Difficulty with seti@home GPU usage on ubuntu 16.04 LTS under BOINC 7.6.31 (x64)

Message boards : GPUs : Difficulty with seti@home GPU usage on ubuntu 16.04 LTS under BOINC 7.6.31 (x64)
Message board moderation

To post messages, you must log in.

AuthorMessage
Bill Roberts

Send message
Joined: 28 Jul 07
Posts: 7
United States
Message 72010 - Posted: 30 Aug 2016, 22:20:27 UTC

Note: I'm also posing this issue on seti@home

Two issues noted:

1. After reboot and starting BOINC Manager, the preferences (Activity) show all 3 options as "always", instead of "use preferences".

2. Even after setting all 3 options to "use preferences", the seti@home project runs tasks that use an extraordinary amount of GPU cycles, enough to noticeably show down response to the mouse, keyboard, running applications, display and Internet activity. If I select "suspend GPU" the problem stops instantly and remains OK until I allow GPU activity again.

The problem may be with BOINC instead of seti@home, but I'm not willing to connect another project to experiment. What's happening now is bad enough.

I've been a seti@home contributor on dozens of computers since 26 May 1999, so I'd also rather not see answers appropriate to new or inexpert users.
ID: 72010 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15478
Netherlands
Message 72011 - Posted: 30 Aug 2016, 22:40:58 UTC - in response to Message 72010.  

Whose BOINC? Berkeley's, or the one from Ubuntu?

1. It can take a little while for BOINC Manager to populate all settings correctly. So wait a minute, especially when large swats of data have to be read from a slow hard drive.

2. Just setting "use preferences" won't do magic if you didn't set any preferences to follow. So if you don't set (the default) "Suspend GPU computing when computer is in use", you can't expect BOINC to do anything other than give it full blast on that GPU. Same goes for other "suspend when..." and "use at most..." settings.
ID: 72011 · Report as offensive
Profile Agentb
Avatar

Send message
Joined: 30 May 15
Posts: 265
United Kingdom
Message 72012 - Posted: 30 Aug 2016, 22:53:05 UTC - in response to Message 72010.  

Note: I'm also posing this issue on seti@home


I'm sure they'll help, you might mention the error messages you've not noticed in your failing GPU tasks.
ID: 72012 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15478
Netherlands
Message 72013 - Posted: 31 Aug 2016, 1:04:47 UTC - in response to Message 72012.  

I could be wrong, but I don't think one can run CUDA60 (6.0) on a GeForce 8500 GT (511MB).
ID: 72013 · Report as offensive
Bill Roberts

Send message
Joined: 28 Jul 07
Posts: 7
United States
Message 72015 - Posted: 31 Aug 2016, 5:13:08 UTC - in response to Message 72010.  

1. I know how to set preferences and they are set appropriately.
2. There are no "error" messages; the other tasks on the PC just respond extremely slowly.
3. The GPU is available and works. This is a recent development.
ID: 72015 · Report as offensive
Profile Agentb
Avatar

Send message
Joined: 30 May 15
Posts: 265
United Kingdom
Message 72061 - Posted: 31 Aug 2016, 18:55:35 UTC - in response to Message 72013.  

I could be wrong, but I don't think one can run CUDA60 (6.0) on a GeForce 8500 GT (511MB).


I could be wrong too, but i don't think one can see one's task errors if the seti is down.
ID: 72061 · Report as offensive
Bill Roberts

Send message
Joined: 28 Jul 07
Posts: 7
United States
Message 72070 - Posted: 31 Aug 2016, 21:09:55 UTC - in response to Message 72061.  
Last modified: 31 Aug 2016, 21:10:17 UTC

This thread is going off subject. It is not about whether or not a GPU works or generates task error messages.

The GPU works and there are no error messages.

The only issue is that the process is using apparently too many GPU cycles, leading to interference with the other tasks on the computer.
ID: 72070 · Report as offensive
Richard Haselgrove
Volunteer tester
Help desk expert

Send message
Joined: 5 Oct 06
Posts: 5078
United Kingdom
Message 72071 - Posted: 31 Aug 2016, 21:24:46 UTC - in response to Message 72070.  
Last modified: 31 Aug 2016, 21:25:27 UTC

Open BOINC Manager in Advanced view, and look down the 'Application' column.

GPU applications will have a plan_class name in brackets after the version number. Some of these will start with the word 'cuda', others with the word 'opencl'.

Can you correlate whether the acceptable computer behaviour was observed (previously or currently) when one type of application was running the active task, and the interfering behaviour with the other type of application active? If so, which way round are they?

If that doesn't provide a clue, is there a correlation with what you see in the 'name' column for the active GPU task? Some task names start with the the identifier ap_, some with a date in the format ddmmyy, and some with the identifier blc and a number. Again, can you associate any task type with the behaviour you're seeing - which are good, and which are unacceptable?
ID: 72071 · Report as offensive
Profile Agentb
Avatar

Send message
Joined: 30 May 15
Posts: 265
United Kingdom
Message 72072 - Posted: 31 Aug 2016, 21:51:19 UTC - in response to Message 72070.  

This thread is going off subject. It is not about whether or not a GPU works or generates task error messages.

The GPU works and there are no error messages.

The only issue is that the process is using apparently too many GPU cycles, leading to interference with the other tasks on the computer.


Way off topic, the seti has returned.

Even further off topic we find there Bill Robert's believe it or not list of GPU tasks with errors

I may be wrong, perhaps the "out of memory" error messages may suggest something.
ID: 72072 · Report as offensive
Richard Haselgrove
Volunteer tester
Help desk expert

Send message
Joined: 5 Oct 06
Posts: 5078
United Kingdom
Message 72073 - Posted: 31 Aug 2016, 22:14:58 UTC

A further indicator can be found in Application details for host 7321608.

Neither of these applications:

AstroPulse v7 7.08 x86_64-pc-linux-gnu (opencl_nvidia_cc1)
SETI@home v8 8.01 x86_64-pc-linux-gnu (cuda60)

shows any sign of a validated or completed task during the lifetime of the machine: the line "Average processing rate" is missing from both sections.
ID: 72073 · Report as offensive
Bill Roberts

Send message
Joined: 28 Jul 07
Posts: 7
United States
Message 72078 - Posted: 1 Sep 2016, 2:39:49 UTC - in response to Message 72071.  

I see 4 tasks with (cuda60). All are currently "GPU suspended - user request". I have currently suspended all GPU tasks so that I can get other work done, including responding to this thread.

It has been quite some time since I had time to investigate this issue, so I have actually suspended the whole project for a couple of months.

I don't recall this being an issue in the more distant past.
ID: 72078 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15478
Netherlands
Message 72107 - Posted: 1 Sep 2016, 13:11:12 UTC
Last modified: 1 Sep 2016, 13:18:33 UTC

Bill Roberts wrote:
...that use an extraordinary amount of GPU cycles, enough to noticeably show down response to the mouse, keyboard, running applications, display and Internet activity.

Bill Roberts wrote:
I don't recall this being an issue in the more distant past.

Since day one of GPU computing we at BOINC have warned that this will slow down screen draws, mouse movements, keyboard use and possibly other things as well, and on damaged hardware can show artifacts being drawn on the screen. That's because you're putting a lot of strain on the system, more than you would by just using the CPU. And where the CPU can be throttled into using part of the cores, and even then perhaps half of one, with GPU computing this is still a pipe-dream. The GPU can only be used fully (on) or not (off).

The primary use for a GPU is to draw everything on screen, add to that that it then has to do intricate calculations on its shader cores and that will amount to extra heat and slowness of things people want to do on their computer. Which is why from the beginning of GPU computing the default option has been to only use the GPU when the system is idle.

Now, on to the situation at Seti. A peek at the Seti@Home applications teaches us a couple of things:
1. The previous Seti@Home v7 did not have a CUDA application for Nvidia under Linux, only an OpenCL application.
2. Astropulse v7 has got OpenCL applications for Nvidia under Linux, but no CUDA applications.
3. The present Seti@Home v8 has got one CUDA 6.0 application and two OpenCL applications for Nvidia under Linux.
4. Your Nvidia Geforce 8500 GT is rated for a maximum of CUDA 1.3, but has OpenCL 1.0 capability.

With 4. in mind, what does that mean? Well, you may be able to install CUDA 6.0 drivers, but not use CUDA 6.0. The same way as you can at maximum use DirectX 10, but install an operating system with DirectX 11 or 12. The GPU will still only use at max Dx10, so same way here, you install CUDA 6.0 drivers, but the GPU can't do any better than 1.3, it cannot add instructions to its hardware that the newer CUDA versions can do. You need newer hardware that's CUDA 6 compliant for that, which is the Pascal range, Nvidia Titan, GTX 1060-1080 territory.

Seti only has a CUDA6.0 application which it will send to all Nvidia GPUs, whether they can use it or not. I still find that a very strange method, as it would be more prudent to actually check the plan_classes and see if your GPU fits one. In this case it doesn't, and if it doesn't to then send the plain OpenCL application your way as that will always work, as ALL Nvidia GPUs from the first one capable of doing these calculations, have OpenCL 1.0 on board.

What I'd do if I were you, and you want to continue to run work on the GPU when the computer is in use, is to go over to the Seti forums, post in Number Cruching and ask for help on how to set up your system so it runs with the Nvidia OpenCL application and not CUDA 6.0
Whether or not that is done through anonymous platform, or perhaps that it's possible to nudge everything in the right direction is something I cannot answer, but they can.

One thing though, when people ask questions, do them a favour and be so kind to answer. At Seti, best not start off like you did here that you don't want answers towards new and inexperienced users. You've had several people in this thread trying to help, asking a couple of things. We have had no answers on any of the questions, some of us were even rudely told to buzz off because you know what you're doing, when evidently you do not. You're the one asking us for help on an issue you have, so telling people you know better... I can tell you if you keep that up, people will just drop interest and continue with what they were doing before you crossed their paths, while you stay behind and won't be helped.

Or else, just turn your preferences back to "Suspend GPU computing when computer is in use".

Good luck on your further endeavours.
ID: 72107 · Report as offensive
Bill Roberts

Send message
Joined: 28 Jul 07
Posts: 7
United States
Message 72124 - Posted: 1 Sep 2016, 15:33:55 UTC - in response to Message 72073.  

Where can I find more information about "Application Details for host 7321608"?
ID: 72124 · Report as offensive
Bill Roberts

Send message
Joined: 28 Jul 07
Posts: 7
United States
Message 72125 - Posted: 1 Sep 2016, 15:38:38 UTC - in response to Message 72107.  

I had no intent on annoying anyone with this query, so if I have offended anyone, I apologize. This last answer is actually the one I was hoping to get first. It tells me what the issue is and points me in a direction to proceed. Thank you.
ID: 72125 · Report as offensive
Profile Jord
Volunteer tester
Help desk expert
Avatar

Send message
Joined: 29 Aug 05
Posts: 15478
Netherlands
Message 72126 - Posted: 1 Sep 2016, 15:55:36 UTC - in response to Message 72124.  

Where can I find more information about "Application Details for host 7321608"?

When you log in on the Seti website, go to Your Account, look for the entry Computers on this account and click the View link.

You'll get on a personalized version of this page, where you can click on the Details link of your computer, or the Tasks it lists. It's how we gathered our information about your system.
ID: 72126 · Report as offensive
Richard Haselgrove
Volunteer tester
Help desk expert

Send message
Joined: 5 Oct 06
Posts: 5078
United Kingdom
Message 72129 - Posted: 1 Sep 2016, 16:45:57 UTC - in response to Message 72124.  

Or just click on the link I made in message 72073

I'm afraid it's one of the parts of the BOINC web code which is under-documented, but it can be very useful. Here are some of the current details for your computer, in a less fancy format.

SETI@home v8 8.00 x86_64-pc-linux-gnu
Number of tasks completed	333
Max tasks per day		37
Number of tasks today		26
Consecutive valid tasks		6
Average processing rate		15.43 GFLOPS
Average turnaround time		3.51 days

SETI@home v8 8.01 x86_64-pc-linux-gnu (cuda60)
Number of tasks completed	0
Max tasks per day		1
Number of tasks today		8
Consecutive valid tasks		0
Average turnaround time		0.00 days

Let's try to pick apart some of that. I've picked out the two application types where new tasks have been allocated to your computer today - 26 for processing on your CPU at the time I took that copy, 8 tasks for GPU.

The most significant line for assessing the health of your machine is the number 'completed'. These are tasks which ran to a normal end-of-run, and produced validated results, for which you will have received credit. Having zero completed cuda60 tasks is a big alarm signal.

Max tasks per day: BOINC rations the distribution of tasks to hosts which exhibit processing errors. 37 per day is a relatively low limit, suggesting there have been problems recently with your CPU application too, but a limit of one per day for your cuda60 application is another big alarm signal. [Quite why BOINC allocated you 8 cuda60 tasks today, while saying your limit is 1, is another of the mysteries of BOINC we haven't been able to fathom yet.]

Average processing rate: when a task both validates and runs to the full normal exit point (some SETI tasks exit early, by design), the running time is used to assess the effective speed of the application on your device. This figure is the average over many tasks, and helps BOINC to allocate the correct amount of work to keep your machine busy for the length of time you specify in your preferences.

The APR for your CPU - 15.43 GFLOPS - is considerably faster than the "Measured floating point speed 3140.11 million ops/sec" of your computer - a measure of how efficient the SETI applications have become in recent years.

But because every one of your cuda60 tasks so far has ended in an error, BOINC hasn't been able to begin estimating a processing speed yet - the final alarm signal.

By way of comparison, here's a similar display for one of my machines - Application details for host 3755243

SETI@home v8 (anonymous platform, CPU)
Number of tasks completed	837
Max tasks per day		935
Number of tasks today		0
Consecutive valid tasks		903
Average processing rate		36.62 GFLOPS
Average turnaround time		1.71 days

SETI@home v8 (anonymous platform, NVIDIA GPU)
Number of tasks completed	30972
Max tasks per day		22413
Number of tasks today		10
Consecutive valid tasks		22380
Average processing rate		160.57 GFLOPS
Average turnaround time		0.43 days

(I'm winding down at the moment after a friendly challenge between teams, and starting up other projects again, which accounts for the low number of tasks today. But I think you can see why your figures strike me as being low, and worth investigating further.)
ID: 72129 · Report as offensive
Juha
Volunteer developer
Volunteer tester
Help desk expert

Send message
Joined: 20 Nov 12
Posts: 801
Finland
Message 72141 - Posted: 1 Sep 2016, 18:02:21 UTC - in response to Message 72013.  

I could be wrong, but I don't think one can run CUDA60 (6.0) on a GeForce 8500 GT (511MB).


G80 GPUs were only deprecated in CUDA Toolkit 6.0, not yet removed. 6.0 won't be optimal for those GPUs of course.

@Bill

Since the app seems to be bombing out on memory allocations freeing GPU memory is needed. These days web browsers for example likes to do GPU acceleration. Disabling the acceleration might free GPU memory.

I find it odd that you say the app is using too much GPU but in the task list there's only one GPU task with some amount of run time. Maybe the driver is crashing/recovering. Check syslog for any GPU related messages.
ID: 72141 · Report as offensive

Message boards : GPUs : Difficulty with seti@home GPU usage on ubuntu 16.04 LTS under BOINC 7.6.31 (x64)

Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.