With the albert4.37 akos optimized you got more output disabling HT. With the new S5 application does that still hold true? ( Still have days of 4.37 til I get there) Any comparasins HT enabled/disabled anyone can give please?
Maybe this is a better place to post my Question rather than the Bruce Allen info thread, I hope :)
I finished and got credit for 5 of the S5 units so far on my P4 2.4Gig. All taking very close to 1.75 hours and averageing 11.98 credits per hour. This computer was agerageing 29.01855 credits per hour on the S4 units. With the S5 it is about the same per hour as with Rosetta.
The other system Celeron D 2.93 is still running the S4 units, averageing 55 min per unit and 36.091 credits per hour.
Credits per hour on the S4 units was calculated over 24 units on each system. Will have to get more of the S5 units (includeing longer ones) to get a good average.
Cheers
Ray
EDIT
The calculation of S4 times can be found at this link.
Feel free to enter your oun times into that for comparison.
Does that mean your computer will do the work faster and get
equal or more credit than with the stock apps.?
Such as happens over at S@H?
An optimised einstein app would claim the same as a stock app because the credits are now decided server side, not by your client. This is good because even tho an optomised app gets the work done faster it will still get the same amount of credit granted as the stock app. And we all know that more work done in less time equalls more work done in a day. And if your after credits then you also get more credit per day. And no optimised client is needed to get fair credit.
Have one S5 WU 35% done with 4hr 28min CPU-time. That adds up to about 12½ hours total for the 100%.
12½hr is too much IMO.
The frequent checkpoints in this project justifies it a bit.
But sometimes my crunching computer is only on for a few hours a day (if on at all), and with a couple of projects more claiming cycles 12hrs is a VERY long time for a single WU.
I know the increased WU-sized helps the database, and you want to make room for more cruchers. But with the S4 WU's and Akos optimized S41 app I crunched in about 1hr. The crunch time is now 12 times as high. That way I only have a result and upload 1/12th of the time compared to S4.
If the 12x increase repeats with others, then this project should be able to handle 12 times more crunchers, at least database-wise.
Don't want to be negative, but I can't see E@H increasing its host by a factor of 12 the next months.
Perhaps you should aim a little lower. WU-times increased by 2-4 times perhaps?
I think the final goal is to have the s5 work units run as long as old s4 classic ones did. My thinking on the subject is as follows.
S4 akos apps were ~5-7x as fast as 4.37.
S4 offical beta apps incorperated some of AKos's work and were ~3x faster than stock.
S5 work units have 5x as much work as s4 ones.
With the current s5 app work units take 2x long as stock s4 ones. With the remainder of the akos optimizations worked into the app runtime will be about the same as the start of s4.
Also while most people on this board were running optimized apps, the majority of crunchers in general were probably unaware of anything akos was doing at all and still running standard apps. Consequently the ammount of headroom from the bigger work units is alot smaller than the 12x it would be if everyone was running akos apps in s4.
This is still better than when I joined in June 2005, average time them was 12 hours and average credit 75 per unit as I remember. That worked out to about 6.25 credits per hour.
My P4 only seems to be able to get S5 units now, but the faster Celeron just got 17 of the S4 units. Looks like the system is selecting certain system types to get the S5 first. All the S5 units so far are the short ones on my system, about 100 Min. each, should get some longer ones soon.
Ray
And now it looks like the credit chasers are about to start a repeat preformance at einstein with the start of the s5 run , lol. Wonder what project they will go to next. There is no other project that has optimised apps for them to run to, oh my. What ever will they do. Guess they will have to learn to enjoy doing the science like the rest of us.
If you're getting more credit with less cpu time using optimized apps,
than those of us who crunch by the rules with stock apps, you don't
consider that cheating? or undermining the system?
With the albert4.37 akos
)
With the albert4.37 akos optimized you got more output disabling HT. With the new S5 application does that still hold true? ( Still have days of 4.37 til I get there) Any comparasins HT enabled/disabled anyone can give please?
Maybe this is a better place to post my Question rather than the Bruce Allen info thread, I hope :)
I finished and got credit for
)
I finished and got credit for 5 of the S5 units so far on my P4 2.4Gig. All taking very close to 1.75 hours and averageing 11.98 credits per hour. This computer was agerageing 29.01855 credits per hour on the S4 units. With the S5 it is about the same per hour as with Rosetta.
The other system Celeron D 2.93 is still running the S4 units, averageing 55 min per unit and 36.091 credits per hour.
Credits per hour on the S4 units was calculated over 24 units on each system. Will have to get more of the S5 units (includeing longer ones) to get a good average.
Cheers
Ray
EDIT
The calculation of S4 times can be found at this link.
Feel free to enter your oun times into that for comparison.
Try the Pizza@Home project, good crunching.
First long S5 WU done. 8:04 h
)
First long S5 WU done. 8:04 h on stock AMD 64 3500+.
http://einsteinathome.org/task/34524825
RE: We need Optomized
)
An optimised einstein app would claim the same as a stock app because the credits are now decided server side, not by your client. This is good because even tho an optomised app gets the work done faster it will still get the same amount of credit granted as the stock app. And we all know that more work done in less time equalls more work done in a day. And if your after credits then you also get more credit per day. And no optimised client is needed to get fair credit.
98SE XP2500+ @ 2.1 GHz Boinc v5.8.8
RE: We need Optomized
)
No, we don't need them, but we would like them to get more units a day completed.
Ray
Try the Pizza@Home project, good crunching.
Have one S5 WU 35% done with
)
Have one S5 WU 35% done with 4hr 28min CPU-time. That adds up to about 12½ hours total for the 100%.
12½hr is too much IMO.
The frequent checkpoints in this project justifies it a bit.
But sometimes my crunching computer is only on for a few hours a day (if on at all), and with a couple of projects more claiming cycles 12hrs is a VERY long time for a single WU.
I know the increased WU-sized helps the database, and you want to make room for more cruchers. But with the S4 WU's and Akos optimized S41 app I crunched in about 1hr. The crunch time is now 12 times as high. That way I only have a result and upload 1/12th of the time compared to S4.
If the 12x increase repeats with others, then this project should be able to handle 12 times more crunchers, at least database-wise.
Don't want to be negative, but I can't see E@H increasing its host by a factor of 12 the next months.
Perhaps you should aim a little lower. WU-times increased by 2-4 times perhaps?
- Knorr
I think the final goal is to
)
I think the final goal is to have the s5 work units run as long as old s4 classic ones did. My thinking on the subject is as follows.
S4 akos apps were ~5-7x as fast as 4.37.
S4 offical beta apps incorperated some of AKos's work and were ~3x faster than stock.
S5 work units have 5x as much work as s4 ones.
With the current s5 app work units take 2x long as stock s4 ones. With the remainder of the akos optimizations worked into the app runtime will be about the same as the start of s4.
Also while most people on this board were running optimized apps, the majority of crunchers in general were probably unaware of anything akos was doing at all and still running standard apps. Consequently the ammount of headroom from the bigger work units is alot smaller than the 12x it would be if everyone was running akos apps in s4.
This is still better than
)
This is still better than when I joined in June 2005, average time them was 12 hours and average credit 75 per unit as I remember. That worked out to about 6.25 credits per hour.
My P4 only seems to be able to get S5 units now, but the faster Celeron just got 17 of the S4 units. Looks like the system is selecting certain system types to get the S5 first. All the S5 units so far are the short ones on my system, about 100 Min. each, should get some longer ones soon.
Ray
Try the Pizza@Home project, good crunching.
And now it looks like the
)
And now it looks like the credit chasers are about to start a repeat preformance at einstein with the start of the s5 run , lol. Wonder what project they will go to next. There is no other project that has optimised apps for them to run to, oh my. What ever will they do. Guess they will have to learn to enjoy doing the science like the rest of us.
Taken from another post.
If you're getting more credit
)
If you're getting more credit with less cpu time using optimized apps,
than those of us who crunch by the rules with stock apps, you don't
consider that cheating? or undermining the system?