BRP4 Status page needs an update?

Exard3k
Exard3k
Joined: 25 Jul 21
Posts: 66
Credit: 56155179
RAC: 0

I just setup a new OS and

I just setup a new OS and everything is fine now. PopOS 22.04 had some problems and I don't know what was wrong with the system (basically identical packages installed), but with Kubuntu 22.04 everything is being detected and working (after renaming the name of the GPU in co_proc.xml and changing file attribute).

Thanks Steve! :)

P.S.: Behold, the new iGPU crunching machine has arrived! :)

 

edit: Wow that iGPU thread is using quite some memory bandwidth....normal RAM as VRAM, leaves little for anything else on the system (including CPU BOINC threads). But performance is like 30x CPU threads anyway, so that makes up for the deficits.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5875
Credit: 118473821535
RAC: 25985691

Exard3k wrote:... everything

Exard3k wrote:
... everything is being detected and working (after renaming the name of the GPU in co_proc.xml and changing file attribute).

The name of the file is coproc_info.xml and it might be a bit of a problem for the staff when they discover all the results that wouldn't normally get validated that are now in their database.

Please realise (if you read Bernd's announcement of the new BRP4/BRP4G {Arecibo,large}/BRP7 stuff in Technical News) that BRP4 is NOT supposed to be done by Intel GPUs anymore.  You are supposed to do the Arecibo, large stuff (BRP4G) which is a bundle of 8 with a credit award of 8 x 62.5 = 500.   Please read about the insane stress on the servers from handling large numbers of small tasks that led to the decision to create these bundles.

The applications page shows a Windows app for BRP4G.  There doesn't appear to be one for Linux.  If there was, at least there would also be a quorum partner to validate your result.  Just because there is no BRP4G app, you really shouldn't fudge things to use BRP4 instead.  The whole sorry history of non-validating results for BRP4 when using Intel GPUs is in this thread.  You should read it some time, particularly comments from Bikeman, Bernd, Christian Beer (former staff member) and Richard Haselgrove.  It's been quite a saga over many years.

Apart from server stress, please realise that most of the small BRP4 tasks have an initial replication of just 1.  That means that most of what you're returning is not being checked for being valid.  My guess is that maybe only 5-10% of your allocated tasks come with a quorum partner.  From memory, it's a fairly small fraction. 

I just had a quick look at your results.  There are 21 invalid and a significantly larger number of 'inconclusives'.  When you realise that all these come from the small percentage that are actually being checked against another result, there must be many more that would have failed validation if there had been quorum partners for everything.

I have no idea what this might mean for the integrity of what is now in the database and what the staff might have to do to remove unacceptable results.  I think you should at least confirm with the staff that they are happy for you to be doing this fudge.  I wonder how the poor PhD student might feel about having to deal with suspect data.

EDIT:  Here is another short thread that will help explain why what you are doing is likely to cause issues for the staff.  It's only a few messages in total but the comments from Richard Haselgrove clearly show the problems when Intel GPUs are competing with the much smaller devices for which BRP4 is now intended.  That was a situation where every result was being validated against a quorum partner and the invalids plus likely to be invalid (ie. inconclusives) from the Intel GPU do dominate.

Cheers,
Gary.

Comment viewing options

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