Posts by Claggy

1) Message boards : Cruncher's Corner : Something must be wrong on beta wus ! (Message 132112)
Posted 12 days ago by Claggy
Will do benchmarking the trick ? ok, i tested it, no changes so far :(

Just have patience, and let it do it's thing, a number of the apps here only report progress at a number of points during their run.

Claggy
2) Message boards : Cruncher's Corner : Something must be wrong on beta wus ! (Message 132109)
Posted 12 days ago by Claggy
Hello, as i am watching from time to time at the wus running time and % , i discovered that the wus reset after running time over of an hour. Whats up ?

Probably nothing, Boinc since about Boinc 7.2.38 will estimate an app's progress if an app doesn't report any progress (it might have a startup phase where it's not crunching),
Once the app does start reporting it's progress it'll drop either to zero, or whatever the checkpoint says.

http://boinc.berkeley.edu/gitweb/?p=boinc-v2.git;a=commit;h=9136a369d4e15cc727c06b55b50c833e184bf9fc

client: if app doesn't report fraction done, estimate it

http://boinc.berkeley.edu/gitweb/?p=boinc-v2.git;a=commit;h=34f252870310b18c7cbe3e71573daff6b01e768c

client: if app doesn't report fraction done, estimate fraction done in a way that converges to but never reaches 100%.


Claggy
3) Message boards : Problems and Bug Reports : Work Running 100% Of Time Despite Settings? (Message 131994)
Posted 17 days ago by Claggy
Has anyone else seen this? I've been away from Einstein for a while and just got my first batch of work units, which although they have a deadline weeks away are running 100% of the time, i.e. not suspending after 60 minutes to allow other projects to go, despite my settings to the contrary.
It's not a dire problem, but am curious as to why it would happen.

That setting is minimum, ie don't switch between tasks more often than 60 minutes, it doesn't mean Boinc has to switch 'tasks' at the 60 minute mark,
you're been away from Einstein for a while, and will have built a bit of debt to Einstein, so Boinc will crunch Einstein for a bit,

Boinc 7

The new CPU + GPU scheduler and work fetch module
OK, now, you have by now noticed that your old BOINC 6 did things differently than this BOINC 7.0 does things. That's because the old 6 was deemed broken and the new 7.0 is deemed fixed, but with some bugs still included.

The REC-based scheduler
What the client does is, it keeps a record of how much CPU/GPU time a certain project has recently seen - the recent estimated credit (REC). It compares this figure with the project share that has been set. A project that has worked less than its share will get priority in both scheduling (running tasks) and work fetch. Then as it gets crunch time its REC increases and another project will get to the head of the queue. Over time you get a more or less good distribution of crunching time according to resource share.

Points of note:
- GPUs are very productive so lead to high REC. If you run GPU projects alongside CPU ones on similar shares the GPU project sees virtually no CPU.
- CPU and GPU are scheduled separately.
- A project like SETI will probably stay pretty high up in the queue, since getting tasks is hit and miss due to their bandwidth being overworked.
-- However, setting a small 'additional days' cache will help getting tasks from Seti, since BOINC will ask more often, thereby increasing your chances.


Claggy
4) Message boards : Technical News : Binary Radio Pulsar Search (Arecibo) "BRP4" - Arecibo data status (Message 131626)
Posted 32 days ago by Claggy
Any news on BRP4 (GPU version)?

I can see that there are "tasks to send" in BRP4 queue... but I'm not receiving them (only BRP4 GPU enabled).

I'd rather see 100% completion that 1300-something days to completion...

Don't get BRP4 (Binary Radio Pulsar Search (Arecibo)) and BRP4G (Binary Radio Pulsar Search (Arecibo, GPU)) mixed up, these are different subprojects,
the 1300 days of work available is for BRP4 subproject, if you have Arm device running Linux or Android, or an Intel GPU on Windows, then you'll get work from that subproject.

Nvidia and AMD GPUs can get work from the BRP5 (Binary Radio Pulsar Search (Perseus Arm Survey)) subproject, there's plenty of work available for it.

Claggy
5) Message boards : Cruncher's Corner : New NVIDIA driver available (Message 131514)
Posted 38 days ago by Claggy
Does anyone know which CUDAs Einstein uses?

http://einstein.phys.uwm.edu/result.php?resultid=440532175

[quote]<stderr_txt>
Activated exception handling...
[15:38:46][4800][INFO ] Starting data processing...
[15:38:46][4800][INFO ] CUDA global memory status (initial GPU state, including context):
------> Used in total: 628 MB (1421 MB free / 2049 MB total) -> Used by this application (assuming a single GPU task): 0 MB
[15:38:46][4800][INFO ] Using CUDA device #0 "GeForce GTX 660" (0 CUDA cores / 0.00 GFLOPS)
[15:38:46][4800][INFO ] Version of installed CUDA driver: 6000
[15:38:46][4800][INFO ] Version of CUDA driver API used: 3020
[15:38:47][4800][INFO ] Checkpoint file unavailable: status.cpt (No such file or directory).
------> Starting from scratch...quote]

Claggy
6) Message boards : Problems and Bug Reports : Long WUs (Message 131379)
Posted 47 days ago by Claggy
Einstein keeps sending me long WUs that can't all be finished before the deadline. I've got some 45+ hour WUs running at the moment, two of which will miss the 7 June deadline. These WUs are coming in 'High priority' from the moment I receive them and are delaying work for other projects. Luckily LHC has no work for me at the moment - maybe because Einstein is hogging all my computing time. Please, somebody sort this out.

What are your cache settings? If you're getting too much work to be completed in time, then you're set the cache settings too high, and only you can lower them, a day or so should be sufficient.

Claggy
7) Message boards : Cruncher's Corner : Parallella, Raspberry Pi, FPGA & All That Stuff (Message 131348)
Posted 48 days ago by Claggy
FFT on Arm Mali GPU anyone?:

Optimizing Fast Fourier Transformation on ARM Mali GPUs

Claggy
8) Message boards : Problems and Bug Reports : screensaver doesn't work (Message 131288)
Posted 54 days ago by Claggy
I look to optimize my graphics processors attributes. So I need just the opposite everyone is posting about.

I would like to know if you can turn OFF the Screensaver?

Thank You in Advance

It's the same as any other Windows Screensaver, go to screensavers, and select (None) instead of BOINC.

Claggy
9) Message boards : Problems and Bug Reports : Error while computing Gamma-ray pulsar search #3 v1.11 (FGRPopencl-ati) (Message 131245)
Posted 56 days ago by Claggy
If clinfo is providing accurate information then I think you might have found the problem:

Max memory allocation: 134217728
and
Error allocating device memory: 268435456 bytes (error: -61)

I don't claim to have any knowledge at all about AMD cards but from the above I would guess that the card supports memory allocation of 128MB at a time and the app tries to allocate 256MB in one go, so ain't gonna work...
Might be driver dependent but more likely a hardware limit...

Anyone with more knowledge about AMD cards around?

First thing to try is a later driver, earlier (windows) AMD APP runtimes has different 'Max memory allocation' values across different Catalyst versions, you're running APP runtime 1214.3, that is somewhere between Cat 13.4 and 13.9, Try Cat 14.4 if possible:

ATI Driver Version Cheat Sheet

Claggy
10) Message boards : Cruncher's Corner : Parallella, Raspberry Pi, FPGA & All That Stuff (Message 131185)
Posted 62 days ago by Claggy
You'll want the Fake Hardware Clock just like on the Pi:

No more back to 1970

Bored to back to 1970? :lol:
Install fake-hwclock.

sudo apt-get install fake-hwclock


Claggy

Apparently Ubuntu doesn't install ntp by default but Debian does. They have something called ntpdate which is supposed to run at boot time. So I have also done a "sudo apt-get install ntp"

Just installed fake-hwclock on mine, not done ntp yet, have looked up in Synaptic Package Manager, saw ntpdate was installed (client for setting system time from ntp servers), haven't bothered with ntp yet.

Claggy


Next 10

Home · Your account · Message boards

This material is based upon work supported by the National Science Foundation (NSF) under Grants PHY-1104902, PHY-1104617 and PHY-1105572 and by the Max Planck Gesellschaft (MPG). Any opinions, findings, and conclusions or recommendations expressed in this material are those of the investigators and do not necessarily reflect the views of the NSF or the MPG.

Copyright © 2014 Bruce Allen