You obviously have a very decent machine with >10K total credits after just a couple of days :-).
Quote:
What is the Avg. Work Done? How is it calculated?
If you head over to the BOINC Website and click on the 'documentation' link under the 'Volunteer' heading near the top of the page, you will be taken to a Wiki where you can find information on many of these sorts of questions. If you scroll down through all the links on the main page of the Wiki, you will find one in the miscellaneous section near the bottom of the page which is directly about credit. While you are on the main page, take a look at all the other available topics as well in case you find other things of interest.
Quote:
Why is it constantly going up even though my computer is not increasing in speed?
Recent Average Credit (RAC) is a moving average taken over 10 days (I think) of your machine's production. On day one, after first joining, its value will be the average of 9 days of zero production and one day of full production. On day two, .... - I'm sure you get the picture.
It is also being affected by delays in granting credit if your 'wingman' is a bit slower in returning results than yourself. It wouldn't be unusual for it to take a couple of weeks or more to reach a semblance of stability.
It is also being affected by delays in granting credit if your 'wingman' is a bit slower in returning results than yourself.
Do we ( attempt to ) match 'like with like' with the locality scheduling features ? He's certainly at the top end of performance, low turnaround times indeed. He may peak RAC in the top 100 I think.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
It is also being affected by delays in granting credit if your 'wingman' is a bit slower in returning results than yourself.
Do we ( attempt to ) match 'like with like' with the locality scheduling features ?
No. Locality scheduling just keeps track of which hosts already have which LIGO data files and so tries to minimise unnecessary downloads of additional LIGO data wherever possible. It may well be that a very fast host and a very slow host are paired up with the same LIGO data files and (together with other hosts as well) are crunching on the same series of tasks.
Quote:
He's certainly at the top end of performance, low turnaround times indeed. He may peak RAC in the top 100 I think.
Even if so, there really is no problem with being paired up with an old, slow host. If you think about it, it's much more preferable to be paired up with a slow host that returns one result every two days than with a fast host with an 8 day cache :-). Yes, they do exist :-).
Why is it constantly going up even though my computer is not increasing in speed?
Recent Average Credit (RAC) is a moving average taken over 10 days (I think) of your machine's production. On day one, after first joining, its value will be the average of 9 days of zero production and one day of full production. On day two, .... - I'm sure you get the picture.
In your example, the RAC would only go up. But it doesn't. You forgot about the decay. ;-)
Each time new credit is added, the RAC is recalculated. Once every week it's also divided in half. This latter counts for RAC to go down, especially when no credit is added or when you stop crunching completely.
For the complete drop-down (albeit somewhat old and perhaps not completely correct anymore versus the new credit system), see RAC in the Unofficial Wiki.
Thanks guys! I also found the built-in benchmarker so I will test out a few different configurations tomorrow to see what clock speeds make for the best crunching!
see what clock speeds make for the best crunching!
Apart from pure speed, decent memory size and bandwidth, a major determinant is floating point efficiency. There has been much work devoted to optimising the relevant code sections. For GW tasks at least, one part is processor bound, the other memory bound. I can't remember where we are with the Intel vs AMD debate, probably on the Intel side last I recall. Of course there is the Power PC or whatever Apple is serving these days. Also while the screensavers are nice, they do slow the crunching.
@Gary :
Quote:
Even if so, there really is no problem with being paired up with an old, slow host. If you think about it, it's much more preferable to be paired up with a slow host that returns one result every two days than with a fast host with an 8 day cache :-). Yes, they do exist :-).
Well put. I never thought of it that way.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Aside from BSOD's and freezing, if a tiny calculation error takes place, what signs or errors should I look for?
I'm not aware of the E@H apps giving BSOD's/freezing per se. The calculation errors ( if any ) get picked up back at the server side ( ie. the E@H validators, after you have returned the workunit ) by various tests including with respect to other quorum members' results for that work unit - a given unit of work is replicated and sent to several distinct hosts for precisely that purpose.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Well my computer is being a poopy pants and not wanting to run at 4.2GHz anymore, I think silicon degradation has permanently forced me down to 4GHz. On the plus side I can now run my memory at 1905MHz CL7 as before it was 1676MHz CL6. I'll work on my memory timings a little bit this week and try to get them down quite a bit to make up for it.
What is Avg Work Done (Credits)?
)
Hi Kuntz - welcome to Einstein!
You obviously have a very decent machine with >10K total credits after just a couple of days :-).
If you head over to the BOINC Website and click on the 'documentation' link under the 'Volunteer' heading near the top of the page, you will be taken to a Wiki where you can find information on many of these sorts of questions. If you scroll down through all the links on the main page of the Wiki, you will find one in the miscellaneous section near the bottom of the page which is directly about credit. While you are on the main page, take a look at all the other available topics as well in case you find other things of interest.
Recent Average Credit (RAC) is a moving average taken over 10 days (I think) of your machine's production. On day one, after first joining, its value will be the average of 9 days of zero production and one day of full production. On day two, .... - I'm sure you get the picture.
It is also being affected by delays in granting credit if your 'wingman' is a bit slower in returning results than yourself. It wouldn't be unusual for it to take a couple of weeks or more to reach a semblance of stability.
Cheers,
Gary.
RE: It is also being
)
Do we ( attempt to ) match 'like with like' with the locality scheduling features ? He's certainly at the top end of performance, low turnaround times indeed. He may peak RAC in the top 100 I think.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
RE: RE: It is also being
)
No. Locality scheduling just keeps track of which hosts already have which LIGO data files and so tries to minimise unnecessary downloads of additional LIGO data wherever possible. It may well be that a very fast host and a very slow host are paired up with the same LIGO data files and (together with other hosts as well) are crunching on the same series of tasks.
Even if so, there really is no problem with being paired up with an old, slow host. If you think about it, it's much more preferable to be paired up with a slow host that returns one result every two days than with a fast host with an 8 day cache :-). Yes, they do exist :-).
Cheers,
Gary.
RE: RE: Why is it
)
In your example, the RAC would only go up. But it doesn't. You forgot about the decay. ;-)
Each time new credit is added, the RAC is recalculated. Once every week it's also divided in half. This latter counts for RAC to go down, especially when no credit is added or when you stop crunching completely.
For the complete drop-down (albeit somewhat old and perhaps not completely correct anymore versus the new credit system), see RAC in the Unofficial Wiki.
Thanks guys! I also found the
)
Thanks guys! I also found the built-in benchmarker so I will test out a few different configurations tomorrow to see what clock speeds make for the best crunching!
RE: see what clock speeds
)
Apart from pure speed, decent memory size and bandwidth, a major determinant is floating point efficiency. There has been much work devoted to optimising the relevant code sections. For GW tasks at least, one part is processor bound, the other memory bound. I can't remember where we are with the Intel vs AMD debate, probably on the Intel side last I recall. Of course there is the Power PC or whatever Apple is serving these days. Also while the screensavers are nice, they do slow the crunching.
@Gary :
Well put. I never thought of it that way.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Aside from BSOD's and
)
Aside from BSOD's and freezing, if a tiny calculation error takes place, what signs or errors should I look for?
RE: Aside from BSOD's and
)
I'm not aware of the E@H apps giving BSOD's/freezing per se. The calculation errors ( if any ) get picked up back at the server side ( ie. the E@H validators, after you have returned the workunit ) by various tests including with respect to other quorum members' results for that work unit - a given unit of work is replicated and sent to several distinct hosts for precisely that purpose.
Cheers, Mike.
I have made this letter longer than usual because I lack the time to make it shorter ...
... and my other CPU is a Ryzen 5950X :-) Blaise Pascal
Well my computer is being a
)
Well my computer is being a poopy pants and not wanting to run at 4.2GHz anymore, I think silicon degradation has permanently forced me down to 4GHz. On the plus side I can now run my memory at 1905MHz CL7 as before it was 1676MHz CL6. I'll work on my memory timings a little bit this week and try to get them down quite a bit to make up for it.