Feel free to have a look on this host.
It's a dual P4 2800 Xeon System (with hyperthreading on, so 4 CPU's are available) with 3 CPUs dedicated to E@H.
S5RI times: 18.000 - 19.000 sec. and 54 Cr.
S5R2 times: 89.000 - 90.000 sec. and 204 Cr.
seems fair to me...
Right now the only comparison I have is on this host.
S5R1 times 9200 - 9300 sec and 53 credit
S5R2 times 47000 - 49000 sec and 77 credit.
This is on a dual core AMD 4800+. I have 2 more in my cache that are calling for 29 hours so I'm interested in seeing how they come out stat wise. One of these is at 50% and time for completion should come down to 14 hours instead of 29. Guess my time calulator for units will need to relearn things.
edited to correct S5R2 label and time to completion update.
Is there going to be any limitation regarding BOINC client ? I still have some clients running Boinc 4.x.
I have, too, on some architectures nothing newer is available. No, no limitation planned other than what existed for S4 - it should be 4.19 or newer.
Our credit system will work even with older clients. The credit is fixed for each workunit on the server side, and measurements are taken so that the claimed credit should be what is granted at the end. For this to work one needs a newer client, with older clients the granted credit may be different from what is claimed, that's all.
And btw. yes, there is a large variation between runtimes of the different WUs. In principle they get longer with higher frequency. The credit rises proportionally, so if you want to compare run times of machines, look for Workunits that claim the same credit (and use newer clients). I'll try to get some chart for WU-time vs. frequency to post it here.
OK, I received two early S5R2 units (I have a dual processor). They came across with expected runtimes of roughly 12 and 14 hours. They ran for 42 and 44 hours, respectively (it's really depressing when the "time to completion" is going up instead of down). In S5R2 "faster hierarchical search", i was getting somewhere around 53-54 credits per WU, and they were taking 7-8 hours. For the two forty-plus hour runs, I'm getting under 100 credits each -- total of about 160 for the two. So, S5R1, I would get (conservatively) 500-550 credits for 86 hours of CPU work -- and now I'm getting 160.
I'm back to S5R1 work now, but I've downloaded another S5R2 workunit for later. This one shows as 149 hours (that's more than six days). I'm dreading what will happen when it starts.
Bernd and company, you might want to rethink the size of these WUs. I have a pretty big fast machine; most people won't be able to finish the WUs within the deadlines. And the credit is REALLY off -- not just a little!
is it possible that the Linux application is much faster than the Win-App
Would be quite surprising, the output from gcc on Linux is usually slightly slower than what we get from the M$ compiler, but is not impossible. Honestly during the past weeks we have been so busy fixing mere showstoppers that I didn't do a platform / compiler comparison with the latest code versions. Thanks for pointing me to that, I'll look into it.
OK, I received two early S5R2 units (I have a dual processor). They came across with expected runtimes of roughly 12 and 14 hours. They ran for 42 and 44 hours, respectively (it's really depressing when the "time to completion" is going up instead of down). In S5R2 "faster hierarchical search", i was getting somewhere around 53-54 credits per WU, and they were taking 7-8 hours. For the two forty-plus hour runs, I'm getting under 100 credits each -- total of about 160 for the two. So, S5R1, I would get (conservatively) 500-550 credits for 86 hours of CPU work -- and now I'm getting 160.
These early WUs have a wrong credit information sent to you. The credit will get corrected, as Bruce Allen wrote in his post in this thread.
Quote:
I'm back to S5R1 work now, but I've downloaded another S5R2 workunit for later. This one shows as 149 hours (that's more than six days). I'm dreading what will happen when it starts.
This one has the corrected credit information and therefore is mixing up your Boinc client. Don't worry about that, it will not take that long. ;-)
But it will take a couple of WUs until the client will show reasonably correct times again.
I am also worried about the run times of the newer wu's. Typically on the machine that has it I get 3-4 hour per wu, this one has run for 8½ and is saying it is only 36% finished.
Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.
Argh, I don't believe it, my X2@2,63 GHz is crunching a 182.7 Hz WU, and after 5h it has already finished 27.5%!
How much time will slower hosts need for these WUs?
Is it possible that we are using an app that is a kind of AMD unfriendly? ;-)
Some wrong or missing compiler switches?
This would explain, why the Linux app is so dramatically faster than the Win app, at least on AMD.
My longest WU(210.3 Hz) crunched so far took 56,850 sec for 260,21 credits with an AMD XP 2600+@2.2GHz(Linux). This is really ok! But the time my X2(Win) will need, is totally unsatisfying!
Tonight, I will run the X2 under Linux to compare the results. Maybe that tracks down the problem.
RE: Feel free to have a
)
Right now the only comparison I have is on this host.
S5R1 times 9200 - 9300 sec and 53 credit
S5R2 times 47000 - 49000 sec and 77 credit.
This is on a dual core AMD 4800+. I have 2 more in my cache that are calling for 29 hours so I'm interested in seeing how they come out stat wise. One of these is at 50% and time for completion should come down to 14 hours instead of 29. Guess my time calulator for units will need to relearn things.
edited to correct S5R2 label and time to completion update.
Arion
RE: Is there going to be
)
This is a current subject of debate over at SETI, so I have to ask you... Why are you still using 4.x?
RE: Is there going to be
)
I have, too, on some architectures nothing newer is available. No, no limitation planned other than what existed for S4 - it should be 4.19 or newer.
Our credit system will work even with older clients. The credit is fixed for each workunit on the server side, and measurements are taken so that the claimed credit should be what is granted at the end. For this to work one needs a newer client, with older clients the granted credit may be different from what is claimed, that's all.
And btw. yes, there is a large variation between runtimes of the different WUs. In principle they get longer with higher frequency. The credit rises proportionally, so if you want to compare run times of machines, look for Workunits that claim the same credit (and use newer clients). I'll try to get some chart for WU-time vs. frequency to post it here.
BM
BM
Hi Bernd, is it possible
)
Hi Bernd,
is it possible that the Linux application is much faster than the Win-App?
I got the following results about the credit-reduction, which should be 30%:
Athlon64 4400+ (Win XP): ~ -40%
Athlon XP 2600+ (Linux): ~ -15.4%
Athlon XP 3000+ (Linux): ~ -12,8%
All results are compared to at least 3 different groups of "old" WUs.
And both Linux hosts get more credits/h than the faster Win-host(per core)!
cu,
Michael
I should test that...
)
I should test that... dual-booting can be useful ;-)
OK, I received two early S5R2
)
OK, I received two early S5R2 units (I have a dual processor). They came across with expected runtimes of roughly 12 and 14 hours. They ran for 42 and 44 hours, respectively (it's really depressing when the "time to completion" is going up instead of down). In S5R2 "faster hierarchical search", i was getting somewhere around 53-54 credits per WU, and they were taking 7-8 hours. For the two forty-plus hour runs, I'm getting under 100 credits each -- total of about 160 for the two. So, S5R1, I would get (conservatively) 500-550 credits for 86 hours of CPU work -- and now I'm getting 160.
I'm back to S5R1 work now, but I've downloaded another S5R2 workunit for later. This one shows as 149 hours (that's more than six days). I'm dreading what will happen when it starts.
Bernd and company, you might want to rethink the size of these WUs. I have a pretty big fast machine; most people won't be able to finish the WUs within the deadlines. And the credit is REALLY off -- not just a little!
RE: is it possible that the
)
Would be quite surprising, the output from gcc on Linux is usually slightly slower than what we get from the M$ compiler, but is not impossible. Honestly during the past weeks we have been so busy fixing mere showstoppers that I didn't do a platform / compiler comparison with the latest code versions. Thanks for pointing me to that, I'll look into it.
BM
BM
RE: OK, I received two
)
These early WUs have a wrong credit information sent to you. The credit will get corrected, as Bruce Allen wrote in his post in this thread.
This one has the corrected credit information and therefore is mixing up your Boinc client. Don't worry about that, it will not take that long. ;-)
But it will take a couple of WUs until the client will show reasonably correct times again.
cu,
Michael
I am also worried about the
)
I am also worried about the run times of the newer wu's. Typically on the machine that has it I get 3-4 hour per wu, this one has run for 8½ and is saying it is only 36% finished.
Wave upon wave of demented avengers march cheerfully out of obscurity into the dream.
Argh, I don't believe it, my
)
Argh, I don't believe it, my X2@2,63 GHz is crunching a 182.7 Hz WU, and after 5h it has already finished 27.5%!
How much time will slower hosts need for these WUs?
Is it possible that we are using an app that is a kind of AMD unfriendly? ;-)
Some wrong or missing compiler switches?
This would explain, why the Linux app is so dramatically faster than the Win app, at least on AMD.
My longest WU(210.3 Hz) crunched so far took 56,850 sec for 260,21 credits with an AMD XP 2600+@2.2GHz(Linux). This is really ok! But the time my X2(Win) will need, is totally unsatisfying!
Tonight, I will run the X2 under Linux to compare the results. Maybe that tracks down the problem.
cu,
Micha