Aurum wrote:There is a real need have some logic to better match the scale of a WU to the capabilities of the GPU.
Absolutely. The V7.4.4 client has a default setting of 10. There's a large percentage of Donors who don't change that setting, and since most of the new GPU projects are bigger than that, there's a large percentage of projects that are rarely assigned...
Or it could be that, as a whole, it's easier to find out how to get into Area 51 than it is to "just download and properly run" FAH...
Aurum wrote:There is a real need have some logic to better match the scale of a WU to the capabilities of the GPU.
Absolutely. The V7.4.4 client has a default setting of 10. There's a large percentage of Donors who don't change that setting, and since most of the new GPU projects are bigger than that, there's a large percentage of projects that are rarely assigned.
Current plan: The next released client will have a larger default setting so owners of projects that are are larger than 10 can be processed at a reasonable across-the-board rate.
BTW, all my folders are running v7.4.15 beta.
You seem to be saying the opposite of the problem I'm addressing. The donors need control over what jobs they get to optimize their folding farm performance. I think you implied in an earlier post that option max-packet-size has been deprecated and that the only value that's used is the default setting, e.g. Normal=10MB. If it worked that would go a long way towards letting donors make good use of older cards. Here you say "...default setting of 10. There's a large percentage of Donors who don't change that..." implying that maybe option max-packet-size does work but is not being used. Unless there's another to change the default from 10, but I haven't discovered it yet.
I've set max-packet-size=Big on my RX 480s but have yet to see a Big WU DLed, the last two have been 14.9 and 5 MB.
We also need an option dump-if-ETA-futile to eliminate WUs that cannot possibly be finished before their Expiration date.
Suppose you owned one of the almost all projects which can be assigned to the default client are suspended. How much science would you be getting done? What you would you do about it: {Hint: One option would be to disable the assignment restriction that's keeping your projects from being assigned.] The restriction can only be re-enabled when "everybody" upgrades to 7.4.15+. In any case, I'll bet all large projects will currently assign if max-packe-size=10.
I'm not sure which projects are being assigned if max-packet-size=5 (small). Do you have any data? I'm trying max-packet=7 on a "slow" GPU but I don't have any reportable results yet. My faster GPUs have no trouble with a "big" setting. (except maybe Project 10496)
In general the assignment server should decide which packets sizes to send to which GPUs, so a donor does not need to care.
There should be this option max-packet-size working when the user knows better and wants to configure as he likes.
The idea of a dump-if-ETA-futile also can be decided by the assignment server for a given GPU to not assign in this case.
An option for the donor to do this would only be a work around for what the assignment server or project did wrong in the first place.
@Foldy: I don't disagree with you, but the fundamental design of the assignment process is based on the relatively sparse GPU information that the client uploads to the server. While some future client could be developed that provides additional information and the Servers could be upgraded to process all that additional information, including tabulated data describing every GPU, I do not 3xpect that to ever happen. Just updating GPUS.txt to incorporate much more limited data when new GPUs are released doesn't happen quickly enough to satisfy everybody.
As I' sure you already know, the backlog of "it-would-be-nice" features grows faster than the list of essential improvements. Moreover, designing an automatic system that replaces your ability to tweak the assignment process won't be popular with advanced Donors. Designing a complex system that makes perfect assignments may be possible, but it would certainly violate the KISS principle and it's unlikely to happen.
Hi Bruce, I've been watching for a few days and I don't think max-packet-size works. Maybe there's not enough variety on the server for it to work. I had my RX 480s set to max-packet-size=50 but they just got some jobs ~2.5 MB. I've seen lots of the opposite setting max-packet-size=4, 5 or 6 and getting 19MB plus.
Aurum wrote:Hi Bruce, I've been watching for a few days and I don't think max-packet-size works.
As I said, I think it works for CPU assignments. In it's present form, it doesn't work on most GPU projects ... but there's a plan to make it work again.
