wish i could get an einstien or 2 to compare. if all alberts were the same you are actually running almost twice as fast as i- little over an hour per? nice! if the others skew you up that seems like a good thing- usually i am the high guesser in this game and get drug down, but i m new so not much to complare to. they have the linux of this newer boinc optimised (raised someone's i think from 50's to 100's but cant remember) and they were waiting on visual studio 2006 for the doze version. i used your reported not credited results to avoid others skew.
i m gonna hafta dust off the socket a and boinc bench- to me it seems the amd64's arent holding a candle to the older ones... but then we were used to running at 2 ghz cheep now its expensive to get over 2.2 ghz... gotta love it; my newer technology is physically running slower than my old one. and by a lot.
Quote:
jl,
Even comparing credit/sec doesn't work in this case. I'm also doing Seti, and running an optimized Seti app for faster production there, and using an optimized (higher benchmarking) Boinc client to keep those credit claims at near-normal levels. Fine, as far as Seti goes, more production should be rewarded with more credit. The stickler comes when Einstein is in the mix. Because there is no Einstein optimized app, the high-marking Boinc client skews my credit clains upward, and thus the claim/sec is out with the dishwater. Other than that, your measure is a good one, and would apply well in ordinary circumstances.
My rig is described here, at Warhawk's request. It's 5hr Einstein times made it one of the fastest single-proc rigs on Windows.
The Altivec version of the E@H app uses hand optimized asembly in some places. The dev team has tried doing the same with x86, but was unable to get a meaningful gain in performance over normal compiled code. This could mean one of two things, either the p4/a64 architectures have much better compilers, or the people who attempted the handwritten asm are much better thinking in the PPC paradigm. If the former is the case, the performance gap is probably due to architetural differences. IBM designed the PPC with a vector unit significantly more capable than any other major processor family on the market. IF the E@H algorythm is capable of taking full advantage of it a large performance lead isn't the least unexpected.
wish i could get an einstien
)
wish i could get an einstien or 2 to compare. if all alberts were the same you are actually running almost twice as fast as i- little over an hour per? nice! if the others skew you up that seems like a good thing- usually i am the high guesser in this game and get drug down, but i m new so not much to complare to. they have the linux of this newer boinc optimised (raised someone's i think from 50's to 100's but cant remember) and they were waiting on visual studio 2006 for the doze version. i used your reported not credited results to avoid others skew.
i m gonna hafta dust off the socket a and boinc bench- to me it seems the amd64's arent holding a candle to the older ones... but then we were used to running at 2 ghz cheep now its expensive to get over 2.2 ghz... gotta love it; my newer technology is physically running slower than my old one. and by a lot.
Not to be difficult, but
)
Not to be difficult, but there is an optimized Einstein@Home application. It runs on the G4/G5 using Altivec ...
RE: Not to be difficult,
)
optimised.....?
I thought the default E@H app has already been optimised.....
The Altivec version of the
)
The Altivec version of the E@H app uses hand optimized asembly in some places. The dev team has tried doing the same with x86, but was unable to get a meaningful gain in performance over normal compiled code. This could mean one of two things, either the p4/a64 architectures have much better compilers, or the people who attempted the handwritten asm are much better thinking in the PPC paradigm. If the former is the case, the performance gap is probably due to architetural differences. IBM designed the PPC with a vector unit significantly more capable than any other major processor family on the market. IF the E@H algorythm is capable of taking full advantage of it a large performance lead isn't the least unexpected.
finally got an einstein; let
)
finally got an einstein; let c what it takes....