Posts by Richard Haselgrove

1) Message boards : Problems and Bug Reports : BRP4 Intel GPU app feedback thread (Message 135469)
Posted 1 day ago by Richard Haselgrove
All we know at this stage is that the newer Intel drivers, and the older Einstein apps, are not compatible with each other. I haven't seen anyone with technical/development expertise - on this or any other project - investigate what's going wrong, and identify who has to fix what.

You're right with this.. but it did work using the older drivers and it's still the same executable. Does this leave any realistic wiggle room that it was not Intel who broke something?

MrS

Yes. For a worked example, take the launch of GPGPUs into the Distributed Computing world with NVidia's donation of a SETI@Home application in 2008. All went well (give or take a few bugfixes) until the release of the Fermi GPU architecture in 2010. It turned out that NVidia had taken a few 'optimisation' shortcuts in their own application, and when Fermi came along, the rules were tightened, and the application didn't work any more.

OK, so that's card firmware rather than a driver, but the principle is still close: something which a developer assumed (with supporting experience) would work in an early release runtime, may not be supported in the same way with a more mature release.
2) Message boards : Cruncher's Corner : How to setup app_info.xml? (Message 135375)
Posted 2 days ago by Richard Haselgrove
Refer to the Anonymous platform (app_info.xml) documentation.

You need to supply a

<file_info>
...
</file_info>

block for each file, otherwise the <file_ref> in <app_version> doesn't point to anything. Hence the 'missing file' error message.
3) Message boards : Problems and Bug Reports : project temp down for maintenance (Message 135357)
Posted 3 days ago by Richard Haselgrove
Well, on Friday we did indeed did some work on the scheduler, which would lead to short periods of "project maintenance" (usually <5min).

On top of that there is still a team challenge running on E@H until Monday.

BM

And SETI is on an extended downtime - a science database needs rebuilding, and the other sub-project has processed all available data. Many refugees will have ended up here.
4) Message boards : Problems and Bug Reports : BRP4 Intel GPU app feedback thread (Message 135329)
Posted 3 days ago by Richard Haselgrove
Yes, it's a shame Intel doesn't fix this.

MrS

All we know at this stage is that the newer Intel drivers, and the older Einstein apps, are not compatible with each other. I haven't seen anyone with technical/development expertise - on this or any other project - investigate what's going wrong, and identify who has to fix what.
5) Message boards : Cruncher's Corner : Boinc now supports ASIC (Message 135327)
Posted 3 days ago by Richard Haselgrove
Well, if Einstein (or any other project) needed to calculate massive amounts of hashes, the mining hardware could be used. That's pretty unlikely, though. Even if those hashing functions could be used for parts of the code, there'd be a serious bottleneck transferring data to the miners over USB and getting results back. From the point of view of a CPU or GPU PCIe is already extremely slow ;)

Regarding Bitcoin Utopia to finance BOINC projects: considering the 'coin hype I can understand why people would want to do this. However, those 'coins are set up so that you can only win if you're way more efficient than the others. As soon as you use the the same equipment and software as them, everyone "looses" (=pays more for electricity than they get back). When that cross-over happens obviously depends on your local electricity price. But overall I don't think it's a "sustainable business".

And when I see alternative 'coin schemes being created, simply because the ASICs for Bitcoin are too strong to compete with using less efficient hardware.. I could simply puke. It's like "Damn it, they've got far better hardware. Let's create a slightly different task, which is essentially the same, but can not be performed by their hardware. Then our inefficient hardware is best for that job again!"

MrS

Add to that, the people who design, build, and *ahem* "test" ASIC hardware get first dibs at sweeping up the easy pickings from the new alternative coins.
6) Message boards : Technical News : Gravitational Wave search S6Bucket Follow-up #1 launched (Message 135251)
Posted 4 days ago by Richard Haselgrove
Hallo BM!
Are these new apps running for test with high priority? I hope so.

Kind regards and happy crunching
Martin

That's not something the project can control. 'High Priority' is set by your local client if it needs to gueue-jump the task in order to meet the deadline.
7) Message boards : Cruncher's Corner : CUDA 6.5? (Message 135136)
Posted 6 days ago by Richard Haselgrove
Cuda 6.5 would definitively exclude all compute capability 1.x cards - not only the remaining legacy 8xxx, 9xxx, and 2xx ranges, but also some Quadro FX and even early Tesla models.
8) Message boards : Problems and Bug Reports : Einstein, cc_config exclusions and Work unit limits (Message 135050)
Posted 8 days ago by Richard Haselgrove
It works fine with them, but you'll need to catch the server logs at the right time - i.e. when the request you're interested in has just (most recently) happened. We don't have access to a full history, just the most recent.
9) Message boards : Problems and Bug Reports : Einstein, cc_config exclusions and Work unit limits (Message 135044)
Posted 8 days ago by Richard Haselgrove
So,

Having sufficient time on my hands to test some things here, I've noticed an issue with receiving work units.

I've set up my cc_config to maximize my gpus while running Einstein. What I've found is that after the initial request for work units, the server will not provide my computer with sufficient work units after all work is completed.

Example, I request a 5 days worth of work and receive it initially. But after I have crunched all that work and only have 2-3 left, the server will deny my computer any more work. It is only when I am completely out will it parsal out 2 or 3 work units rather than a 5 day allotment.

There are no other projects requesting work.

Resetting the project has no effect.

The only way I have found to resolve this is it remove the project completely and reinstall it.

Ideas?

As always, read the logs.

You have two sets available here:

1) <work_fetch_debug> in your local BOINC client, to see what work you're requesting and why (or why not)
2) Server logs, to see what request was received and how it was dealt with. At the moment, both http://einstein.phys.uwm.edu/host_sched_logs/10267/10267709 and http://einstein.phys.uwm.edu/host_sched_logs/11681/11681771 are indicating that no work was requested on the last contact. So look to your client first - sending no work when none is requested doesn't amount to a "denial", in my book.
10) Message boards : Problems and Bug Reports : All Einstein cpu tasks stuck at 98% (Message 134800)
Posted 13 days ago by Richard Haselgrove
I forgot to comment on the CPU support needed in my previous post.

Archae86's observations is still true, the Intel GPU app needs a full CPU core to support it or the run time of the tasks really blows up, it's on the scale of taking hours compared to about 15 min per task. Checking in windows task manager the currently running Intel GPU tasks has an elapsed time of 10 min but has only used 21 seconds of CPU time, mostly the process reports 0% CPU usage with the occasional 1-2% blip.

Also confirmed, in response to Archae86's post. Tried my Haswell i5 HD4600 (driver version 10.18.10.3621) with all four cores loaded: runtime increased ~7x (on the basis of taking seven minutes to reach 10%, against the normal 11 minutes for the full task).


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