Video card not seen?
24 Feb 2016 16:27:48 UTC
Topic 198464
(moderation:
Put a new comp with an R9 Nano, to see how it goes. The thing is the manages doesn't seem to notice it. It runs tasks only on the CPU.
I updated my preferences online, as 'use ATI GPU' was disabled, so I corrected it. And then click 'update' several times, restarted and checked again. Anyone knows why?
Video card not seen?
)
People might have better luck giving you advice if you unhide your computers.
Yap, I don't know why I hid
)
Yap, I don't know why I hid them. They should be public now.
OK, my fault, I installed
)
OK, my fault, I installed Boinc before the AMD Driver:)
But now it's something else: it shows up, but how do I know it actually uses it?
On OpenHardwareMonitor in "Load - GPU Core" it stays at 0%. Isn't it supposed to be 100%? I made all the settings I know to use the graphic card to the full.
[/img]
Does BOINCmgr say it is
)
Does BOINCmgr say it is running a GPU task? And which application (you seem to have fetched an enormous amount of work for many of them)? Does it show the task as progressing? A little benign neglect might be called for right now.
It doesn't show GPU, indeed.
)
It doesn't show GPU, indeed. Many of them are in 'waiting to run' status. Is shows them as progressing, but very small increments. Like 0.003% or 0.004%. It was the same without the GPU (only on the CPU), so I guess it's not ok to have the same step with it (after the reinstall).
The workload settings I made were 3 days work and max another 7 days download and store on HDD. Is it much 3 and 7?
RE: The workload settings I
)
Yes, ten days total is a lot of work, especially since Einstein currently has an important task type issuing with deadlines not longer than seven days.
But the bigger problem is that since your host is new to the game, the productivity estimation is likely to be wildly wrong. The way it figures out how fast your machine is, and what fraction of the time you actually have it running work, is by watching it.
"Waiting to run" is just right for the great majority of your huge stash of work. The tiny increments may just be "simulated" progress guessed by BOINCmgr before it has gotten any actual progress update from the application. Because of this, you may see a sudden transition in claimed progress--which can easily take the form of a big step back.
But the question was: does BOINCmgr show any GPU jobs to be with status of "running"?
After about 100 lines (in
)
After about 100 lines (in advanced view) with 'waiting to run'(few) and 'ready to start' (majority), there's a row with a running task where it says CPU+GPU.
But, the load on GPU Core is 0%. It should be 100%, if I'm not wrong.
RE: After about 100 lines
)
That particular task is attempting to use the GPU. The ones showing 'waiting to run' would seem to have attempted to start but then have been suspended by BOINC for some reason. That's not normal.
Since you have a machine with 8 virtual cores, you would also expect to see 8 'running' CPU tasks. If there are none actually running, that's also not at all normal. You really need to be in Advanced view to regain control of this and the very first thing to do is go to the project tab, click on the Einstein project to select it and click on the 'No new tasks' button just in case subsequent actions prompt BOINC into wanting to ask for more work (which you really don't need) :-).
Then go back to the tasks tab. You will see a number of columns with a header on each. The borders between columns can be dragged to give appropriate widths so you can see as much of the data as you desire. Column headers are clickable so you can sort rows in normal or reverse order, alphabetically or numerically depending on the data in the column. I often click on the progress column so I can see all tasks with greater than 0.000% progress (ie tasks that are actually running) at the top of the list.
How many tasks can you actually find that are running and making progress?
If you find some of these, exactly what does it say (for each one) in the status column?
I will attempt to respond quickly if you can answer the above quickly.
It will never be 100% and it shouldn't be zero. Because of the power of that GPU it may show a fluctuating value averaging somewhere in the 30-80% range, perhaps :-).
The most likely reason it shows 0% is because (for some reason) BOINC can detect it but can't use it. I'm not familiar with BOINC 7.6.22. The highest one I've used is 7.2.42 for Linux. My guess about what is wrong is that you don't have some particular component of the OpenCL libs in the driver package you are using. I only use Linux so I have no understanding of how drivers/libs work under Windows. There may be some clue about this in the BOINC startup messages that you can look at through the event log. If you post those, a Windows user may be able to help.
If no CPU tasks seem to be running, you could try 'suspending' (in BOINC Manager - tasks tab) the bulk of all your tasks, leaving just say about 10 CPU tasks and a couple of GPU tasks and see if that causes any of the remaining ones to start crunching.
You're not attempting to run multiple concurrent GPU tasks by any chance, are you?
EDIT: I just noticed in your tasks list for that machine, you now have a completed and validated BRP4G task. It took a really long time to crunch suggesting that your drivers are working but that you need to not be running CPU tasks on all virtual CPU cores. Try going to the Tools -> Computing preferences -> processor usage tab and setting the number of cores to 87%. This should allow only 7 CPU tasks to run. If that doesn't speed things up, try 75% which should reduce the number to 6 running CPU tasks.
Cheers,
Gary.
It's true, there are 8
)
It's true, there are 8 running tasks. I have modified to 75% (6 cores) and updated it. Those 8 have 'running' in the status column, and they run on CPU (I guess). There's another one (9th) which also has 'running' status, with both CPU+GPU.
It's true about the GPU. The Catalyst window shows exactly that: ups and downs from like 10 to 90%. I was expecting a straight line for that, like the one for the fan's rpm. :)
I guess not :) although I don't know if or how you can control that.
I have to take look at the log file, but I guess the driver is up to date. Would it be better to install an older Boinc manager?
RE: It's true, there are 8
)
OK, you now have a second BRP4G returned, still way too slow but a little faster than the first one. Of the 8 'running' CPU tasks, can you confirm there are now just 6 'running' and 2 'waiting to run'? If there are still 8 running you haven't got the change to 75% of CPU cores to stick. If that change has been successful, the current GPU task should really be flying in comparison to previously. Check the % progress figure for that task. It should be increasing at least 0.015% every single second - say 1% in less than a minute or so.
That sound very much like the expected behaviour.
Good! If you don't know then you are obviously not trying to run multiple tasks :-). One less complication for the moment.
It seems likely (once you confirm the proper rate of progress along the lines mentioned above) that your GPU may be crunching at a reasonable speed. Can you give some indication of the elapsed time (forget the estimate still remaining) and the current % progress for each task currently running. For the CPU tasks, they will show in BOINC Manager under the 'Application' heading where the first word will either be 'Gravitational' or 'Gamma'. With 2 BRP4G GPU tasks having been completed in very long times, this will have made the estimated crunch times for everything ridiculously high to the point that everything is in panic mode for the moment. As you complete tasks in proper times, these crazy estimates will be reduced down but it will be a slow process needing lots of completed tasks for BOINC to make the full adjustment.
The easiest way to get out of panic mode quickly is to suspend the bulk of tasks on the tasks tab, leaving just enough ready to start as the current ones finish. When you have suspended enough individual tasks, the status will no longer show 'running, high priority' which surely must be showing at the moment. You should pay attention and resume tasks as required to make sure the crunching process doesn't stall for lack of available work. As each task finishes, the estimates will continue to drop and you will be able to resume more and more.
Once you have completed tasks of each type, you will be able to estimate how many you will be able to complete before deadline. At some point in the next couple of days you will need to start working out how many of each type you should abort. When you are ready to do this make sure you have proper work cache settings to avoid any future over-fetch problem. Make yourself aware of the deadlines for the different science runs.
Cheers,
Gary.