Too much credit compared to others like SETI
1 Oct 2016 18:17:31 UTC
Topic 201588
I run Einstein, SETI, and POEM on my GPU. I've noticed Einstein and POEM give way more credit per hour than SETI does. This is irritating, as I wanted to give equal time to each project. Can't an agreement be reached like 1 credit per so many flops?
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
People over at SETI are only
People over at SETI are only getting half the credit they were two years ago. Credit here at Einstein has remained
relatively the same over the same time period.
look up creditscrew over at SETI its one of the neames they call creditnew over there.
I've asked over there and
I've asked over there and they seem to think they're being fair and everyone else isn't. There is no wrong and right amount of credit, you just need to use the same system, it can't be that hard, surely!
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
Peter Hucker wrote: Can't an
As it happens there is/was a definition, which you may review here.
Many of the projects don't make any direct attempt at all to adhere to this definition, though some probably try to make credit on their project bear a recognizable relation to some reference. I suspect strongly that some have deliberately give extra high credit in hopes of luring users.
That does not, I think, describe Einstein.
As mentioned above, a side effect of how CreditNew was actually implemented and behaves in practice has been that credit at SETI has wandered a long way from origin.
I suggest you take this up elsewhere than Einstein.
archae86 wrote:Peter Hucker
Where is "elsewhere"? All
Where is "elsewhere"? All the projects have to simply agree on a credit speed. It's that simple.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
Elsewhere is probably it
Elsewhere is probably it developers bulletin board, where they have had an ongoing discussion regarding credits.
Unfortunately, Seti has it's credit system directed by Dr. A (if I am remembering correctly)
The argument has been brought up many times to him about what is wrong with their credit system and how to improve it.
Unfortunately, he doesn't think there is a problem and since he is busy with other projects, it is not even a low priority for him.
Einstein, in my opinion, has the right approach and wish it was copied by several other projects including Seti.
My 2 cents.
Zalster
Well looking at the link
Well looking at the link given above, and estimating the points I should have got recently, it looks like Einstein is correct and SETI isn't. Depending on how annoyed I get, I may well withdraw from the projects that aren't doing things properly.
The other thing that annoys me is projects that aren't utilising GPUs, which are the main part of my (and presumably a lot of people's) computing power.
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
I've adjusted my settings so
I've adjusted my settings so SETI gets a tenth of the points of the other projects, that seems to make them get the same computing time. And the same goes for those that don't make GPU work (although they're listed as doing so on the master list, which is why I signed up to them in the first place).
If this page takes an hour to load, reduce posts per page to 20 in your settings, then the tinpot 486 Einstein uses can handle it.
Peter Hucker wrote:Where is
Why "must" they agree? Projects admins assign credit as they see fit and they may have good reason to change them. It may be easier to just stop adding them up across projects.
If you want to engage in the never ending credit debate ... version four is
http://boinc.berkeley.edu/dev/forum_thread.php?id=10953#68956
Peter Hucker wrote:The other
Productive use of GPUs involves, at the minimum, a suitable problem, suitable tools, and (usually) appreciable application development effort aimed specifically at the narrow problem of effective GPU use. There is no universal truth that problems are suitable for parallel processing in a way that makes fruitful use of a GPU remotely possible, and a smaller set of those which are potential parallel runners fit the offered architectures moderately well. Still fewer implement that parallel efficient approach with modest application of the available development tools.
Yes, when a good fit is found for an interesting problem they offer amazing advantages, but if everything could run well on them supercomputers would not be built with nearly so strong a proportion of conventional CPUs (and problems run on them already need to be able to fit to significant parallel running to make it sensible to use the modern sort of supercomputer at all.)