The link is live again. (SETI K7-08)
But I don't do any funny remarks and I don't promise anything.
I tried to run it on one of my PC again, but it did lot of computing errors, so I will do a revisit on this code.
...and I have a SSE3 processor now ( no mobo yet ). :-)
Great. I have got a pack from England. It will be the mobo. :-)
So, I will start to do X41 tomorrow.
Sounds great! The new apps are already so fast it's amazing that there is still room for improvement. Can't wait to see what you cook up this time.
There are 10^11 stars in the galaxy. That used to be a huge number. But it's only a hundred billion. It's less than the national deficit! We used to call them astronomical numbers. Now we should call them economical numbers. - Richard Feynman
Cool. How much of a speedup should we be expecting? In theory I could see a doubling of speed, but competition for other resources will probably make it somewhat less.
Anyone have any speed comparision between S39L and S41.06?
My two Pentium III hosts and my Banias Pentium M are all substantially faster (20%+) on S41.06 than they were on S-39L. My Gallatin P4 running hyperthreaded is currently running S40.04--as S41.06 was slower than that by a little. I don't have all the intermediate steps to say, but suspect S-39L would be about breakeven with S41.06 for that Gallatin host. Different P4's are showing different responses.
You've hidden your computers, so I can't give specific advice, but for simplicity, so long as all your machines support SSE, probably an overall switch from S-39L to S41.06 would be a good idea.
You've hidden your computers, so I can't give specific advice, but for simplicity, so long as all your machines support SSE, probably an overall switch from S-39L to S41.06 would be a good idea.
That really depends. On my dual Cranford box (3.4 GHz Xeon, 200MHz FSB, 16kB L1 data cache) U41.02 (which is supposedly marginally faster than S41.x) performs roughly the same as S39L: long results (r1_1436.5) take somewhere between 5920 and 6100 seconds and times are not much different for either turboAlbert.
U41.02 really is on afterburners on Opterons (approx 2/3 time compared to D40).
That really depends. On my dual Cranford box (3.4 GHz Xeon, 200MHz FSB, 16kB L1 data cache) U41.02 (which is supposedly marginally faster than S41.x) performs roughly the same as S39L: long results (r1_1436.5) take somewhere between 5920 and 6100 seconds and times are not much different for either turboAlbert.
I would like to notice that Sxx codes have an SSE/FPU overlapping technique that is good for AMD CPUs, but the intel processors dislike it. C/D/U/S41.xx are based on the previous versions, just they executes less and simpler instructions. So the pure code would be faster than previous versions, but I have only one intel-based laptop for checking optimizations and it doesn't show me significant disadvantage with overlapped code because it has a modern Dothan core.
RE: The link is live again.
)
I tried to run it on one of my PC again, but it did lot of computing errors, so I will do a revisit on this code.
...and I have a SSE3 processor now ( no mobo yet ). :-)
RE: ...and I have a SSE3
)
Great. I have got a pack from England. It will be the mobo. :-)
So, I will start to do X41 tomorrow.
RE: RE: Great. I have got
)
Sounds great! The new apps are already so fast it's amazing that there is still room for improvement. Can't wait to see what you cook up this time.
There are 10^11 stars in the galaxy. That used to be a huge number. But it's only a hundred billion. It's less than the national deficit! We used to call them astronomical numbers. Now we should call them economical numbers. - Richard Feynman
RE: RE: ...and I have a
)
Intel SS3 WinXP. ;)
me-[at]-rescam.org
Cool. How much of a speedup
)
Cool. How much of a speedup should we be expecting? In theory I could see a doubling of speed, but competition for other resources will probably make it somewhat less.
It sounds like S41.06 is
)
It sounds like S41.06 is stable.
Anyone have any speed comparision between S39L and S41.06?
I've been very happy with S39L. Looking at my results list I've got 315 completed with no client errors and no validation errors.
For an amd host it's
)
For an amd host it's massively faster. IIRC s40.04 is the fastest on a p4.
RE: Anyone have any speed
)
My two Pentium III hosts and my Banias Pentium M are all substantially faster (20%+) on S41.06 than they were on S-39L. My Gallatin P4 running hyperthreaded is currently running S40.04--as S41.06 was slower than that by a little. I don't have all the intermediate steps to say, but suspect S-39L would be about breakeven with S41.06 for that Gallatin host. Different P4's are showing different responses.
You've hidden your computers, so I can't give specific advice, but for simplicity, so long as all your machines support SSE, probably an overall switch from S-39L to S41.06 would be a good idea.
RE: You've hidden your
)
That really depends. On my dual Cranford box (3.4 GHz Xeon, 200MHz FSB, 16kB L1 data cache) U41.02 (which is supposedly marginally faster than S41.x) performs roughly the same as S39L: long results (r1_1436.5) take somewhere between 5920 and 6100 seconds and times are not much different for either turboAlbert.
U41.02 really is on afterburners on Opterons (approx 2/3 time compared to D40).
Metod ...
RE: That really depends. On
)
I would like to notice that Sxx codes have an SSE/FPU overlapping technique that is good for AMD CPUs, but the intel processors dislike it. C/D/U/S41.xx are based on the previous versions, just they executes less and simpler instructions. So the pure code would be faster than previous versions, but I have only one intel-based laptop for checking optimizations and it doesn't show me significant disadvantage with overlapped code because it has a modern Dothan core.