Folding Forum

On my GTX 660 Ti, I haven't gotten Core_17 on either normal or advanced for a couple of weeks or so (currently P8018 on Advanced, Win7 64-bit). That is OK with me if that is what FAH needs, but maybe it is a glitch (?).

How about you set all your slots to finish. Uninstall F@H (including the data files). Reinstall it and use the default settings for the GPU Slot (if they are correctly mapped to the physical GPU) and see what the result is.

PantherX wrote:How about you set all your slots to finish. Uninstall F@H (including the data files). Reinstall it and use the default settings for the GPU Slot (if they are correctly mapped to the physical GPU) and see what the result is.

Thanks, I tried that, but still picked up a Core_15 (P8018) on Advanced. It is no big deal for me, I will just wait until they get the server problem fixed, whatever it is.

EDIT: I think I have fixed it another way. I set the Beta flag, and picked up a Core_17, and before it was finished I set the Advanced flag. It then picked up another Core_17 (P8900). So maybe it changed servers, or else identifies itself differently now?

I experimented yesterday...

(1) Following the advice posted earlier, I set one GPU slot to finish. After it was done I paused the other two and shut the client down. I restarted the client and then restarted the finished GPU slot which promptly downloaded a p8108.

(2) I then deleted all three GPU slots and dumped the p8108s, sorry I don't like doing that... uninstalled the client and deleted the data folder. Reinstalled the client which set itself up with 1 CPU slot and 3 GPU slots. All three GPU slots downloaded p8108.

(3) Deleted all of the slots and dumped the WUs, sorry again. I setup three GPU slots with the client-type beta flag. All three slots download p8900, strange because I don't think p8900 is a beta project. After the slots are running for a while I reset the client-type flag to advanced. Overnight all three slots finished the WUs and download three more p8900.

Conclusion? From what I see It seems that the client develops an affinity for a work server and you get work from that server. If p8108 is the highest priority right now then according to Pande I should have downloaded those overnight regardless of the flag setting.

Will be interesting to see how long I get p8900 before going back to p8108...

Thats a very bad way to experiment by abandoning work units! Anyone here could have told you there is no such thing as server affinity. Now all those wasted work units have to expire many days from now, slowing down the science. It also slows down each following generation in that line of work units because the next WU isn't sent until the wasted WUs get reassigned, completed and returned.

The Assignment Server simply reads your config and assigns work based on the server's weighting. See the Server Status page for the weight settings, and also the percent assigned column.

7im wrote:Thats a very bad way to experiment by abandoning work units! Anyone here could have told you there is no such thing as server affinity. Now all those wasted work units have to expire many days from now, slowing down the science. It also slows down each following generation in that line of work units because the next WU isn't sent until the wasted WUs get reassigned, completed and returned.

The Assignment Server simply reads your config and assigns work based on the server's weighting. See the Server Status page for the weight settings, and also the percent assigned column.

Yes I agree and I am sorry, however I can't help but feel that something is not working correctly. At a stretch from the first week of November through the first week of December I ran nothing but p8108 on all three of my GPUs, completing either 7 or 8 WUs per day. Concurrently, other members of my team folding with GPUs were seeing a mix of Core15 and Core17 WUs. At some point luck of the draw or priority no longer explains actual WU assignment in my case.

I am not complaining and I will continue to donate my electricity and gear because I believe in the work. I also have skin in the game with cancer in the family and I worry about my children. I hope that cumulatively the work that gets done will save my kids some of the troubles our family has experienced in the past.

Are you using the same client settings as your team members?

7im wrote:Are you using the same client settings as your team members?

IDK, I will have to check.

Another question that comes to mind is the contribution choice (cancer, alzhiemers, etc). I selected cancer, how does that weigh into WU assignment?

AFAIK the choice is not yet fully implemented server side but there has been at least one other report that specifying a cause may result in not getting certain types of work. Try switching it to Any and see if that helps.

7im wrote:Thats a very bad way to experiment by abandoning work units! Anyone here could have told you there is no such thing as server affinity. Now all those wasted work units have to expire many days from now, slowing down the science. It also slows down each following generation in that line of work units because the next WU isn't sent until the wasted WUs get reassigned, completed and returned.

The Assignment Server simply reads your config and assigns work based on the server's weighting. See the Server Status page for the weight settings, and also the percent assigned column.

Really? Was he a naughty boy? Did you tell him there was no such thing as server affinity? And how then do you explain his findings? There is significant evidence that clients do indeed get "stuck" on particular work.

And your signature is not an excuse to be high and mighty.

Blasko9 wrote:I setup three GPU slots with the client-type beta flag. All three slots download p8900, strange because I don't think p8900 is a beta project. After the slots are running for a while I reset the client-type flag to advanced. Overnight all three slots finished the WUs and download three more p8900.

That is what I see, except that after getting the additional P8900 with the Advanced flag, it reverted back to a Core_15 work unit again.

So I get P8900 reliably only if I set the beta flag, even though it is not supposed to be a beta project insofar as I know. Whether that is being implemented by accident or on purpose by PG to ration the work units I don't know, but the simple thing to do is just set the beta flag. If there is a better way, I would like to know. (But I think I will go back to Advanced, and see if I get a mix of work units or not.)

Specific servers may be listed as having multiple projects. This information is shown on the project summary page. Although the project summary is designed to automatically stop showing projects when a specific server is non-functional, it doesn't do a very good job of suspending the listings of projects which may have zero WUs to assign. (I don't know why.)

The server status page shows which servers are active and gives information about how many WUs are available. Information is also given about whether the server is configured for standard FAH projects, for Adv projects, for Beta projects, and perhaps Internal projects as well as the platform (operating system) for which it's configured. Unfortunately it's impossible to know how many WUs fall into those classifications.

P8900 is the only project being assigned by 171.64.65.69. That server is configured for the possible assignments to assign projects from to Windows or Linux requesting Advanced, Beta or Internal projects or to Windows for FAH projects. Many of those seven possibilities may have WUs and many may not. About the only useful information found there is that a Linux requesting a FAH assignment (i.e.: non-beta, non-adv) won't get one. P8900 may be moved from one classification to another without any visible changes.

Other GPU servers that are assigning work include 171.67.108.44, 171.67.108.11, 171.67.108.21, and 171.67.108.142 but the first three have only low priority projects so you won't get any of those projects unless you have a GPU that can't be accommodated by 171.64.65.69 or 171.67.108.142 so p8018 may be the only other non-beta project available to certain type(s) of GPUs.

I believe these statements are accurate at the time I looked at the data. I do not have access to a history log indicating if any applicable projects were added or went out of service recently, which might be especially important if the changes were brief.

With selections that limited, concluding specific "facts" about the assignment process won't produce reliable conclusions.

While Nvidia seems to have problems getting P8900-wu's, they atleast appears to get a steady supply of other GPU-work. :zzzzzzz:

For Amd on the other hand, was sure there was roughly 1100 active Amd's last week, but now it's dropped to only 326.
So is this a problem with the stats for Amd-cards, a lack of work or another reason?

Rattledagger wrote:So is this a problem with the stats for Amd-cards, a lack of work or another reason?

Your guess is as good as mine ... unless we start getting reports from donors who are failing to get work.

Remember to compare the anticipated duration of a WU on specific hardware with the timeout that drops GPUs out of the active category. (I would have to look up those numbers before making an educated guess.)

Rattledagger wrote: For Amd on the other hand, was sure there was roughly 1100 active Amd's last week, but now it's dropped to only 326.

Can I ask where you are getting that stat? If true (and I'm not doubting you) that is incredibly low.