It looks like my AMD 6000+ machine is taking a bit of a performance hit with the S5R3's. They're taking just a bit over half as long to do as it would a monster S5R2, but only giving me about one-third of the credit.
This box is running Kubuntu 7.04 Linux.
Two comments on your experience.
1. Of your three S5R3 results, one has successful quorum partner--a Core 2 Duo running XP. That person saw about a 3% degradation from their S5R2 resuls on your worst (slowest) result, much less degradation than you report.
2. Your experience is another example of strong CPU time variation for results in the same series with the same credit:
all three results are of the form h1_0545.10_S5R2__rrr_S5R3a_n
Thanks -- very helpful! Other users, please post your observations.
Bruce
My host 982234 a Core 2 Q6600 Quad currently OC'c to 3.24 GHz has been processing 655.5 credit results in about 91000 CPU seconds recently using beta 4.40
Its first S5R3 results are on the production 4.01 science ap, and took 29130 CPU seconds to claim 220.28 credits
4.40 S5R2 138.8 CPU seconds/credit
4.01 S5R3 132.2 CPU seconds/credit
so on this tiny sample, I got 5% more credit/hour on S5R3 4.01 than on S5R2 4.40.
caveats:
#1
It will be at least another day before I can begin to say whether I'm getting the large variation from result to result reported by others, probably longer, as my host has downloaded all the consecutive results from
h1_0542.70_S5R2__157_S5R3a_2 to
h1_0542.70_S5R2__111_S5R3a_0
so I'm only sampling sequential results in that part of the sequence, and can only do twelve a day at most on this host.
#2
As others have pointed out, 4.40 already had a large penalty compared to the dominant production S5R2 ap in credit, presumably because of debugging code added late in the run. I don't have a good number for this penalty, but believe it was on the order of 15%.
The second result even took a bit longer for the same credit. No way this is good performance...
I can't say anything about the question if they validate, since I have a wingman who keeps getting client errors, but I had no indication of trouble. Crunching is just slow- or else this box is getting too little credit.
We generated the first Workunits of S5R3, some of you probably already got them.
The code of the Apps hasn't changed very much compared to the 4.4x Apps of S5R2. If you would run the same S5R2 workunit with both of them you'd find the S5R3 App being slightly faster.
As some paramaters of the Workunit split-up have changed, we needed to adjust our "credit normalization factor" (roughly credit per template). We may have to tune this again in the next few days. Please watch your credit per hour ratio and report here (and also take into account a normal fluctuation of +-5%).
BM
After the first unit crunched I needed only 17,5 hours thats half the time the S5R2 units but I only got a third of the credits, you have to ajust the credit value more than 5% its around 30 % the ajustment needed to get to the same value of last units.( 124000 seconds = 640 credits, now 62000seconds = 219 credits)
We generated the first Workunits of S5R3, some of you probably already got them.
The code of the Apps hasn't changed very much compared to the 4.4x Apps of S5R2. If you would run the same S5R2 workunit with both of them you'd find the S5R3 App being slightly faster.
As some paramaters of the Workunit split-up have changed, we needed to adjust our "credit normalization factor" (roughly credit per template). We may have to tune this again in the next few days. Please watch your credit per hour ratio and report here (and also take into account a normal fluctuation of +-5%).
BM
After the first unit crunched I needed only 17,5 hours thats half the time the S5R2 units but I only got a third of the credits, you have to ajust the credit value more than 5% its around 30 % the ajustment needed to get to the same value of last units.( 124000 seconds = 640 credits, now 62000seconds = 219 credits)
To be more accurate the data are 18.6 sec/ credit in S5R2 and 12.9 sec/ credit in S5R3a on a Mac Book Pro dual core, Darwin
After the first unit crunched I needed only 17,5 hours thats half the time the S5R2 units but I only got a third of the credits, you have to ajust the credit value more than 5% its around 30 % the ajustment needed to get to the same value of last units.( 124000 seconds = 640 credits, now 62000seconds = 219 credits)
To be more accurate the data are 18.6 sec/ credit in S5R2 and 12.9 sec/ credit in S5R3a on a Mac Book Pro dual core, Darwin
I'm not sure what you did wrong, but those numbers are impossible. What you've listed are 193/279credit/hour = 4600/6700 credit/day.
The rates of the person you quoted are 193 and 283 sec/credit.
After the first unit crunched I needed only 17,5 hours thats half the time the S5R2 units but I only got a third of the credits, you have to ajust the credit value more than 5% its around 30 % the ajustment needed to get to the same value of last units.( 124000 seconds = 640 credits, now 62000seconds = 219 credits)
To be more accurate the data are 18.6 sec/ credit in S5R2 and 12.9 sec/ credit in S5R3a on a Mac Book Pro dual core, Darwin
I'm not sure what you did wrong, but those numbers are impossible. What you've listed are 193/279credit/hour = 4600/6700 credit/day.
The rates of the person you quoted are 193 and 283 sec/credit.
Sorry you are right, but still its a huge difference
RE: RE: It looks like my
)
Yeah, I see that now. I hadn't looked before.
The strange part. . .
With the monster 'R2's, this machine used to significantly outperform the XP-loaded Core 2 E6700's. This is quite the reversal of fortunes.
RE: Thanks -- very helpful!
)
My host 982234 a Core 2 Q6600 Quad currently OC'c to 3.24 GHz has been processing 655.5 credit results in about 91000 CPU seconds recently using beta 4.40
Its first S5R3 results are on the production 4.01 science ap, and took 29130 CPU seconds to claim 220.28 credits
4.40 S5R2 138.8 CPU seconds/credit
4.01 S5R3 132.2 CPU seconds/credit
so on this tiny sample, I got 5% more credit/hour on S5R3 4.01 than on S5R2 4.40.
caveats:
#1
It will be at least another day before I can begin to say whether I'm getting the large variation from result to result reported by others, probably longer, as my host has downloaded all the consecutive results from
h1_0542.70_S5R2__157_S5R3a_2 to
h1_0542.70_S5R2__111_S5R3a_0
so I'm only sampling sequential results in that part of the sequence, and can only do twelve a day at most on this host.
#2
As others have pointed out, 4.40 already had a large penalty compared to the dominant production S5R2 ap in credit, presumably because of debugging code added late in the run. I don't have a good number for this penalty, but believe it was on the order of 15%.
First 2 s5r3's crunched on
)
First 2 s5r3's crunched on one of my hosts. A64x2 WIndows. 276sec/credit/core in R3 vs 251sec/credit/core in R2. At 10% slow down.
On one of my hosts I found a
)
On one of my hosts I found a big variation in completion times:
h1_0531.15_S5R2__159_S5R3a 79,618.06
h1_0531.15_S5R2__141_S5R3a 80,730.02
h1_0531.15_S5R2__128_S5R3a 87,645.56
h1_0531.15_S5R2__118_S5R3a 91,048.45
h1_0531.15_S5R2__100_S5R3a 82,022.61
That's more than 12% difference.
ok, it's a Pentium D, but as far as I know that's 2 cores and no hyperthreading...
Udo
The second result even took a
)
The second result even took a bit longer for the same credit. No way this is good performance...
I can't say anything about the question if they validate, since I have a wingman who keeps getting client errors, but I had no indication of trouble. Crunching is just slow- or else this box is getting too little credit.
RE: We generated the first
)
After the first unit crunched I needed only 17,5 hours thats half the time the S5R2 units but I only got a third of the credits, you have to ajust the credit value more than 5% its around 30 % the ajustment needed to get to the same value of last units.( 124000 seconds = 640 credits, now 62000seconds = 219 credits)
About my margin. And since
)
About my margin. And since your box is an Intel Darwin whereas mine is an AMD running Windows it's not CPU/OS related, either...
RE: RE: We generated the
)
To be more accurate the data are 18.6 sec/ credit in S5R2 and 12.9 sec/ credit in S5R3a on a Mac Book Pro dual core, Darwin
RE: RE: After the first
)
I'm not sure what you did wrong, but those numbers are impossible. What you've listed are 193/279credit/hour = 4600/6700 credit/day.
The rates of the person you quoted are 193 and 283 sec/credit.
RE: RE: RE: After the
)
Sorry you are right, but still its a huge difference