Wow 50% BOINC increase in performance

[Fozzie]

with just a software tweak and an optomised client.

Cheers Bok.

Woohoo. :elephant:

[Jeff]

Nice one! :cheers:

[Scoofy12]

what you really mean is a 50% increase in credit... since the rosetta code is the same, only the boinc code has changed. all that really means is that the scoring system is seriously broken.

[PY 222]

Ahh... you've found the optimized clients.

Welcome to the Club mate. :thumbs:

[n7vxj]

So someone tell me how to install the optomized amd client on my linux box?

[Bok]

Check your PM's

Bok

[Fozzie]

the optomised client is only part of the change the rest is clockgen and a little tweak in FSB.

Turned my benchmark from around 2100 to 3000 with more OCs to come.

[Bok]

aha, so you've done it then, what are you up to then ?? I got your board/proc to 2.4 without trying to go any higher :)

Bok

[Scoofy12]

well, i still stand by my assertion that the scoring system is broken... my benchmarks went from
2005-10-20 11:31:43 [---] Benchmark results:
2005-10-20 11:31:43 [---] Number of CPUs: 2
2005-10-20 11:31:43 [---] 686 double precision MIPS (Whetstone) per CPU
2005-10-20 11:31:43 [---] 1298 integer MIPS (Dhrystone) per CPU
2005-10-20 11:31:43 [---] Finished CPU benchmarks

to

2005-10-20 11:32:55 [---] Benchmark results:
2005-10-20 11:32:55 [---] Number of CPUs: 1
2005-10-20 11:32:55 [---] 2881 double precision MIPS (Whetstone) per CPU
2005-10-20 11:32:55 [---] 4481 integer MIPS (Dhrystone) per CPU

this is a benchmark improvement of over 100% (not 200%, the newer version counted the HT cpu as one instead of 2) for FP, and 72% for integer.

i don't know if credit scales linearly with benchmark scores, but
1) rosetta is compiled separately without these optimizations, and so does not run any faster, only the credit goes up, and
2)even if that wasn't the case, whetstone/dhrystone is just about the absolute worst possible way to benchmark anything; its results are completely meaningless because the benchmark code is like 10 lines long, with tight, easily optimized loops, and its performance is a very poor predictor of actual application performance.

[Mustard]

Originally posted by Scoofy12
what you really mean is a 50% increase in credit... since the rosetta code is the same, only the boinc code has changed. all that really means is that the scoring system is seriously broken.



Sigh............ boinc.......... If we remember back a bit, this system was designed by an individual that said that the crunchers (speaking of people) can't be trusted. Surely, something from Berkeley couldn't be poorly designed?????

:rotfl:

In seriousness though,I really wouldn't say it is broken. Rather poorly designed. However as it sits, everyone out there in cruncherland can use optimized clients, and overclock, etc. But it does beg the question of why they just didn't put out optimized clients from the git-go, and leave all the players on an equal footing...... if we are hinting at or going to talk ethics here.

And of course each boinc project has scoring differences despite the effort/push to try and have cpu_effort relate to equal points across the various boinc projects.

Myself, I fail to grasp why there needs to be any equality in points across the various projects. I also (like you) fail to understand why the ws/ds was used to base a "benchmark" when there are better ways to do things. But then, who ever said that the boinc way was the best way anyway???? And again, I admit that I am not a boinc fan at all.