

If your estimates are still too high, boinc will think you have much more work available than you actually have. Just allow it to settle. If you keep 'additional work' small, you'll top up frequently and have a higher chance of keeping your cache full.

NB if you completely run out of tasks (GPu and CPU are handled separately), you may need to do a few manual updates to get something again and thus purge the line.
____________
A person who won't read has no advantage over one who can't read. (Mark Twain)

strue87
Send message
Joined: 20 Mar 13
Posts: 13
Credit: 476,694
RAC: 0
United States
Message 1350363 - Posted: 25 Mar 2013, 2:29:00 UTC - in response to Message 1350326.

I have tried running the Update button to get the schedular to run but it always defers it for 5 minutes and never gets around to running. I also set my BOINC preferences to have 7 days of work on hand but still it wont download more than 1 or 2 wu's at a time. I now only have 5 wu's waiting to start but I will let it keep running and see if it downloads any more.

Server won't let you update in shorter intervalls than 5 minutes.

If your estimates are still too high, boinc will think you have much more work available than you actually have. Just allow it to settle. If you keep 'additional work' small, you'll top up frequently and have a higher chance of keeping your cache full.

NB if you completely run out of tasks (GPu and CPU are handled separately), you may need to do a few manual updates to get something again and thus purge the line.[/quote]

Ok thanks for the info. I will let it run and see if it get's itself figured out. I know the CPU Processors quit trying to solve CUDA cores so that is a plus. I set my BOINC settings to have 10 days of work (the maximum) trying to get it to have something to work on come 1 April.

1 · 2 · Next

Message boards : Number crunching : CUDA Processing

Copyright © 2014 University of California