S5R2 vs S5R3
20 Sep 2007 9:28:37 UTC
Topic 193154
(moderation:
I'd like to get some information on what S5R3 is going to bring to the table as far as times/credits and is it going to be stable?
Back in mid-June I pulled all my systems off of einstein and switched them to climateprediction.net as I wasn't willing to wait 7 1/2 weeks for credits on work completed earlier. I had a number of these fall through the cracks because of the long crunch times and whatever wingman I was getting decided to pull the plug. Got a bit disheartning after a while.
I came back about a month ago and run a couple of my systems for about a week and return times seemed to be a bit more responsive. Will the new app have all, if not most of the beta error tracking taken out so that we won't have such long extended processing times?
Mostly what I'm trying to say is that I've missed my einstein@home running and hoping things have improved enough to come back.
Any feedback greatly appreciated.
Arion
PS... I have 2 systems that are nearing their completion times for climateprediction that I'd like to bring back. (maybe a week or so left)
edited for additional PS comment
S5R2 vs S5R3
)
WU sizes are going to be smaller than s5r2. All the major bugs have been wiped out in the current app, IIRC the reported error rate was down to only a hundred/day project wide several apps and bug fixes ago.
RE: Mostly what I'm trying
)
Perfect timing, S5R3 will start in just a few days.
Workunit runtime will on average be close to that of the shortest WUs of S5R2, or very roughly one third IIRC compared to that of the ca 650 credits units most common these days.
As always, the credits for WU will depend only on the workunit and will be calibrated at the start of the run to be "fair" compared to other BOINC projects.
The run will start with code similar (but not equal) to the one used now but will be switched to "Akosified" (= optimized) code later in the run, according to Bernd's most recent announcements.
The app is quite stable now, the very annoying cross-platform validation problem was fixed some months ago.
So you are most welcome to give S5R3 a try in a few days.
Happy crunching
H-BE
RE: Perfect timing, S5R3
)
Even though I haven't been crunching along for most of the summer I have been trying to keep abreast of what's going on here. My biggest problem is that I have all 64 bit AMD chips here an they were taking a beating. If that's fixed it's time to watch final climateprediction estimated completion dates more closely.
Thanks
I'm not sure if it's 100%
)
I'm not sure if it's 100% fixed (under Windows) but it is certainly much better... Bikeman, you know sth more definite? I've mainly been crunching with my (Intel) Laptop the last weeks.
RE: I'm not sure if it's
)
I think, he knows. But won´t tell us ;)
Lets hope, the new work arrives before the weekend. We could get low on work...
RE: My biggest problem is
)
Fwiw, my 64 bit AMD 3500+ seems to be doing quite well! :)
Finally, 64 bit versions of
)
Finally, 64 bit versions of BOINC will automatically download clients for E@H (32 bit, not native 64 bit, but the FPU is the same anyway and 64 bit code would not be that much faster). So no manual workaround is needed anymore!
There is no longer any AMD specific problem that I know of (other than Intel producing the faster CPUs at the moment, I guess ;-) ).
It will be interesting to see how different CPUs will perform once SSE, vectorized versions of the client will be introduced later in S5R3.
The Pentium IIIs implementation of SSE was not that hot, the Pentium 4 (Netburst) ... well...let's not talk about that anymore. Pentium M and first generation Core Duos still had only 64 bit data paths in the SSE unit, so a 128 bit wide SSE operation would be split internally into two micro operations, which is kind of not so cool for a vector operation.
I'm not sure about the AMD K7 chips. First generation K8 had not so hot SSE support, I think.
But finally, the Core 2 CPUs are said to have really excellent implementations of SSE, like true 128 bit data paths within the SSE units, and less constraints for parallel processing. Will the AMDs be able to compete ?? As I said, this will be exciting.
BRM
Hey ;-) don't make my Core
)
Hey ;-) don't make my Core Duo sound so bad. I do hope it will show a decent performance in the next run.
RE: Hey ;-) don't make my
)
:-) I have a Mac Mini with a Core Duo myself (too bad, ony a few months later and I would have got a Core 2 Duo Mac mini for less...sigh....), and even the Core Duos are really doing well against the P4s :-). So I feel with you.
CU
H-BE
RE: Finally, 64 bit
)
Has this been implemented yet, or will it come in when the S5R3 run begins?
TTFN - Pete.