Posts by Claggy
log in
1) Message boards : News : Variable Sized Wokunits & New CPU Applications (Message 18939)
Posted 1391 days ago by Claggy
I see there is a "Mobile" location choice for individual systems, but no settings pages for that choice. It also would be nice to be able to use that location resource.

Later Boinc server software has had that preference removed, as it was a lot of work to implement, so wasn't done.
the Official Android Boinc only looks at the default location/venue anyway, you may move it to a different location/venue and that server side project preferences will be used,
But if you have 'Use CPU' on the default venue set to NO, the Official Android Boinc won't ask for work, there is a changeset in place to fix that, but Rom hasn't built a new Android client with that fix in place yet.

Claggy
2) Message boards : News : Variable Sized Wokunits & New CPU Applications (Message 18929)
Posted 1393 days ago by Claggy
Lastly, there still seems to be an issue with multiple venues and plan class choices. A fix will be put in place as soon as I figure out why it isn't working properly.

Yep, my Windows x64 i7-2600K/HD7770 host has been crunching OpenCL ATI/AMD GPU tasks exclusively,

But the Application details page has stats for these ones too:

solo_collatz 6.06 i686-pc-linux-gnu (opencl_intel_cpu),

solo_collatz 6.02 x86_64-pc-linux-gnu (cuda40)

solo_collatz 6.00 armv6l-unknown-linux-gnueabi

http://boinc.thesonntags.com/collatz/host_app_versions.php?hostid=20542

Claggy
3) Message boards : News : Android Applications Now Available (Message 18799)
Posted 1415 days ago by Claggy
Still no work for my two phones.
Probably "dont need" - the buffer of WUs is probably full of *simap* W-units. :(

The Official Boinc client for Android doesn't use venues other than the default, and even then most of the web preferences are overridden in the client,
while you may move the host to a different venue, and the server will display the host being on that venue, the client is still looking at the default,
If you have 'Use CPU' set to 'No' on the default, you won't get any work, the solution is move it to the default, and any non-Android hosts away from the default,
Yes, I've mentioned this on the mailing lists after Jorg encountered it (and arkayn beforehand, and others), just waiting for someone to take some notice.

Claggy
4) Message boards : Number crunching : Intel GPUs not showing on the Boinc 'Choosing BOINC projects' page for Collatz (Message 17946)
Posted 1514 days ago by Claggy
That's sorted now, thanks.

Claggy
5) Message boards : Number crunching : Intel GPUs not showing on the Boinc 'Choosing BOINC projects' page for Collatz (Message 17926)
Posted 1515 days ago by Claggy
Or, maybe it can and I just don't know how. :-)

The url points to Collatz's server!!!

Claggy
6) Message boards : Number crunching : Clashing deadlines with Primegrid. (Message 17860)
Posted 1522 days ago by Claggy
I have an unattended pc at home running BOINC with two projects: Collatz (normal and solo, Nvidia GPU) and Primegrid (pps sieve only, Nvidia GPU). No CPU projects (I am mining Primecoins with my 6-cores CPU).
4/5ths of time scheduling is on Primegrid, 1/5th on Collatz.
Primegrid expiration deadlines are much shorter (3 days).
Because of this, Boinc Manager does not fulfill the time slices and attributes priority to Primegrid, whose workunits always expire earlier.
I end up having a number of Collatz units expired "Not started by deadline - canceled", one expired with no answer, a number of self-aborted Primegrid's WU (aborted by user).
Presumably, Boinc forced Primegrid to stop by aborting the running WUs, while Collatz didn't have any time left for its WUs.

How do I solve this problem?

Run a smaller cache size, so PrimeGrid tasks aren't always in High Priority (aka Earliest Deadline Mode), then Boinc will be able to do it's normal round robin scheduling. (Once it's paid back the debt to Collatz)

Claggy
7) Message boards : Number crunching : How to: reduce CPU usage while running on Intel GPU (Message 17856)
Posted 1523 days ago by Claggy
Raistmer has made a lot of progress on reducing CPU usage for his Nvidia Astropulse app after finding a Bug and working around it:

http://setiathome.berkeley.edu/forum_thread.php?id=73290&postid=1443401

Very nasty bug (feature?) was discovered recently in attempt to reduce CPU usage of OpenCL NV AP app: asynchronous buffer reads actually done as synchronous ones.
Bug was filed via nVidia CUDA registered developer program.

and for reducing CPU usage on his Intel GPU Astropulse and Multibeam v7 app after studying Einstein's Intel OpenCL app:

http://setiathome.berkeley.edu/forum_thread.php?id=73290&postid=1443601

EDIT:thanks to Einstein@home project we now have further improved oclFFT library that works much more precise on IntelGPUs than original one (and whole idea of too bad native_sin/cos precision on Intel GPU is inspired by conversations with Oliver) also, comparison between SETI and Einstein's sourses reveal some difference in synchronisation usage (namely, they use direct clFinish() call instead of other possible methods). Extention of this approach to quite ridiculous degree (to call clFinish after EACH OpenCL call) allowed to decrease CPU usage on IntelGPUs considerably (need to note that Einstein's code doesn't use clFinish in that way, but looks like very such (and definitely not the one I could call "normal") way of usage provides best results on Intel GPUs).
On nVidia GPUs such excessive clFinish calls do nothing but increase total runtime and CPU overhead (as one could expect from general considerations)


Ideas to try.

Claggy
8) Message boards : Number crunching : Intel GPUs not showing on the Boinc 'Choosing BOINC projects' page for Collatz (Message 17848)
Posted 1524 days ago by Claggy
DA has been updating the 'Choosing BOINC projects' page with icons, I reported that there were no projects with Intel GPUs,
he has fixed Einstein, but Collatz doesn't show up, probably because Collatz's get_project_config.php is outdated:

The BOINC web site gets the platforms from
http://boinc.thesonntags.com/collatz/get_project_config.php
which isn't showing the Intel GPU platform.

This discrepancy could be because of
- a bug in get_project_config.php
- Collatz has out-of-date get_project_config.php
- permission problem on the cached file of platforms
(html/cache/platform_list.xml) preventing it from being updated.

-- David


Claggy
9) Message boards : Number crunching : Optimizing Collatz v6.xx OpenCL and CUDA Applications (Message 17088)
Posted 1656 days ago by Claggy
You will need to restart BOINC, and then the app for it to take effect.

No, you don't need to restart Boinc, the whole point of having a config file is so Stock users can supply cmd parameters to the apps, and that changes can be done on the fly,
either suspending GPU usage momentary, or suspending and resuming the running GPU task will cause the app to use the new parameters.
(If you were putting cmd parameters into an app_info, then yes a Boinc restart is required as the app_info is only read on startup)

Claggy
10) Message boards : Number crunching : Not Requesting Tasks.... I want tasks! (Message 17013)
Posted 1665 days ago by Claggy
What would you need an app_info for over here? I can't think of any uses for CC which would not be covered in a better way by the app_config.

MrS

For testing a later app than is supplied as Stock?

Claggy
11) Message boards : Number crunching : ATI application Q&A (Message 17004)
Posted 1666 days ago by Claggy
Didn't want to open a new topic about this, but does CPU power matter in OpenCL applications since it's using an entire core when operating? Just trying to run WU's as efficient as possible.

Have a read of:

Optimizing Collatz 4.07 Applications

Setting a lower 'kernels_per_reduction' and/or a higher 'sleep' value should help lower CPU usage.

Claggy
12) Message boards : Number crunching : Not Requesting Tasks.... I want tasks! (Message 16992)
Posted 1667 days ago by Claggy
If that doesn't work, let me know and I'll figure something out.

Eric at Seti Beta is still trying to get anonymous platform to work for Intel GPUs without much luck, he has done a few fixes:

Astropulse 6.05 for OpenCL on Intel GPUs released.

Changeset 4e338e94 in boinc-v2

Changeset 03e64f72 in boinc-v2

So I think some debugging is still in order.

Claggy
13) Message boards : Number crunching : Should solo_collatz run on 4770? (Message 16980)
Posted 1668 days ago by Claggy
So any idea why this may be if it isn't the OpenCL issue?

Your HD 4700/4800 hosts are running Boinc 6.10.56 and 6.12.34, those Boinc version's don't do OpenCL detection, so the server has no way of knowing if your GPUs are OpenCL capable,
Upgrade to Boinc 7.0.64 and Boinc will be able to detect if OpenCL capability is available.

Claggy
14) Message boards : Number crunching : Should solo_collatz run on 4770? (Message 16865)
Posted 1677 days ago by Claggy
Doh! Maybe I just dreamed that the HD 3870's were OpenCL compatible because they were double precision.

Don't worry, not all the Boinc dev's realised that not all ATI GPUs were OpenCL capable either, which causes confusion when mixing OpenCL capable and non OpenCL capable GPUs.

http://setiathome.berkeley.edu/forum_thread.php?id=71192&sort_style=6&start=0

Claggy
15) Message boards : Number crunching : Should solo_collatz run on 4770? (Message 16861)
Posted 1677 days ago by Claggy
Well I'm running a 4870 and 4850 and I can't get solo to give me any tasks. I've checked the preferences and I should be getting them.


BOINC is not recognizing that your GPUs can run OpenCL. While it recognizes they can run CAL, there is no mention of OpenCL. The driver version reported by BOINC shows 1.4.1546 which does not support OpenCL.
So... either upgrade to the latest legacy driver (I know 1.4.1734 works OK) or downgrade to the Catalyst 11.6 driver (the last one that supported OpenCL on HD 38xx and HD 4xxx GPUs).

Why? AMD dropped OpenCL support for the older GPUs when they came out with OpenCL 1.1 drivers as the HD 38xx and HD 4xxx GPUs hardware was not compatible with OpenCL 1.0 (even though the HD 4xxx GPUs were still available in stores). So you either have to run an old pre-OpenCL 1.1 driver or the driver made specifically for legacy GPUs.

AMD Never dropped OpenCL support on HD38xx GPUs, HD38xx GPUs never had OpenCL support in the first place, Here's a post by Ageless showing the initial GPU requirements from Steam SDK 2.0:

ATI Radeon™ HD 5870
ATI Radeon™ HD 5850
ATI Radeon™ HD 5770
ATI Radeon™ HD 5750
ATI Radeon™ HD 4890
ATI Radeon™ HD 4870 X2
ATI Radeon™ HD 4870
ATI Radeon™ HD 4850 X2
ATI Radeon™ HD 4850
ATI Radeon™ HD 4830
ATI Radeon™ HD 4770
ATI Radeon™ HD 4670
ATI Radeon™ HD 4650
ATI Radeon™ HD 4550
ATI Radeon™ HD 4350

ATI FirePro™ V8750
ATI FirePro™ V8700
ATI FirePro™ V7750
ATI FirePro™ V5700
ATI FirePro™ V3750

AMD FireStream™ 9270
AMD FireStream™ 9250

ATI Mobility Radeon™ HD 4870
ATI Mobility Radeon™ HD 4860
ATI Mobility Radeon™ HD 4850 X2
ATI Mobility Radeon™ HD 4850
ATI Mobility Radeon™ HD 4830
ATI Mobility Radeon™ HD 4670
ATI Mobility Radeon™ HD 4650
ATI Mobility Radeon™ HD 4500 Series
ATI Mobility Radeon™ HD 4300 Series

ATI Mobility FirePro™ M7740

ATI Radeon™ Embedded E4690 Discrete GPU


The archived sdk-v2-0 page now (it did once) has no mention of GPU requirements, only CPU requirements (But note it is part of the GPU Tools Archive):

http://developer.amd.com/resources/archive/archived-tools/gpu-tools-archive/ati-stream-software-development-kit-sdk-v2-0-beta-program/

Claggy
16) Message boards : News : Intel GPUs Supported (Message 16834)
Posted 1679 days ago by Claggy
Slicker, do you know if the app works under anonymous platform? Seti and Albert@home are having difficulties getting work when using anonymous platform.

Claggy
17) Message boards : News : Server Upgrade on Thursday, April 18th (Message 16296)
Posted 1736 days ago by Claggy

I wonder whether there is an issue getting work for the Intel GPU when there is a second non-Intel GPU installed.

You are right, there also 2 nvidia videocards installed:
19/04/2013 18:02:10 | | CUDA: NVIDIA GPU 0: GeForce GTX 460 (driver version 314.22, CUDA version 5.0, compute capability 2.1, 1024MB, 951MB available, 1089 GFLOPS peak)
19/04/2013 18:02:10 | | CUDA: NVIDIA GPU 1: GeForce GTX 460 (driver version 314.22, CUDA version 5.0, compute capability 2.1, 1024MB, 951MB available, 1089 GFLOPS peak)
19/04/2013 18:02:10 | | OpenCL: NVIDIA GPU 0: GeForce GTX 460 (driver version 314.22, device version OpenCL 1.1 CUDA, 1024MB, 951MB available, 1089 GFLOPS peak)
19/04/2013 18:02:10 | | OpenCL: NVIDIA GPU 1: GeForce GTX 460 (driver version 314.22, device version OpenCL 1.1 CUDA, 1024MB, 951MB available, 1089 GFLOPS peak)
19/04/2013 18:02:10 | | OpenCL: Intel GPU 0: Intel(R) HD Graphics 4000 (driver version 9.17.10.2932, device version OpenCL 1.1, 728MB, 728MB available, 45 GFLOPS peak)


What Boinc version are you running?

Claggy
18) Message boards : News : Server Upgrade on Thursday, April 18th (Message 16288)
Posted 1737 days ago by Claggy
This upgrade should enable Collatz to support OpenCL applications on Intel embedded GPUs.

Forgive my ignorance, but how do I know or check if my i5 lappie is a candidate for this. At start-up, the BOINC manager log says "No usable GPU's found". Is it a matter of simply updating the driver? The Windows Device Mangler indicates it's using version 8.15.10.2253 from 11/28/10. And, off course, after running the driver update search, it tells me that the "driver is up to date".

Cheers!

MarkR

You need an Ivy Bridge CPU for the GPU to be used a an OpenCL device, the Sandy Bridge CPU's GPU is not supported.

Also, even if it was supported, you'd need a later Boinc version that has Intel GPU OpenCL detection support too, Boinc 7.0.28 is too old.

Claggy
19) Message boards : News : Collatz 4.xx Windows OpenCL App for AMD 32 and 64 bit Released (Message 16243)
Posted 1742 days ago by Claggy
Urs Echternacht fould the reason the Setiathome AMD/ATI MB and AP apps were failing on Cat 13.2 and Cat 13.3 (APP runtime 1124.2) was because that APP runtime doesn't like the build option "-fno-bin-llvmir" from OpenCL compiler options,
removing that option when building the apps (or it might be when the apps build it's compilations) allowed the apps to work again, while i doubt it's the problem here, it's something to try:

http://devgurus.amd.com/message/1288791#1288791

Claggy
20) Message boards : News : Collatz 4.xx CUDA and CPU Applications Released for Windows x86_64 (Message 16208)
Posted 1748 days ago by Claggy
I don't know if nobody reads this thread or nobody cares, but an answer would be appreciated.

As arkayn has already said, the config file need eithier to be empty when you receive if from the project, and you add your wanted parameters (have a separate readme with info on what to out in the config file),
Or you have it fully featured with all the parameters, But it's got to stay the same file size, it's O.K to change a 16 to a 15, but what you can't do is add extra charcters or an extra line, as that changes the file size, and Boinc will redownload it.

Claggy


Crap. I just ran into this again. What events cause a new file to be downloaded?

Also, I think I saw someone mention that it would be fixed in a later version of BOINC? I am having the same problem with 7.0.60.

There is no fix in later Boinc's, the project has to supply an empty config file, then you are free to put whatever in the config file.

Claggy


Next 20

Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.