Anonymous Platform, server-side settings problem?
11 Feb 2014 8:32:46 UTC
Topic 197378
(moderation:
One of my machine is having a mix of ATI+CUDA+Intel_GPU.
(R9-280X, GTX670, 4770K).
Crunching GPU-only BRP5(Perseus); but I wanted to run 3xATI and 2xCUDA.
This isn't possible via standard Preference/Profile & app_config.xml - as that one setting is for all GPU.
So I'm using app_info.xml.
Crunch ATI+CUDA+Intel_GPU well when first transitioned to app_info.xml (existing workunits).
= No problem with app_info.xml.
But since then I can get ATI & Intel_GPU work, but _not_ CUDA workunits.
Here's the latest scheduler log.
2014-02-11 08:08:41.7896 [PID=13845] Request: [USER#xxxxx] [HOST#8522935] [IP xxx.xxx.xxx.189] client 7.2.33
2014-02-11 08:08:41.8071 [PID=13845] [send] effective_ncpus 4 max_jobs_on_host_cpu 999999 max_jobs_on_host 999999
2014-02-11 08:08:41.8071 [PID=13845] [send] effective_ngpus 3 max_jobs_on_host_gpu 999999
2014-02-11 08:08:41.8071 [PID=13845] [send] Not using matchmaker scheduling; Not using EDF sim
2014-02-11 08:08:41.8071 [PID=13845] [send] CPU: req 0.00 sec, 0.00 instances; est delay 0.00
2014-02-11 08:08:41.8071 [PID=13845] [send] CUDA: req 12211.98 sec, 1.00 instances; est delay 0.00
2014-02-11 08:08:41.8071 [PID=13845] [send] ATI: req 0.00 sec, 0.00 instances; est delay 0.00
2014-02-11 08:08:41.8071 [PID=13845] [send] Intel GPU: req 0.00 sec, 0.00 instances; est delay 0.00
2014-02-11 08:08:41.8071 [PID=13845] [send] work_req_seconds: 12211.98 secs
2014-02-11 08:08:41.8072 [PID=13845] [send] available disk 45.28 GB, work_buf_min 21600
2014-02-11 08:08:41.8072 [PID=13845] [send] active_frac 0.999938 on_frac 0.999811 DCF 1.080330
2014-02-11 08:08:41.8072 [PID=13845] Anonymous platform app versions:
2014-02-11 08:08:41.8072 [PID=13845] app: einsteinbinary_BRP5 ver: 139
2014-02-11 08:08:41.8072 [PID=13845] app: einsteinbinary_BRP5 ver: 139
2014-02-11 08:08:41.8072 [PID=13845] app: einsteinbinary_BRP4 ver: 134
2014-02-11 08:08:41.8107 [PID=13845] [send] [HOST#8522935] is reliable
2014-02-11 08:08:41.8108 [PID=13845] [send] set_trust: random choice for error rate 0.000010: yes
2014-02-11 08:08:41.8109 [PID=13845] [mixed] sending locality work first (0.5291)
2014-02-11 08:08:46.3023 [PID=13845] [version]: App 'einstein_S6CasA' (24) not selected
2014-02-11 08:08:46.3025 [PID=13845] [mixed] sending non-locality work second
2014-02-11 08:08:46.3149 [PID=13845] [version] Don't need ATI jobs, skipping anonymous version 139 for einsteinbinary_BRP5 (BRP5-opencl-ati)
2014-02-11 08:08:46.3150 [PID=13845] [version] Don't need CPU jobs, skipping anonymous version 139 for einsteinbinary_BRP5 (BRP5-cuda32-nv301)
2014-02-11 08:08:46.3150 [PID=13845] [send] Didn't find anonymous platform app for einsteinbinary_BRP5
2014-02-11 08:08:46.3218 [PID=13845] [debug] [HOST#8522935] MSG(high) No work sent
2014-02-11 08:08:46.3218 [PID=13845] Sending reply to [HOST#8522935]: 0 results, delay req 60.00
2014-02-11 08:08:46.3222 [PID=13845] Scheduler ran 4.539 seconds
Note the "[version] Don't need CPU jobs, skipping anonymous version 139 for einsteinbinary_BRP5 (BRP5-cuda32-nv301)".
Is there something wrong with server-side settings?
Thanks in advance.
terencewee*
Sicituradastra.
terencewee*
Sicituradastra.
Anonymous Platform, server-side settings problem?
)
Hi!
It would be best to see your app_info.xml file to check if there's anything wrong with that.
Better still, you might want to check the section about app_config.xml files here : http://boinc.berkeley.edu/wiki/Client_configuration, because I think this is better suited than app_info.xml for what you have in mind.
Cheers
HBE
Thanks Bikeman. I'd reset
)
Thanks Bikeman.
I'd reset the project to clear off unwanted files and switched to app_config.xml.
I missed the tag. :)
It's working now.
EDIT: Work units are downloading but is only supported in BOINC 7.3+ (not available yet). So I'm out of luck. Will run x2 on each GPU.
It's still strange why server will skip CUDA plan-class when not requesting CPU work, yes?
terencewee*
Sicituradastra.
RE: EDIT: Work units are
)
If you feel adventurous you can always download all the Alpha, Beta and release versions of Boinc here http://boinc.berkeley.edu/dl/?C=M;O=D
Version 7.2.39 and an APP file is very helpful in making the newer R series of AMD gpu's work everywhere.
RE: EDIT: Work units are
)
I wouldn't believe that, it was implemented in Boinc 7.2.30, and should be in every dev version afterwards (But not the recommended releases):
Preliminary Change Log 7.2.29 --> 7.2.30:
client: allow elements in app_config.xml, allowing users to override the parameters of particular app versions.
Claggy
app_config.xml:
)
app_config.xml:
einsteinbinary_BRP5
BRP5-opencl-ati
0.1
0.33
einsteinbinary_BRP5
BRP5-cuda32-nv301
0.1
0.5
Claggy is correct.
Boinc-client recommended version 7.2.33(x64); did not work.
Boinc-client version 7.2.39(x64) (get it here, works.
Presently running 3x ATI, 2x CUDA.
Thanks guys! :)
terencewee*
Sicituradastra.
RE: app_config.xml:
)
Boinc 7.2.39 has now become the latest recommended Boinc version, you can get it straight from Berkeley's Boinc download page now.
Claggy