Let GPU72 decide
 

Chris: Just wondering — how many participants use the "Let GPU 72 Decide" assignment option? :mellow:

[QUOTE=Chuck;338971]Chris: Just wondering — how many participants use the "Let GPU 72 Decide" assignment option? :mellow:[/QUOTE]

Over the last week, six.

I'm probably one of those 6, for a single GPU (GTX570M).

Ressurecting this thread for a (probably dumb) question:
What´s the difference in criterion between the "What makes sense" and "Let GPU72 decide" options for getting work from GPU72 server? After all, GPU72 decisions are supposed to make sense... :wink:

[QUOTE=lycorn;368722]Ressurecting this thread for a (probably dumb) question:
What´s the difference in criterion between the "What makes sense" and "Let GPU72 decide" options for getting work from GPU72 server? After all, GPU72 decisions are supposed to make sense... :wink:[/QUOTE]

After a lengthy debate, the "Let GPU72 decide" option (via MISFIT, #9) also sets the "Pledge" level to what makes the most sense at the current moment, as well as the range.

GPU72 sometimes varies assignments requested under this assignment type with Jerry's and LaurV's machines (with their knowledge and permission). This is to keep the Category 4 DC and LL clients feed.

OK, I see.
Last night I requested DCTF work under "What makes sense" and got 346xxxxx from 70 to 71. Then I used "Let GPU72 decide" and got 347xxxxx from 69 to 70.
Thx for the info.
I´ll submit the results as soon as the server comes to life.

When picking "Let GPU72 Decide" on LLTF get assignments page ever give DCTF assignments? I'm merely curious. I really don't care if I do LL or DC.

I've got my Linux boxes set up to automatically fetch "Let GPU72 Decide" work from the LLTF page, configured to buffer approximately 2 days of work for each GPU. Is 2 days of buffer delay good for the project? One of my cards takes over a day to do a 73->74 assignment.

[QUOTE=Mark Rose;376548]When picking "Let GPU72 Decide" on LLTF get assignments page ever give DCTF assignments? I'm merely curious. I really don't care if I do LL or DC.[/QUOTE]

No, never. Not via the manual assignments page, nor when using the MISFIT or Python automatic fetching spiders.

(One exception: I have in the past occasionally given LaurV's or Jerry's machines DCTF work if we're under a crunch in the DC "Cat4" range, but that shouldn't be needed now that I'm leveraging on several EC2 g2.2xlarge instances plus my G580 to build up a buffer when needed.)

[QUOTE=Mark Rose;376548]I've got my Linux boxes set up to automatically fetch "Let GPU72 Decide" work from the LLTF page, configured to buffer approximately 2 days of work for each GPU. Is 2 days of buffer delay good for the project?[/QUOTE]

Purrrfect! :smile:

[QUOTE=chalsall;376551]the MISFIT or Python automatic fetching spiders.[/QUOTE]

Speaking of which, perhaps the MISFIT forum and teknohog's [url=https://github.com/teknohog/primetools]primetools[/url] could be linked from the Submission Spider page on gpu72.com? It would help people find them easier, especially primetools, which doesn't have its own forum. :)

[QUOTE=Mark Rose;376558]Speaking of which, perhaps the MISFIT forum and teknohog's [url=https://github.com/teknohog/primetools]primetools[/url] could be linked from the Submission Spider page on gpu72.com? It would help people find them easier, especially primetools, which doesn't have its own forum. :)[/QUOTE]

An excellent idea. Soon; but not today...

I'm currently having an "argument" with a new Samsung washing machine my girlfriend bought this last Saturday. The control interface looks like something from a space-ship. (Much water was to be mopped up.)

It seems to claim it is smarter than both of us put together (it might not be wrong :wink:).

[QUOTE=chalsall;376551]I'm leveraging on several EC2 g2.2xlarge instances plus my G580 to build up a buffer when needed.[/QUOTE]

That inspired me. After tweaking all night, I've now got 10 M2050 cards (at a total of $0.80/hr or less) crunching DCTF. Baked an AMI so I can do persistent spot requests. Each card gets a 100 GHz-d (9 hour) buffer. It backs up over rsync every minute, so if I lose one, I can dig through the backup and rescue the work. Had to release back quite a few DCTF exponents are part of the debugging through. Hopefully that didn't mess anything up on GPU72's end.

Now to see how long I feel like paying for this experiment...


Edit: looks like I hit a race condition. I ended up getting one set of assignments 9 times, when all 5 machines (10 cards) launched at once. My guess is a non-isolated read in the database. I'll have to hack in a random delay for polling for work.