Is it possible to add a usage of CPU for idle and for non-idle state.
I want to let my computer use more % of the CPU when he is idle.

Is it possible to have an option like:
On GPUs, use at most x% of N ComputeUnits. So that the first (1-x)*N CU are not used by BOINC. Maybe this could help decrease the screen lag for projects which need to modify their application to avoid this issue.

To be able to manually suspend the current gpu project so the other gpu project starts up until such time you suspend that, and/or resume the first gpu project. It is such a pain to have to suspend gpu work completely just to surf the web because for example "help conquer cancer" tasks in the world community grid project make the fps drop to about 10 intermittently, when gpugrid would work fine for that period.

i want to propose the function of parallel computation for any downloaded task, so the task was splitted for chunks and those chunks can be run in parallel threads loading all cpu cores, right now it is doing this by the server but when i left the only task it is working for about few days loading only 1 core - so it is 25% for me, while this could be more and creating chunks is not so big deal i believe. and this what user could change in preferences prob if he wants to load more tasks from server or crunch it one by one choosing prioritization for each maybe depending on it's deadline status.

creating chunks is not so big deal i believe

Wrong. Some types of computations/algorithms do not lend themselves to parallelization. Some you can't break into chunks or parallel threads. Anyway, it would not be BOINC's job to do that. It is the job of each project to create multithread applications if they can and if they want to.

It would be nice to have an option for "snooze Gpu(X) where there today is snooze GPU down at the windows clock icon when right-clicking the BM icon.

Since I might want to use the GPU, maybe to see a movie and that is connected to my primary display, then it would be nice to snooze that particular GPU and still use all other Gpu's for Boinc.

Today the only way to temporarily do this is to snooze all Gpu's or write a config file.

A snooze Gpu(X) would make it alot smoother.

Get rid of the automatic benchmarking, it only causes problems...it runs while I am actively doing other activities with my computer generating erroneous benchmarks...it clears the GPU wu in the middle of a long run thus restarting the wu from scatch...

Actually that brings up another wish...separate CPUs from GPUs in terms of scheduling and such. Square peg in round hole.

Actually that brings up another wish...separate CPUs from GPUs in terms of scheduling and such. Square peg in round hole.

I think that has long been on their to-do list, but I certainly need it too.

Get rid of the automatic benchmarking

Since BOINC 6.10, the benchmarking only happens:
1. when you've upgraded or downgraded the client.
2. when you've exited & restarted the client, after it has run for 5 days or more.

It does not benchmark on any other times, unless you specifically told it to do a benchmark through the menu.

it clears the GPU wu in the middle of a long run thus restarting the wu from scatch...

Since BOINC 7.0, BOINC leaves GPU tasks in memory when doing benchmarks.

As such:

6/13/2013 11:40:07 PM | Einstein@Home | [coproc] ATI instance 0: confirming for PA0070_01021_57_0
6/13/2013 11:40:45 PM |  | Running CPU benchmarks
6/13/2013 11:40:45 PM |  | Suspending computation - CPU benchmarks in progress
6/13/2013 11:40:45 PM | Einstein@Home | [cpu_sched] Preempting PA0070_01021_57_0 (left in memory)
6/13/2013 11:40:45 PM | Einstein@Home | [task] task_state=SUSPENDED for PA0070_01021_57_0 from suspend

CPU tasks are still unloaded from memory prior to the benchmarks.

Actually that brings up another wish...separate CPUs from GPUs in terms of scheduling and such.

BOINC 7.0 has separate schedulers (and such) for CPU and GPU. These schedulers have been rewritten from the ground up. For 7.2 we're testing the new work fetch module in the present 7.1 range.

Ageless, thanks for the quick feedback.

1. when you've upgraded or downgraded the client.
2. when you've exited & restarted the client, after it has run for 5 days or more.

Since BOINC 7.0, BOINC leaves GPU tasks in memory when doing benchmarks.

So I understand correctly...
Running it after client changes makes sense, no wories there.
But #2...
Is it
a) running the benchmarks every 5 days regardless of start/restart of client?
b) running the benchmark after a restart of the client (i) immediatley, (ii) after 5 days after the restart, or (iii) only after restart if 5 days of continuious running pass prior to the restart?

I ask because I had two different computers run benchmarks today. One I restarted in the morning, auto benchmarking about 5-1/2 hour later. The other has been running for days non-stop.

* * *

BOINC might leave the GPU tasks in memory, but it still somehow messed up the GPU runs, returning errors. It may be app dependent as a number of GPU apps are also using the CPU concurrently. Possibly an error is generated if a GPU looks for the CPU, but the CPU is offline due to benchmarking.

BOINC 7.0 has separate schedulers (and such) for CPU and GPU. These schedulers have been rewritten from the ground up. For 7.2 we're testing the new work fetch module in the present 7.1 range.

The "and such" for me means setting separate minimum and maximum additional work buffers for the CPU and GPU work. Is that possible?

But #2...
Is it
a) running the benchmarks every 5 days regardless of start/restart of client?
b) running the benchmark after a restart of the client (i) immediatley, (ii) after 5 days after the restart, or (iii) only after restart if 5 days of continuious running pass prior to the restart?

Before the change, the client would benchmark every 5 days, as long as the client was running.
After the change, the client will only benchmark immediately upon restart (before anything is loaded), and only do so if prior to the exit & restart the client had been running for 5 days or more.

But it's (remotely) possible that your client starts with the --run_cpu_benchmarks attribute. You can also run it with the --skip_cpu_benchmarks attribute. Guess what that does? :)

BOINC might leave the GPU tasks in memory, but it still somehow messed up the GPU runs, returning errors. It may be app dependent as a number of GPU apps are also using the CPU concurrently. Possibly an error is generated if a GPU looks for the CPU, but the CPU is offline due to benchmarking.

Can also be coincidence, that the GPU app crashed for another reason just around the time a benchmark was done. Without knowledge of the actual error message, stderr.txt entries and/or link to the affected computer, it's guessing.

---

BOINC 7.0 has separate schedulers (and such) for CPU and GPU. These schedulers have been rewritten from the ground up. For 7.2 we're testing the new work fetch module in the present 7.1 range.

The "and such" for me means setting separate minimum and maximum additional work buffers for the CPU and GPU work. Is that possible?

BOINC 7.2 will have rewritten work fetch modules, separate ones for CPU and GPU.
You can test that with 7.1.15, links in the change log thread. Do know, once you start using development versions, these will have bugs. They can crash your system. But all the while be harmless as well.

And know the numbering:
Major dot odd dot revision == Development/alpha client.
Major dot even dot revision == Release Candidate or Recommended Client.

7.1.xx is always development. When using one, you'll have to check in regularly to see if newer (better) versions are available and update.

---

Thanks. That doesn't seem to be implemented in 7.1.5 yet, but knowing that it will come along in 7.2 is good enough for me.

Actually that brings up another wish...separate CPUs from GPUs in terms of scheduling and such.

BOINC 7.0 has separate schedulers (and such) for CPU and GPU. These schedulers have been rewritten from the ground up. For 7.2 we're testing the new work fetch module in the present 7.1 range.

To be honest, I find that the v7.0 scheduler is much more tightly integrated between CPU and GPU than the old one was.

Up to and including v6.12, there was separate recording and accounting for 'debt' between projects, for the different compute units. A lot of time spent crunching with a GPU for a project didn't affect its CPU resource share.

But with v7, all crunching (whichever compute unit) goes into a 'common pot' of REC - and if you crunch GPU for a project, BOINC vary rarely schedules CPU work for that same project. They're interlinked.

i want to propose the function of parallel computation for any downloaded task, so the task was splitted for chunks and those chunks can be run in parallel threads loading all cpu cores...

Answer from the developers: Not going to be done, ever.

It would be nice to have an option for "snooze Gpu(X) where there today is snooze GPU down at the windows clock icon when right-clicking the BM icon.

As per the developers: this is a reasonable request but we're not going to do it soon.

They asked me to make it a trac ticket, which I did: [trac]#1292[/trac]

When a particular device is configured to run multiple WUs simultaneously, I'd like to see BM attempt to retrieve sufficient WUs to fully utilize that device.

Forex: Currently, I have Einstein set to run 4 tasks simultaneously on my GPU. At the time I write this, there are only three WUs running, there are no GPU WUs in the cache, and when I update to get more work, I am told: 'Not requesting tasks: don't need'.

EDIT:
I see now this has already been requested per http://boinc.berkeley.edu/trac/ticket/1239

Consider this a second. :)

not just the columns in the task portion but the tabs as well. we should be able to slide the tabs left or right to our preferences

+1

Hi,


I couldn't find a "new features request" or "feedback" link, but I have a configuration related suggestion.

It would be useful to be able to set "Use at most <###.##> % GPU time" separately from "Use at most <###.##> % CPU time".

My GPU is newer and runs cooler than my CPU. Some projects rely almost entirely on the GPU. It is spring here in Argentina and hot days are coming soon.


Best regards,
Nicola Bernardelli