Author |
Message |
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
I just installed new BOINC server software. The primary feature of the new server is that this should permanently fix the problem of erroneously getting CPU apps when you request GPU apps.
New with this release of the the BOINC server:
- CPU/GPU bug fixed.
- New PPS-Sieve, GFN (short), and GFN (WR) plan-classes. As a result of the new plan-classes, all host computers running these projects will need to download these applications again. In most cases, it's the exact same software as before, only with different names.
- Improved project selection:
- You now select GFN and GFN-WR independently. This allows you to use different shift override values for long and short tasks, and lets you choose CPU and GPU independently.
- If you select CPU tasks but have the "Use CPU" box unchecked, you get a warning that the tasks are disabled. The same applies to CUDA and ATI apps.
- If you turn on "Use CPUs" but don't select any tasks, PPS-LLR is automatically selected for you. Likewise, if you select "Use CUDA" or "Use ATI" and don't select any CUDA or ATI apps (respectively), PPS-Sieve (CUDA or ATI) will be selected for you. This prevents a situation where your host would continuously try to get work without the server being able to deliver any work because nothing is selected.
- New CUDA versions of GeneferCUDA have been installed for Windows and Linux. This is version 3.1.2-1 (2.05 in the BOINC numbering scheme), and improves CUDA error handling. The Mac version will be installed later when it becomes available.
____________
My lucky number is 75898524288+1
|
|
|
|
Nice job Mike! :)
____________
PrimeGrid Challenge Overall standings --- Last update: From Pi to Paddy (2016)
|
|
|
|
Fixing the CPU/GPU bug is great news. Now I will sign onto PPS Sieve more often.
____________
|
|
|
|
Since upgrade one of my computers has not gotten PPS Sieve tasks.
203015 is in the School venue along with 174013. The first computer gets no ATI tasks, but the second one is humming away fine.
Times are UTC:
Project Date Message
PrimeGrid 4/16/2013 22:12:43 update requested by user
PrimeGrid 4/16/2013 22:12:43 Sending scheduler request: Requested by user.
PrimeGrid 4/16/2013 22:12:43 Requesting new tasks for ATI
PrimeGrid 4/16/2013 22:12:45 Scheduler request completed: got 0 new tasks
Last unit received was 16 Apr 2013 | 11:47:46 UTC
Very odd.
____________
My lucky numbers are 121*2^4553899-1 and 3756801695685*2^666669±1
My movie https://vimeo.com/manage/videos/502242 |
|
|
|
UPDATE:
Neither computer is getting the PPS Sieve for ATI. The second computer got one for the CPU (not noticing that right away cause the remote software I use didn't show me it was a CPU task).
So, after the update I got a CPU task not marked.
I am getting not ATI tasks.
I have shut ATI off for now and moved to a different project until someone can take a closer look at this.
____________
My lucky numbers are 121*2^4553899-1 and 3756801695685*2^666669±1
My movie https://vimeo.com/manage/videos/502242 |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
Since upgrade one of my computers has not gotten PPS Sieve tasks.
203015 is in the School venue along with 174013. The first computer gets no ATI tasks, but the second one is humming away fine.
Times are UTC:
Project Date Message
PrimeGrid 4/16/2013 22:12:43 update requested by user
PrimeGrid 4/16/2013 22:12:43 Sending scheduler request: Requested by user.
PrimeGrid 4/16/2013 22:12:43 Requesting new tasks for ATI
PrimeGrid 4/16/2013 22:12:45 Scheduler request completed: got 0 new tasks
Last unit received was 16 Apr 2013 | 11:47:46 UTC
Very odd.
It seems no ATI tasks have gone out, but CPU and CUDA tasks are getting sent just fine.
I don't have an ATI GPU, so I can't test that directly, although I think it was tested on the test server. There's two possibilities:
1) I messed something up setting up the new app versions. I'll double check that it's set up correctly.
2) I messed something up in the plan_class definition for ATI. I may not know enough about ATI to recognize what's wrong, so I'll post the plan class definition here and perhaps somebody will spot something obvious that's wrong.
<plan_class>
<name> atiPPSsieve </name>
<gpu_type> amd </gpu_type>
<cal/>
<need_ati_libs/>
<min_driver_version> 1300000 </min_driver_version>
<min_gpu_ram_mb> 250 </min_gpu_ram_mb>
<gpu_ram_used_mb> 250 </gpu_ram_used_mb>
<gpu_peak_flops_scale> .99 </gpu_peak_flops_scale>
<cpu_frac> .01 </cpu_frac>
<project_prefs_tag> pps_sieve_types </project_prefs_tag>
<project_prefs_regex> ati </project_prefs_regex>
</plan_class>
____________
My lucky number is 75898524288+1 |
|
|
|
Oddly, though I am only getting GFN (short) WUs on my lone nVidia box, they are viewed by external stats sites and the like as "Genefer (World Record)" !
I checked multiple times here and at the machine itself, and they are indeed short tasks, except the calculated runtime is ridiculously long - roughly as long as, funnily enough, the average WR task.
Beyond that minor cosmetic glitch, nice to see everything is running quite smoothly :)
____________
|
|
|
|
Maybe it needs <opencl/> instead of <cal/> as long as AMD OpenCL support required
____________
|
|
|
pvhSend message
Joined: 28 Feb 11 Posts: 5 ID: 88694 Credit: 231,583,127 RAC: 0
 
|
I am not getting any work for my AMD GPU either. Maybe <gpu_type> needs to be ati instead of amd?? |
|
|
|
I don't get new PPS Sieve tasks for a nVidia GPU.
Mi 17 Apr 2013 10:40:34 CEST | PrimeGrid | [sched_op] Starting scheduler request
Mi 17 Apr 2013 10:40:34 CEST | PrimeGrid | Sending scheduler request: Requested by user.
Mi 17 Apr 2013 10:40:34 CEST | PrimeGrid | Requesting new tasks for NVIDIA GPU
Mi 17 Apr 2013 10:40:34 CEST | PrimeGrid | [sched_op] CPU work request: 0.00 seconds; 0.00 CPUs
Mi 17 Apr 2013 10:40:34 CEST | PrimeGrid | [sched_op] NVIDIA GPU work request: 43200.86 seconds; 1.00 GPUs
Mi 17 Apr 2013 10:40:35 CEST | PrimeGrid | Scheduler request completed: got 0 new tasks
Mi 17 Apr 2013 10:40:35 CEST | PrimeGrid | [sched_op] Server version 701
Mi 17 Apr 2013 10:40:35 CEST | PrimeGrid | Project requested delay of 7 seconds
Mi 17 Apr 2013 10:40:35 CEST | PrimeGrid | [sched_op] Deferring communication for 7 sec
Mi 17 Apr 2013 10:40:35 CEST | PrimeGrid | [sched_op] Reason: requested by project
Last wu I got is this. |
|
|
|
See this on the task page you linked?
<core_client_version>6.12.34</core_client_version>
< Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
See this on the task page you linked?
<core_client_version>6.12.34</core_client_version>
< Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
I think the ATI bug is understood now. There is (another) bug in the BOINC scheduler code. Fortunately, this one will be easy to fix (which I'll do a little later), but even better, I was able to work around the bug by tweaking the plan_class to circumvent the problem.
Work should be flowing to ATI GPUs again.
Let me know if there's any (more) problems. Special thanks to everyone who helped with diagnosing the problem.
____________
My lucky number is 75898524288+1 |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
Maybe it needs <opencl/> instead of <cal/> as long as AMD OpenCL support required
Maybe. It's working right now, but at the moment I'm requiring neither CAL nor OpenCL support so it will send work to any ATI card. (It also worked just fine with cal/.) I'll dig down and set the requirements more appropriately at a later time. (I.e., I'll ask Ken, "What does this program need to run?")
____________
My lucky number is 75898524288+1 |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
Thanks for your response.
Though I didn't find a notice so far about the changed requirements to run this project.
That's because the requirements didn't change. There was a bug in the new server software that prevented ATI tasks from being sent out. Are you able to get work now?
____________
My lucky number is 75898524288+1 |
|
|
|
Thanks for your response.
Though I didn't find a notice so far about the changed requirements to run this project.
My answer was to the response from NullCoding. You were faster than me typing the message. So it's a bit confusing.
That's because the requirements didn't change. There was a bug in the new server software that prevented ATI tasks from being sent out. Are you able to get work now?
No, the host doesn't get work.
Important to say the host has a nVidia gpu (8600 GT), not an ATI card. |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
No, the host doesn't get work.
Important to say the host has a nVidia gpu (8600 GT), not an ATI card.
Ok, found that problem. The default memory requirement for CUDA is 384MB and your 8600 GT has 256MB. I'm going to lower the requirement to 180 MB and that should fix the problem.
Thanks for letting me know about the problem!
Let me know if that computer still can't get work.
____________
My lucky number is 75898524288+1 |
|
|
|
No, the host doesn't get work.
Important to say the host has a nVidia gpu (8600 GT), not an ATI card.
Ok, found that problem. The default memory requirement for CUDA is 384MB and your 8600 GT has 256MB. I'm going to lower the requirement to 180 MB and that should fix the problem.
Thanks for letting me know about the problem!
Let me know if that computer still can't get work.
Problem solved. Host is getting work.
Thank you! |
|
|
|
I'm not sure if this is a client problem or not.
Since updating to the new plan classes, all (cudaPPSieve) WUs coming into my host have an estimated time to completion of 07:38:08, and appear to be "stuck" at that figure, which sounds suspiciously like the ETC for a CPU rather than GPU task. My actual processing time on a GTX 580 is in the neighborhood of 12.5~13 minutes. After 24 hours at that rate, I would expect to see the ETC figure decrementing toward reality, but nope, still 07:38:08.
The result is that, to get any kind of cushion against server outage, I have to crank the min/max work buffer parameters way up -- something I don't like to do.
I've seen this sort of thing before on another (CPU) project, but I can't for the life of me remember what the solution was.
____________
|
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
I'm not sure if this is a client problem or not.
Since updating to the new plan classes, all (cudaPPSieve) WUs coming into my host have an estimated time to completion of 07:38:08,
I'll look into it, but no promises on how long it will take. I want to make sure work is flowing out correctly first -- then I'll take a look at the timing parameters and see if anything is not right. The thing is so messed up due to the single-DCF problem that sometimes it's hard to tell if it's broken or not -- or even if it makes a difference if it's broken or not.
____________
My lucky number is 75898524288+1 |
|
|
|
It does not appear that I can get any work for my ATI.http://www.primegrid.com/show_host_detail.php?hostid=272106
I am currently running version 7.0.62 (x64) for BOINC. |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
It does not appear that I can get any work for my ATI.http://www.primegrid.com/show_host_detail.php?hostid=272106
I am currently running version 7.0.62 (x64) for BOINC.
Can you post the log from your BOINC client showing the failure to get work? Also please indicate the UTC offset used in the log so I can compare it to the timestamps in the server logs.
____________
My lucky number is 75898524288+1 |
|
|
|
It does not appear that I can get any work for my ATI.http://www.primegrid.com/show_host_detail.php?hostid=272106
I am currently running version 7.0.62 (x64) for BOINC.
Can you post the log from your BOINC client showing the failure to get work? Also please indicate the UTC offset used in the log so I can compare it to the timestamps in the server logs.
4/16/2013 5:44:21 PM | PrimeGrid | Sending scheduler request: To fetch work.
4/16/2013 5:44:21 PM | PrimeGrid | Requesting new tasks for ATI
4/16/2013 5:44:24 PM | PrimeGrid | Scheduler request completed: got 0 new tasks
4/16/2013 5:46:41 PM | PrimeGrid | Sending scheduler request: Requested by user.
4/16/2013 5:46:41 PM | PrimeGrid | Requesting new tasks for ATI
4/16/2013 5:46:43 PM | PrimeGrid | Scheduler request completed: got 0 new tasks
4/17/2013 8:57:56 PM | PrimeGrid | Sending scheduler request: To fetch work.
4/17/2013 8:57:56 PM | PrimeGrid | Requesting new tasks for ATI
4/17/2013 8:58:00 PM | PrimeGrid | Scheduler request completed: got 0 new tasks
4/17/2013 10:20:14 PM | PrimeGrid | Sending scheduler request: To fetch work.
4/17/2013 10:20:14 PM | PrimeGrid | Requesting new tasks for ATI
4/17/2013 10:20:16 PM | PrimeGrid | Scheduler request completed: got 0 new tasks
UTC offset -6:00 (Mountain Daylight time) |
|
|
|
I'm not sure if this is a client problem or not.
Since updating to the new plan classes, all (cudaPPSieve) WUs coming into my host have an estimated time to completion of 07:38:08,
I'll look into it...
Just to add that BOINC does a similar thing for me, except my ETC does rapidly decline to zero. My run times don't seem to be affected. It's happening on both my 560Ti and my 580, on the 560 the actual run time for one WU is just under 29 minutes, but the ETC is still starting at 8 hours 40 mins. For my 580 the times are obviously similar to jjwhalen's. I too assumed that BOINC would adapt after a few hours of the server upgrade, but it hasn't.
I realise that it's hardly the biggest problem on your hands, just letting you know that it's not a one-off. |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
bones,
I changed something -- did that fix the problem?
____________
My lucky number is 75898524288+1 |
|
|
|
bones,
I changed something -- did that fix the problem?
4/18/2013 6:54:20 AM | PrimeGrid | Sending scheduler request: To fetch work.
4/18/2013 6:54:20 AM | PrimeGrid | Requesting new tasks for ATI
4/18/2013 6:54:23 AM | PrimeGrid | Scheduler request completed: got 2 new tasks
Yes, that seemed to help. Thank you. |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
bones,
I changed something -- did that fix the problem?
4/18/2013 6:54:20 AM | PrimeGrid | Sending scheduler request: To fetch work.
4/18/2013 6:54:20 AM | PrimeGrid | Requesting new tasks for ATI
4/18/2013 6:54:23 AM | PrimeGrid | Scheduler request completed: got 2 new tasks
Yes, that seemed to help. Thank you.
Ok, that's good.
For everyone else, the change I made was specific for bones and does NOT apply to anyone else. There was something strange about his user preferences -- I can't explain how they got set the way they did, but it was easy to fix. (If you have the same problem, setting your project preferences again, i.e., selecting which projects to run, should fix the problem.)
So if you're not getting tasks and you think you should, please speak up and I'll see what I can do to remedy the problem.
____________
My lucky number is 75898524288+1 |
|
|
Ken_g6 Volunteer developer
 Send message
Joined: 4 Jul 06 Posts: 929 ID: 3110 Credit: 236,594,205 RAC: 10,765
                           
|
Maybe it needs <opencl/> instead of <cal/> as long as AMD OpenCL support required
Maybe. It's working right now, but at the moment I'm requiring neither CAL nor OpenCL support so it will send work to any ATI card. (It also worked just fine with cal/.) I'll dig down and set the requirements more appropriately at a later time. (I.e., I'll ask Ken, "What does this program need to run?")
And Ken will answer that this program needs OpenCL to run. I'm not exactly sure of the version, but it's fairly old. 4000-series GPUs that support OpenCL should continue to work.
____________
|
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
And Ken will answer that this program needs OpenCL to run. I'm not exactly sure of the version, but it's fairly old. 4000-series GPUs that support OpenCL should continue to work.
Excellent.
I'l set it up so it doesn't have any requirements other than OpenCL. If there's any higher requirements (a specific minimum driver, for example) I can add those is when they're recognized.
____________
My lucky number is 75898524288+1 |
|
|
|
I'm not sure if this is a client problem or not.
Since updating to the new plan classes, all (cudaPPSieve) WUs coming into my host have an estimated time to completion of 07:38:08,
I'll look into it, but no promises on how long it will take. I want to make sure work is flowing out correctly first -- then I'll take a look at the timing parameters and see if anything is not right. The thing is so messed up due to the single-DCF problem that sometimes it's hard to tell if it's broken or not -- or even if it makes a difference if it's broken or not.
when you do get to it can you please put the cuda CPU count back to .22? At .8 and with 2 GPUs it means boinc stops a CPU task to let the GPU run. If I tell the host to use an additional CPU (in my case 7 rather than 6) then while it will run the 6 cpu tasks I want it to it also fetches a 7th task it can't run. |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
I'm not sure if this is a client problem or not.
Since updating to the new plan classes, all (cudaPPSieve) WUs coming into my host have an estimated time to completion of 07:38:08,
I'll look into it, but no promises on how long it will take. I want to make sure work is flowing out correctly first -- then I'll take a look at the timing parameters and see if anything is not right. The thing is so messed up due to the single-DCF problem that sometimes it's hard to tell if it's broken or not -- or even if it makes a difference if it's broken or not.
Thanks to some information from x3mEn and JimB, it seems like this is a 'feature' -- and it might turn out to be a really nice feature.
(Caveat: this is new to me, and there's undoubtedly stuff I don't know about yet. So all of this may be completely wrong...)
You know how we have been complaining about how a single DCF just doesn't work when you have so many wildly different apps? Berkeley may have fixed that. At least they're attempting to fix it. This is very good news.
They have done away with DCF. If you're running 7.0.28 or later on your computer, and the server is a relatively new version (as ours is now), DCF is turned off. It's replaced by a system that tracks performance for each app version. Not each app, but each version of each app. So, there's separate tracking for, say, GFN 2.04 on the CPU, 2.04 on CUDA, and 2.05 on CUDA.
If this works, it's very good news. But like DCF, you'll have to be patient and wait until enough tasks are run for the server to get an accurate estimate of how long tasks take to run on your hardware. Unlike DCF, however, this information is stored on the server, not the host, so if you want to tweak the DCF manually, that's no longer possible.
Here's a link to the BOINC page describing the new system, for anyone who is interested:
http://boinc.berkeley.edu/trac/wiki/RuntimeEstimation
____________
My lucky number is 75898524288+1 |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
Upon further investigation, things are a lot more complicated. It may take a while to get the estimated time code working properly, or working at all.
____________
My lucky number is 75898524288+1 |
|
|
|
So if you're not getting tasks and you think you should, please speak up and I'll see what I can do to remedy the problem.
I have 3 systems with 4000 series ATI cards that have not been able to get any PPS Seive. I have reselected the projects for them, 3 different venues, and two have been re-booted as well with no luck. |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
So if you're not getting tasks and you think you should, please speak up and I'll see what I can do to remedy the problem.
I have 3 systems with 4000 series ATI cards that have not been able to get any PPS Seive. I have reselected the projects for them, 3 different venues, and two have been re-booted as well with no luck.
Could you be a little more specific than "4000 series"? What's the exact model number or numbers, how much ram do they have, and what's the version number of the drivers? I'm assuming you *could* get tasks before, right?
Please give me the host ids of the machines, and leave all three machines trying to download ATI PPS-sieve apps.
To see what's happening, I need to turn on logging that eats up a ton of disk space, so it's important that your computer(s) are actively trying to download tasks.
____________
My lucky number is 75898524288+1 |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
Found out what's happening, sort of.
Here's the important part of the log of one of your computers requesting work:
GPU RAM required min: 262144000.000000, supplied: 0.000000
For some reason, the request your computer is sending is indicating that you don't have any video memory. This isn't happening for other users. I'll dig further...
____________
My lucky number is 75898524288+1 |
|
|
|
I'm not sure if this is a client problem or not.
Since updating to the new plan classes, all (cudaPPSieve) WUs coming into my host have an estimated time to completion of 07:38:08, and appear to be "stuck" at that figure, which sounds suspiciously like the ETC for a CPU rather than GPU task. My actual processing time on a GTX 580 is in the neighborhood of 12.5~13 minutes. After 24 hours at that rate, I would expect to see the ETC figure decrementing toward reality, but nope, still 07:38:08.
The result is that, to get any kind of cushion against server outage, I have to crank the min/max work buffer parameters way up -- something I don't like to do.
I've seen this sort of thing before on another (CPU) project, but I can't for the life of me remember what the solution was.
I have the same problem, it wouldn't worry me except like the poster instead of a cache of 30 or 40 WU's I now get only 1 or 2. Even the slightest bump in the scheduler would have mine & others GPU's running dry.
I also have seen a fix for this problem before by editing a file and resetting it back to a more normal setting, but don't remember the details of how to do it either. Anyone know how? |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
I also have seen a fix for this problem before by editing a file and resetting it back to a more normal setting, but don't remember the details of how to do it either. Anyone know how?
If you're talking about manually setting the DCF value, that will not work anymore if you're using BOINC client 7.0.28 or later.
We're actively working on fixing the problem.
____________
My lucky number is 75898524288+1 |
|
|
|
So if you're not getting tasks and you think you should, please speak up and I'll see what I can do to remedy the problem.
I have 3 systems with 4000 series ATI cards that have not been able to get any PPS Seive. I have reselected the projects for them, 3 different venues, and two have been re-booted as well with no luck.
Could you be a little more specific than "4000 series"? What's the exact model number or numbers, how much ram do they have, and what's the version number of the drivers? I'm assuming you *could* get tasks before, right?
Please give me the host ids of the machines, and leave all three machines trying to download ATI PPS-sieve apps.
To see what's happening, I need to turn on logging that eats up a ton of disk space, so it's important that your computer(s) are actively trying to download tasks.
2 systems have ATI HD4350 with 512MB ram Driver 1.4.900
Host IDs are
67274
88450
1 has ATI HD4650 with 1GB ram Driver 1.4.1664
Host ID
46638
All three were getting tasks fine before. |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
So if you're not getting tasks and you think you should, please speak up and I'll see what I can do to remedy the problem.
I have 3 systems with 4000 series ATI cards that have not been able to get any PPS Seive. I have reselected the projects for them, 3 different venues, and two have been re-booted as well with no luck.
Could you be a little more specific than "4000 series"? What's the exact model number or numbers, how much ram do they have, and what's the version number of the drivers? I'm assuming you *could* get tasks before, right?
Please give me the host ids of the machines, and leave all three machines trying to download ATI PPS-sieve apps.
To see what's happening, I need to turn on logging that eats up a ton of disk space, so it's important that your computer(s) are actively trying to download tasks.
2 systems have ATI HD4350 with 512MB ram Driver 1.4.900
Host IDs are
67274
88450
1 has ATI HD4650 with 1GB ram Driver 1.4.1664
Host ID
46638
All three were getting tasks fine before.
Following up on my previous reply, the request your computers are sending doesn't have the <available_ram> tag that I see in other requests. That may be the reason the server thinks your GPUs don't have sufficient video memory. It's possible the new server code is looking for this tag, which may only come from newer clients. You're running 6.10.58, which is sort of old. If you wish, you could try upgrading one of the computers to a newer client, but I'm hoping that's not required.
____________
My lucky number is 75898524288+1 |
|
|
Honza Volunteer moderator Volunteer tester Project scientist Send message
Joined: 15 Aug 05 Posts: 1931 ID: 352 Credit: 5,702,802,045 RAC: 1,040,148
                                   
|
Here's a link to the BOINC page describing the new system, for anyone who is interested:
http://boinc.berkeley.edu/trac/wiki/RuntimeEstimation
Dealing with different runtime and it's estimation is aiming toward giving proper amount of work based on preferences and finish it before deadline.
Another feature worth attention and could be useful for sieving (where jobsize can be adjusted)
The difference in throughput between a slow processor (e.g. an Android device that runs infrequently) and a fast processor (e.g. a GPU that's always on) can be a factor of 1,000 or more. Having a single job size can therefore present problems:
If the size is small, hosts with GPUs get huge numbers of jobs. This causes performance problems on the client and a high DB load on the server.
If the size is large, slow hosts can't get jobs, or they get jobs that take weeks to finish.
http://boinc.berkeley.edu/trac/wiki/JobSizeMatching
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186 |
|
|
|
So if you're not getting tasks and you think you should, please speak up and I'll see what I can do to remedy the problem.
I have 3 systems with 4000 series ATI cards that have not been able to get any PPS Seive. I have reselected the projects for them, 3 different venues, and two have been re-booted as well with no luck.
Could you be a little more specific than "4000 series"? What's the exact model number or numbers, how much ram do they have, and what's the version number of the drivers? I'm assuming you *could* get tasks before, right?
Please give me the host ids of the machines, and leave all three machines trying to download ATI PPS-sieve apps.
To see what's happening, I need to turn on logging that eats up a ton of disk space, so it's important that your computer(s) are actively trying to download tasks.
2 systems have ATI HD4350 with 512MB ram Driver 1.4.900
Host IDs are
67274
88450
1 has ATI HD4650 with 1GB ram Driver 1.4.1664
Host ID
46638
All three were getting tasks fine before.
Following up on my previous reply, the request your computers are sending doesn't have the <available_ram> tag that I see in other requests. That may be the reason the server thinks your GPUs don't have sufficient video memory. It's possible the new server code is looking for this tag, which may only come from newer clients. You're running 6.10.58, which is sort of old. If you wish, you could try upgrading one of the computers to a newer client, but I'm hoping that's not required.
That's interesting since my machine using an NVIDA GTX480 is using the same client and it doesn't seem to have any trouble getting them.
I'll see about upgrading one this weekend and see how it goes. |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
So if you're not getting tasks and you think you should, please speak up and I'll see what I can do to remedy the problem.
I have 3 systems with 4000 series ATI cards that have not been able to get any PPS Seive. I have reselected the projects for them, 3 different venues, and two have been re-booted as well with no luck.
Could you be a little more specific than "4000 series"? What's the exact model number or numbers, how much ram do they have, and what's the version number of the drivers? I'm assuming you *could* get tasks before, right?
Please give me the host ids of the machines, and leave all three machines trying to download ATI PPS-sieve apps.
To see what's happening, I need to turn on logging that eats up a ton of disk space, so it's important that your computer(s) are actively trying to download tasks.
2 systems have ATI HD4350 with 512MB ram Driver 1.4.900
Host IDs are
67274
88450
1 has ATI HD4650 with 1GB ram Driver 1.4.1664
Host ID
46638
All three were getting tasks fine before.
Following up on my previous reply, the request your computers are sending doesn't have the <available_ram> tag that I see in other requests. That may be the reason the server thinks your GPUs don't have sufficient video memory. It's possible the new server code is looking for this tag, which may only come from newer clients. You're running 6.10.58, which is sort of old. If you wish, you could try upgrading one of the computers to a newer client, but I'm hoping that's not required.
That's interesting since my machine using an NVIDA GTX480 is using the same client and it doesn't seem to have any trouble getting them.
I'll see about upgrading one this weekend and see how it goes.
The Nvidia and ATI processors have very different code in both the boinc server and client, so it's not surprising that this only affects ATI.
What IS surprising is that this isn't working. The server code is clearly written to be backwards compatible with older clients, and if that particular tag is not present it's supposed to use the memory value from a different tag. I'm not yet sure why this isn't working.
____________
My lucky number is 75898524288+1 |
|
|
|
Now the ATI cards a sporadically getting work. Most of time time a request happens work is given, but there are times when no work is given and it takes a while to finally get work.
This has happened on both my hosts that have ATI cards.
____________
My lucky numbers are 121*2^4553899-1 and 3756801695685*2^666669±1
My movie https://vimeo.com/manage/videos/502242 |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
Now the ATI cards a sporadically getting work. Most of time time a request happens work is given, but there are times when no work is given and it takes a while to finally get work.
This has happened on both my hosts that have ATI cards.
You get "no work available", and then on the next try get work?
____________
My lucky number is 75898524288+1 |
|
|
|
You get "no work available", and then on the next try get work?
Requesting new tasks for ATI
Scheduler request completed: got 0 new tasks
It can happen from what I have so far observed, 1 to 5 times when it happens.
____________
My lucky numbers are 121*2^4553899-1 and 3756801695685*2^666669±1
My movie https://vimeo.com/manage/videos/502242 |
|
|
|
I also have seen a fix for this problem before by editing a file and resetting it back to a more normal setting, but don't remember the details of how to do it either. Anyone know how?
If you're talking about manually setting the DCF value, that will not work anymore if you're using BOINC client 7.0.28 or later.
We're actively working on fixing the problem.
Thanks Michael, yes that was what I was referring to. I didn't realize it wasn't possible anymore.
Thanks for the info JN |
|
|
|
Still having an issue keeping the ATI cards continuously working on PPS Sieve. It happens very randomly and most times hand asking for new work will get it, but as I have stated before sometimes I have to ask more than once, but usually no need to go more than 2-3 times. I did have to ask 5 times once.
I know other things are under priority, but just letting you know this is still an issue.
____________
My lucky numbers are 121*2^4553899-1 and 3756801695685*2^666669±1
My movie https://vimeo.com/manage/videos/502242 |
|
|
|
I am not getting work on my AMD HD 7950.
4/22/2013 2:08:40 | | ATI GPU 0: AMD Radeon HD 79x0 series (Tahiti) (CAL version 1.4.1741, 3072MB, 3032MB available, 9856 GFLOPS peak)
4/22/2013 2:08:40 | PrimeGrid | Application uses missing CUDA GPU
4/22/2013 2:08:40 | | Config: report completed tasks immediately
4/22/2013 2:08:40 | | Config: use all coprocessors
4/22/2013 2:08:51 | PrimeGrid | Sending scheduler request: Requested by user.
4/22/2013 2:08:51 | PrimeGrid | Requesting new tasks for ATI
4/22/2013 2:08:53 | PrimeGrid | Scheduler request completed: got 0 new tasks
I was successfully getting tasks until recently. I'm not sure why it's saying that it's missing a CUDA GPU when my task preferences don't ask for CUDA work. |
|
|
|
I'm not sure why it's saying that it's missing a CUDA GPU when my task preferences don't ask for CUDA work.
Possibly you are using app_info.xml, which contains CUDA app description.
____________
|
|
|
|
I searched my computer for that file, and no such file exists.
Additionally, I don't think that would explain why it was fetching work properly until recently. |
|
|
|
It seems BOINC is not able to detect OpenCL driver and falsely reports about missing CUDA GPU.
Could you please tell the version of BOINC client you are using?
In any case you need to achieve the similar BOINC log:
04-Sep-2012 08:11:25 [---] ATI GPU 0: AMD Radeon HD 7700 series (Cape Verde) (CAL version 1.4.1741, 1024MB, 984MB available, 1126 GFLOPS peak)
04-Sep-2012 08:11:25 [---] OpenCL: ATI GPU 0: AMD Radeon HD 7700 series (Cape Verde) (driver version CAL 1.4.1741 (VM), device version OpenCL 1.2 AMD-APP (938.2), 1024MB, 984MB available)
Finally you can try BOINC development version from here, if nothing helps.
____________
|
|
|
|
I was using 7.0.28. I switched to the development version, and now BOINC does properly detect it as an OpenCL device. However, it still doesn't get new tasks:
4/22/2013 10:31:24 | | CAL: ATI GPU 0: AMD Radeon HD 7870/7950/7970 series (Tahiti) (CAL version 1.4.1741, 3072MB, 3032MB available, 9856 GFLOPS peak)
4/22/2013 10:31:24 | | OpenCL: AMD/ATI GPU 0: AMD Radeon HD 7870/7950/7970 series (Tahiti) (driver version 1124.2 (VM), device version OpenCL 1.2 AMD-APP (1124.2), 3072MB, 3032MB available, 9856 GFLOPS peak)
4/22/2013 10:31:24 | | app version refers to missing GPU type CUDA
4/22/2013 10:31:24 | PrimeGrid | Application uses missing CUDA GPU
4/22/2013 10:31:24 | | Config: report completed tasks immediately
4/22/2013 10:31:24 | | Config: use all coprocessors
4/22/2013 10:31:24 | | Version change (7.0.28 -> 7.0.64)
4/22/2013 10:31:49 | PrimeGrid | Sending scheduler request: Requested by user.
4/22/2013 10:31:49 | PrimeGrid | Requesting new tasks for ATI
4/22/2013 10:31:50 | PrimeGrid | Scheduler request completed: got 0 new tasks
|
|
|
|
Actually, after trying an update multiple times, I am observing the behavior pooh_bear is observing: I have to request an update multiple times. For my one sample point it was three times. |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
Actually, after trying an update multiple times, I am observing the behavior pooh_bear is observing: I have to request an update multiple times. For my one sample point it was three times.
I'm not sure (yet) if the problem is a bug or just tuning the server's parameters. I've adjusted some parameters related to handing out jobs. Please post if you notice any changes in behavior.
____________
My lucky number is 75898524288+1 |
|
|
|
4/22/2013 10:31:24 | | app version refers to missing GPU type CUDA
4/22/2013 10:31:24 | PrimeGrid | Application uses missing CUDA GPU
I'm not sure, but usually this message appears when your app_info.xml contains application for missing GPU type.
Please check is there app_info.xml in .\projects\www.primegrid.com\ folder?
____________
|
|
|
|
Tried again, had to request four times.
Also, there is no app_info.xml in \projects\www.primegrid.com\. There aren't any xml files in there. |
|
|
|
Seems to be worse.
____________
My lucky numbers are 121*2^4553899-1 and 3756801695685*2^666669±1
My movie https://vimeo.com/manage/videos/502242 |
|
|
Honza Volunteer moderator Volunteer tester Project scientist Send message
Joined: 15 Aug 05 Posts: 1931 ID: 352 Credit: 5,702,802,045 RAC: 1,040,148
                                   
|
Out of curiosity, I've enabled ATI work - and got work on BOINC 7.0.60.
Notice blue section in log below, those "empty" scheduler request are repeating ad infinitum.
Network usage only 0.01 days, single PPS sieve task is running with remaining estimation of couple of hours (will actually take ~30 minutes on HD7950 while watching HD movie).
22/04/2013 21:08:54 | PrimeGrid | Sending scheduler request: To fetch work.
22/04/2013 21:08:54 | PrimeGrid | Requesting new tasks for CPU and ATI
22/04/2013 21:08:56 | PrimeGrid | Scheduler request completed: got 2 new tasks
22/04/2013 21:08:58 | PrimeGrid | Started download of primegrid_tpsieve_1.39_windows_intelx86__atiPPSsieve.exe
22/04/2013 21:08:58 | PrimeGrid | Started download of LLR_SGS_182750131
22/04/2013 21:08:59 | PrimeGrid | Finished download of LLR_SGS_182750131
22/04/2013 21:09:00 | PrimeGrid | Finished download of primegrid_tpsieve_1.39_windows_intelx86__atiPPSsieve.exe
22/04/2013 21:09:00 | PrimeGrid | Starting task pps_sr2sieve_r_216891957_0 using pps_sr2sieve version 139 (atiPPSsieve) in slot 6
22/04/2013 21:09:06 | PrimeGrid | Sending scheduler request: To fetch work.
22/04/2013 21:09:06 | PrimeGrid | Not requesting tasks
22/04/2013 21:09:08 | PrimeGrid | Scheduler request completed
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186 |
|
|
|
Just a point to possibly help in finding out what is wrong.
I have an ATI that is getting work OK, but the CPU contribution has increased from 6% to 50%. Nothing else of note has been seen here so at least one user is quite happy.
____________
Member team AUSTRALIA
My lucky number is 9291*2^1085585+1 |
|
|
|
Has anyone with problems tried the modded AMD driver posted at DistrRTgen?
It has an older OpenCL driver (12.4) as some have had problems with the newer OpenCL drivers there.
https://www.freerainbowtables.com/phpBB3/viewtopic.php?f=4&t=3826
____________
Largest Primes to Date:
As Double Checker: SR5 109208*5^1816285+1 Dgts-1,269,534
As Initial Finder: SR5 243944*5^1258576-1 Dgts-879,713
|
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
Just a point to possibly help in finding out what is wrong.
I have an ATI that is getting work OK, but the CPU contribution has increased from 6% to 50%. Nothing else of note has been seen here so at least one user is quite happy.
You mean BOINC is saying 0.50 of a CPU correct? Not actually using 50% of a core?
____________
My lucky number is 75898524288+1 |
|
|
Honza Volunteer moderator Volunteer tester Project scientist Send message
Joined: 15 Aug 05 Posts: 1931 ID: 352 Credit: 5,702,802,045 RAC: 1,040,148
                                   
|
Add to my previous post.
Excessive scheduler request with no work requesting means that SGS runs dry on CPU, PPS sieve on GPU continues and only when GPU task is finished, new work both for CPU and GPU is being asked and provided.
I'll update to latest 7.0.64 (from 7.0.60) if it changes anything.
____________
My stats
Badge score: 1*1 + 5*1 + 8*3 + 9*11 + 10*1 + 11*1 + 12*3 = 186 |
|
|
|
Since about the time of the upgrade I have gotten no CUDA jobs. 0, zilch, nada. Before that time I always had these. I have changed NOTHING on the machine. Any ideas?
____________
|
|
|
|
Just a point to possibly help in finding out what is wrong.
I have an ATI that is getting work OK, but the CPU contribution has increased from 6% to 50%. Nothing else of note has been seen here so at least one user is quite happy.
You mean BOINC is saying 0.50 of a CPU correct? Not actually using 50% of a core?
Yes, it is what is being reported, but may not be accurate until a few lobs are done. That is why I am quite happy with things.
____________
Member team AUSTRALIA
My lucky number is 9291*2^1085585+1 |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
Since about the time of the upgrade I have gotten no CUDA jobs. 0, zilch, nada. Before that time I always had these. I have changed NOTHING on the machine. Any ideas?
Details, please. OS, CPU, GPU, Boinc client version, computer's host ID number, relevant portion of the boinc log, and WHICH GPU apps you''re asking for (PPS-Sieve, GFN-short, GFN-long). I suspect you're not using app_info, but if you are, the information in there may be relevant, too. Thank you.
____________
My lucky number is 75898524288+1 |
|
|
|
Since about the time of the upgrade I have gotten no CUDA jobs. 0, zilch, nada. Before that time I always had these. I have changed NOTHING on the machine. Any ideas?
you've started fetching PPS Sieve CPU tasks instead of CUDA tasks. Check that your preferences are correct. |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
I have disabled the "don't use DCF" feature in the new server. This should, in theory, allow DCF to be used the way it was previously. That's hardly ideal, but it should produce run time estimates better than what we've been seeing since the server upgrade.
This is intended to be a quick fix and not a permanent solution. Getting the real fix to work is taking longer than expected.
____________
My lucky number is 75898524288+1 |
|
|
|
Since about the time of the upgrade I have gotten no CUDA jobs. 0, zilch, nada. Before that time I always had these. I have changed NOTHING on the machine. Any ideas?
Details, please. OS, CPU, GPU, Boinc client version, computer's host ID number, relevant portion of the boinc log, and WHICH GPU apps you''re asking for (PPS-Sieve, GFN-short, GFN-long). I suspect you're not using app_info, but if you are, the information in there may be relevant, too. Thank you.
Windows 7
Intel Q6700 (Core 2 Quad)
NVIDIA GeForce 8800 GTS (615MB) driver: 19745
BOINC 7.0.28
Computer ID 167402
Log (most recent, but they're all the same) -
4/26/2013 9:01:15 PM | PrimeGrid | Requesting new tasks for NVIDIA
4/26/2013 9:01:16 PM | PrimeGrid | Scheduler request completed: got 0 new tasks
Nothing has changed at my end in months. No settings, nothing. |
|
|
|
Nothing has changed at my end in months. No settings, nothing.
Take a look at your primegrid preferences. As mentioned in my previous message you are fetching PPS Sieve CPU instead of GPU. |
|
|
|
I have disabled the "don't use DCF" feature in the new server. This should, in theory, allow DCF to be used the way it was previously.
It's working. |
|
|
|
I have disabled the "don't use DCF" feature in the new server. This should, in theory, allow DCF to be used the way it was previously.
It's working.
Ditto here; estimated times are coming down & work reserves are going up :)
|
|
|
Dave  Send message
Joined: 13 Feb 12 Posts: 3062 ID: 130544 Credit: 2,114,592,724 RAC: 1,432,765
                      
|
Complete dittoness. |
|
|
|
Nothing has changed at my end in months. No settings, nothing.
Take a look at your primegrid preferences. As mentioned in my previous message you are fetching PPS Sieve CPU instead of GPU.
Thanks. I wonder how that got changed. Oh well. It's fixed now. |
|
|
|
Just to confirm what's happening with my ETA/ETCs of PPS Sieve WUs (these were taken about 3 minutes apart, same client/PC, Core 2Quad+GTX 560Ti)
Again though, it doesn't matter at all - aside from possibly scaring people new to PG off by the estimated times. Just clarifying what happened after the server upgrade and is still happening now.
I would also prefer it if the CPU usage for GPU tasks was put back to something a lot lower than 0.8%, like j.sheridan. If this was a beneficial and deliberate change then obviously I'm happy to live with it; otherwise, could it be put low-down on the list of alterations to be made please? Cheers. |
|
|
|
I'm seeing the same behavior with PPS Sieve tasks. Comes in showing ~22h estimated, finishes in under an hour.
____________
My lucky #: 60133106^131072+1 (GFN 17-mega) |
|
|
Michael Goetz Volunteer moderator Project administrator
 Send message
Joined: 21 Jan 10 Posts: 13804 ID: 53948 Credit: 345,369,032 RAC: 4,797
                              
|
I'm seeing the same behavior with PPS Sieve tasks. Comes in showing ~22h estimated, finishes in under an hour.
If you're only running one PrimeGrid project, now that DCF is turned back on, over time the estimates will return to normal. If you're running more than one PrimeGrid project, estimates will return to what passes for normal when using DCF with dissimilar multiple PrimeGrid projects.
Eventually I hope to get the new time estimate system working, but there's no ETA for that right now.
____________
My lucky number is 75898524288+1 |
|
|
|
The issue of having to request an update multiple times to get tasks seems to have been resolved for me. |
|
|
|
So if you're not getting tasks and you think you should, please speak up and I'll see what I can do to remedy the problem.
I have 3 systems with 4000 series ATI cards that have not been able to get any PPS Seive. I have reselected the projects for them, 3 different venues, and two have been re-booted as well with no luck.
Could you be a little more specific than "4000 series"? What's the exact model number or numbers, how much ram do they have, and what's the version number of the drivers? I'm assuming you *could* get tasks before, right?
Please give me the host ids of the machines, and leave all three machines trying to download ATI PPS-sieve apps.
To see what's happening, I need to turn on logging that eats up a ton of disk space, so it's important that your computer(s) are actively trying to download tasks.
2 systems have ATI HD4350 with 512MB ram Driver 1.4.900
Host IDs are
67274
88450
1 has ATI HD4650 with 1GB ram Driver 1.4.1664
Host ID
46638
All three were getting tasks fine before.
Following up on my previous reply, the request your computers are sending doesn't have the <available_ram> tag that I see in other requests. That may be the reason the server thinks your GPUs don't have sufficient video memory. It's possible the new server code is looking for this tag, which may only come from newer clients. You're running 6.10.58, which is sort of old. If you wish, you could try upgrading one of the computers to a newer client, but I'm hoping that's not required.
That's interesting since my machine using an NVIDA GTX480 is using the same client and it doesn't seem to have any trouble getting them.
I'll see about upgrading one this weekend and see how it goes.
The Nvidia and ATI processors have very different code in both the boinc server and client, so it's not surprising that this only affects ATI.
What IS surprising is that this isn't working. The server code is clearly written to be backwards compatible with older clients, and if that particular tag is not present it's supposed to use the memory value from a different tag. I'm not yet sure why this isn't working.
I know it's taken me some time to try this.... Just installed BOINC 7.0.28 and now my ATI 4650 is getting work again. Will upgrade my other ATI machines next. |
|
|