BRP4 large, available for which OS`s?

Drago75
Drago75
Joined: 19 Sep 20
Posts: 14
Credit: 20,009,175
RAC: 19,844
Topic 228849

My Odroid N2+ is receiving the short BRP4 units running Android. Wanted to try out the larger version of it but I get error messages on my AMD Linux and AMD Windows hosts claiming there was no app version for those OS available. So what app versions are availble? I saw Keith running them on his Nvidia Jetsons.

Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3,899
Credit: 43,456,632,644
RAC: 62,882,438

Keith is running a custom

Keith is running a custom CUDA application on his Nano, using the GPU. He's not using the ARM CPU for that app, since an ARM app for BRP4G isn't available. you can see all applications available HERE

 

BRP4G (or "large") is really the same data as BRP4, it's just like 8 of them packaged together in the same task. and hence takes 8x longer to run and gets 8x more credits. there's no point to run BRP4G on a slow ARM device, it will take forever to run. the different WU type were intentionally split up to keep the smaller tasks on the slower devices and allowing the faster x86 CPUs and intel GPUs to run the larger batch jobs.

_________________________________________________________________________

Drago75
Drago75
Joined: 19 Sep 20
Posts: 14
Credit: 20,009,175
RAC: 19,844

Hey Ian & Steve. Thanks for

Hey Ian & Steve. Thanks for the clarification. According to the table of available apps both Windows and Linux on AMD are listed so I should get those. Here is the event log. Maybe I am reading it wrong but it sounds like there is no app available. Or is there a different problem?

 

2023-01-03 14:48:56.6386 [PID=850200]   Request: [USER#xxxxx] [HOST#12932717] [IP xxx.xxx.xxx.163] client 7.16.6
2023-01-03 14:48:56.6463 [PID=850200] [debug]   have_master:1 have_working: 1 have_db: 1
2023-01-03 14:48:56.6464 [PID=850200] [debug]   using working prefs
2023-01-03 14:48:56.6464 [PID=850200] [debug]   have db 1; dbmod 1669759766.000000; global mod 1669759766.000000
2023-01-03 14:48:56.6464 [PID=850200]    [send] effective_ncpus 32 max_jobs_on_host_cpu 999999 max_jobs_on_host 999999
2023-01-03 14:48:56.6464 [PID=850200]    [send] effective_ngpus 1 max_jobs_on_host_gpu 999999
2023-01-03 14:48:56.6464 [PID=850200]    [send] Not using matchmaker scheduling; Not using EDF sim
2023-01-03 14:48:56.6464 [PID=850200]    [send] CPU: req 240061.85 sec, 32.00 instances; est delay 0.00
2023-01-03 14:48:56.6464 [PID=850200]    [send] CUDA: req 0.00 sec, 0.00 instances; est delay 0.00
2023-01-03 14:48:56.6464 [PID=850200]    [send] work_req_seconds: 240061.85 secs
2023-01-03 14:48:56.6464 [PID=850200]    [send] available disk 83.23 GB, work_buf_min 172800
2023-01-03 14:48:56.6464 [PID=850200]    [send] active_frac 0.985418 on_frac 0.207668 DCF 1.000000
2023-01-03 14:48:56.6476 [PID=850200]    [mixed] sending non-locality work first (0.5043)
2023-01-03 14:48:56.6679 [PID=850200]    [send] [HOST#12932717] will accept beta work.  Scanning for beta work.
2023-01-03 14:48:56.6724 [PID=850200]    [version] Checking plan class 'opencl-intel_gpu'
2023-01-03 14:48:56.6734 [PID=850200]    [version] reading plan classes from file '/BOINC/projects/EinsteinAtHome/plan_class_spec.xml'
2023-01-03 14:48:56.6735 [PID=850200]    [version] parsed project prefs setting 'gpu_util_brp': 1.000000
2023-01-03 14:48:56.6735 [PID=850200]    [version] No Intel GPU devices found
2023-01-03 14:48:56.6735 [PID=850200]    [version] no app version available: APP#19 (einsteinbinary_BRP4) PLATFORM#7 (x86_64-pc-linux-gnu) min_version 0
2023-01-03 14:48:56.6735 [PID=850200]    [version] no app version available: APP#19 (einsteinbinary_BRP4) PLATFORM#1 (i686-pc-linux-gnu) min_version 0
2023-01-03 14:48:56.6755 [PID=850200]    [mixed] sending locality work second
2023-01-03 14:48:56.6780 [PID=850200] [debug]   [HOST#12932717] MSG(high) No work sent
2023-01-03 14:48:56.6780 [PID=850200] [debug]   [HOST#12932717] MSG(high) see scheduler log messages on https://einsteinathome.org/host/12932717/log
2023-01-03 14:48:56.6780 [PID=850200]    Sending reply to [HOST#12932717]: 0 results, delay req 60.00
2023-01-03 14:48:56.6781 [PID=850200]    Scheduler ran 0.047 seconds
Ian&Steve C.
Ian&Steve C.
Joined: 19 Jan 20
Posts: 3,899
Credit: 43,456,632,644
RAC: 62,882,438

You’re missing that the

You’re missing that the windows and Linux apps are for x86 only. That application does not work on ARM. Totally different CPU architecture and instruction sets. the only ARM apps are the ones listed for BRP4 that you’re already getting. 

edit:

didn't realize you linked to a totally different system now.

your work request on that system doesn't appear to be requesting any CPU work. so that's why you didnt get any. BRP4G is a CPU application now for the project.

the AMD CPU does not have an intel iGPU, so there's nothing to get.
the nvidia GPU cannot run BRP4 since there are no stock apps for it, so there's nothing to get. (Keith's application is for ARM/CUDA and will not work on x86/CUDA platforms)

if you want to run BRP4 on that host, you will need to enable CPU processing with a jhigh enough resource share (or no competing projects) so that BOINC actually requests work for it.

but you're better off running FGRPB1G on that host on the nvidia GPU, or even the BRP7 campaign.

 

_________________________________________________________________________

Drago75
Drago75
Joined: 19 Sep 20
Posts: 14
Credit: 20,009,175
RAC: 19,844

Ok what actually helped was

Ok what actually helped was selecting "Arecibo, fast" and "Meerkat" also. Now I am getting the "Arecibo, large" units. Maybe the names got a little mixed up...

Link
Link
Joined: 15 Mar 20
Posts: 118
Credit: 7,595,036
RAC: 44,203

Ian&Steve C. wrote:BRP4G (or

Ian&Steve C. wrote:
BRP4G (or "large") is really the same data as BRP4, it's just like 8 of them packaged together in the same task. and hence takes 8x longer to run and gets 8x more credits. there's no point to run BRP4G on a slow ARM device, it will take forever to run.

It takes "forever", but at least 20+% of your tasks are not marked invalid because of Intel GPUs and aarch devices. I use on my Android device same app as I got for BRP4 with an app_info file to make it BRP4G (actually both BRP4 and BRP4G so I can switch between them in the web preferences). No issues so far, all tasks validate instantly against the x86 apps, no inconclusives, no invalids.

The invalid rate for BRP4 is insane, just check the valid:invalid ratio on the ssp and compare to BRP4G. Currently it's 4.1:1 for BRP4 and 1124:1 for BRP4G. They should be the about same if all app versions were OK, after all it's same data and same processing.

Definitely worth moving the Odroid to BRP4G, it has the same issue and same high invalid rate I had, it's results are often not validating against those of Intel GPUs, aarch64 app and maybe also the v1.51 (ASIMDPIE_1X) app, there are currently three tasks in the invalid list where Intel GPU or aarch64 validated against that app and not the v1.46 (ASIMDPIE) the Odroid is running.

I reported this a while ago in the Android feedback thread.

.

Drago75
Drago75
Joined: 19 Sep 20
Posts: 14
Credit: 20,009,175
RAC: 19,844

Ok you got a point there with

Ok you got a point there with the high invalid rate. I think I am getting ready to sell the Odroid because it's output level is very low even if all tasks validate. I ran two of the large BRP4 files on my Windows laptop and they came back successfully just waiting for validation. Maybe I'll stick to that app and skip the Android version for now.

Thanks for your help, Ian & Steve

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.