So is the new application going to be optimized for AVX or FMA instructions?
Reply
Quote
I see gene@home PC-IM v1.03....
No new thread?
Reply
Quote
valtercProject administrator
Project tester
Send message
Joined: 30 Oct 13
Posts: 623
Credit: 34,677,535
RAC: 1
I see gene@home PC-IM v1.03....
No new thread?
It's the same application as v1.02 (just one library has been made static)
Reply
Quote
valtercProject administrator
Project tester
Send message
Joined: 30 Oct 13
Posts: 623
Credit: 34,677,535
RAC: 1
So is the new application going to be optimized for AVX or FMA instructions?
They are ready. I just want to wait a little bit before making them available (just to see if there are problems with the base one)
Reply
Quote
valtercProject administrator
Project tester
Send message
Joined: 30 Oct 13
Posts: 623
Credit: 34,677,535
RAC: 1
thanks for the updates and glad new app works. i dont have a clue why people are complaining as u adopted the new boinc credit system and as for all the gridcoin users itll all balance out within 30 days anyway. im not doing solo atm but end of this month i will be.
i did notice your app does take longer, and on applications page for my host i averaged 6.74 gflops and now its under 3 gflops. could be as the information is more accurate now as youve mentioned. boinc rr simulation reflects that as well but thats normal. either way thanks for getting work units up, cheers.
I have multiplied by 1.5 the time estimation, let's see what happens (it may take some time to have some feedback)
Reply
Quote
If you look for strange things....
http://gene.disi.unitn.it/test/workunit.php?wuid=3574433
Reply
Quote
valtercProject administrator
Project tester
Send message
Joined: 30 Oct 13
Posts: 623
Credit: 34,677,535
RAC: 1
If you look for strange things....
http://gene.disi.unitn.it/test/workunit.php?wuid=3574433
Yes, I know, I really don't know why, it seems that the some system component deletes the input file when there is no need to do it. I hope to find a solution soon.......
[edit] we built a multi-threaded version of our work-generator. Sometimes, because of the same timestamps, it generates different input files but with the same file name. This makes boinc crazy. We have a solution for this. I hope that fixing this will also fix the problem you noticed
[edit, again] we fixed the 'same name, different content' problem. Let'see....
Reply
Quote
Do you guys have the system down on purpose? If so do you have an et when it will be back?
Reply
Quote
valtercProject administrator
Project tester
Send message
Joined: 30 Oct 13
Posts: 623
Credit: 34,677,535
RAC: 1
Do you guys have the system down on purpose? If so do you have an et when it will be back?
We still have a problem with our work-generator (the previous fix didn't work). The effect is that some workunits error out (with the download failure error), many of them. We think we have solved this problem. We will apply the fix tomorrow in the morning. In the meantime I will distribute just a few hundreds workunits.
Reply
Quote
valtercProject administrator
Project tester
Send message
Joined: 30 Oct 13
Posts: 623
Credit: 34,677,535
RAC: 1
The new work generator seems to do his job the right way. We started again to distribute work. Thank you all for your understanding.
Reply
Quote
Hi, good job on the new App - I have had lots of them validated in the last three days.
Hovever, the credit system seems still broken.
The last 300 WUs show roughly 20 credits for 1400s on an i5 - which is around 51creds/hour/core
Previously I got 240 cres/hour/core - almost factor 5 more!
So two questions:
- What is the target credit per WU which you want to show?
- What happens to the old WUs - will they get re-evaluated?
Thanks a lot and Cheers
Holger
____________
My stats
Reply
Quote
valtercProject administrator
Project tester
Send message
Joined: 30 Oct 13
Posts: 623
Credit: 34,677,535
RAC: 1
Hi, good job on the new App - I have had lots of them validated in the last three days.
Hovever, the credit system seems still broken.
The last 300 WUs show roughly 20 credits for 1400s on an i5 - which is around 51creds/hour/core
Previously I got 240 cres/hour/core - almost factor 5 more!
So two questions:
- What is the target credit per WU which you want to show?
- What happens to the old WUs - will they get re-evaluated?
Thanks a lot and Cheers
Holger
We don't have any target credit. Since the very beginning we used the default boinc server CreditNew system, so you should expect to get more or less the same credits as other CreditNew projects (like seti@home). During February this year (before we started having problems) for some (at least for me) unknown reasons the system started to grant a lot more credits than usual (and no one complained... ;) This behavior was probably related to a completely wrong workunit's time estimate.
To this end if I had to re-evaluate old (February) workunits I should probably decrease their assigned credit.
We discussed about this and decide to do nothing.
You may simple say that you were lucky to be here in February.
Reply
Quote
Was the system back in February granting more points all over sudden, or were there more users?
https://boincstats.com/en/stats/150/project/detail/credit
Looks like TN-Grid was added to the DC Vault in early January this year.
From there it skyrocketed.
The lower credits per workunit now, are caused by the lowered amount of estimated FLOPS per unit.
Edit: The DC-Vault added TN-Grid to the Vault on 6th of January, many started to try the project earlier.
http://www.team-ninja.com/vbulletin/showthread.php?t=59064
https://boincstats.com/en/stats/150/project/detail/user
Reply
Quote
I looked back and yeah some strange things happened in February.
Sorry admin for my earlier post.
Reply
Quote
valtercProject administrator
Project tester
Send message
Joined: 30 Oct 13
Posts: 623
Credit: 34,677,535
RAC: 1
Was the system back in February granting more points all over sudden, or were there more users?
https://boincstats.com/en/stats/150/project/detail/credit
Looks like TN-Grid was added to the DC Vault in early January this year.
From there it skyrocketed.
The lower credits per workunit now, are caused by the lowered amount of estimated FLOPS per unit.
Edit: The DC-Vault added TN-Grid to the Vault on 6th of January, many started to try the project earlier.
http://www.team-ninja.com/vbulletin/showthread.php?t=59064
https://boincstats.com/en/stats/150/project/detail/user
A lot of things happened since December
* Public invitation code + badges + challenges etc. -> a lot more users
* Two rounds of app optimization
* Other organisms (Ec, Vv, d5-Ec, etc.)
The time estimate, at the beginning, was a polynomial interpolator based on the Ec dataset and the various algorithm parameters. It wasn't perfect but it worked. With version 0.11 (probably also because of the mixing of different organisms/parameters) the time estimates were completely wrong so we had to find another way to compute them. Now the expected FLOPS are more accurate.
Reply
Quote