thanks Ulli! looking forward to giving it a try 
here are my times with the speziall Version, which will be able next week for everyone
http://setiathome.berkeley.edu/results.php?hostid=318394&offset=80
Athlon64 3.200+ not overclocled running with Windows XP1
that are about 45 Minutes for a WU
and my P4 3.2 with HT on an i875 Mobo. also not overclocked running Windows XP1
http://setiathome.berkeley.edu/results.php?hostid=56765&offset=80
that are about 50 Minutes for 2 WUs
for Info
Sir Ulli
i hope this spezial release will up this Days, my Athlon64 3.200 is flying
for Info
S754 Board Athlon64 3.200+ Twinmos BH6 chips not overclocked
http://home.teleos-web.de/ubrinkschmidt/test/Athlon64crunch.jpg :eek:
sir Ulli
for Info
not overclocked…
the only Prob at this time is that the people who biuld this are waiting for an offical Version from Imhell
http://www.intel.com/cd/software/pr.../cwin/index.htm
and
http://www.intel.com/cd/software/pr...b/ipp/index.htm
the Version hat i using is build with the trial 30 Day Version…
so if the Lizense Prob is cleared, i hope this week, this Version will be offical…
for Info
Sir Ulli
thanks for keeping us updated Sir Ulli 
update:)
http://www.guntec.de/Crunch3r/setix86.html
use at own risk…, and dont forgett to make a Backup before install …
for Info with these Version my Athlon64 3.200+ Newcastle 2.2 GHz 512 KB is running a WU in about 45 Minutes:eek: , not overclocked.
Sir Ulli
That’s astounding perfrmance and unclocked as well :eek: Awesome!!! Cheers Ulli
[edit]
sorry Ulli didn’t see you posted the link above!
[/edit]
One a sample of my machines: AMD64 3200 - 40 mins, P4 3Ghz 35 Mins with HT on.
Only trouble now is download enough WU to prevent swapping out to Einstein or Rosetta!
Installed the new client at work and it seemed fine. Just put it on at home on my amd64 3000+ and the completion time jumped from 1:20 to over 12 hours 
gonna give it another try later
edit done a fresh install of BOINC and then installed the SETI client and seems much better… got project suspended for the time being but benchmarks looks better this time
P4 3.4 is running units in about 38-42 minutes (This is while I am also doing CAD work and Folding is running at a low priortity also !)
you think you have a fast computer, so look here
that is fast i think…
sir Ulli
wasnt intendend as bragging about my computer (its work box anyway :(), was just referring that the new client had cut down my wu time by about ten minutes even with me working.
Can someone give me a break-down as to what this all means?
Are these optimised clients just for the boinc manager (if so what sthe point in that)? or for the actual seti execution core? If so, do they make legit results? (remembers the seti classic ‘cheat’ clients that had ppl turning out 10k WU a day)
WHAT DOES IT ALL MEAN?!
Ummm where to start?
OK, there are two pieces of software:
- The Boinc core client, and
- The science application
Taking the science application first, this has been optomised by several people to perform the floating point calculations faster. The most recent optomisations use a clever trick of not having to re-compute calculations that have already been made. This is performed by use of a cache (not to be mixed up with the cache of workunits!). Each time the science app wants to perform a calculation it checks to see if it has been made before and re-uses the result if it has. The result is quite simply astounding. For the stock Seti science app (version 4.18), a WU on one of my 3GHz P4 used to take about 3-4 hours (for two with HT). They now run in under an hour on some machines. On my AMD rigs, I achieve run times down to 35 minutes.
Regarding the accuracy, yes they have been extensively tested (Sir Ulli 
has worked hard in this area and can tell you more). Also, the Seti validation process requires a close match between all results for a particular work unit before granting credit.
Now for the second piece of software:
The Boinc core client performs benchmarks (amongst many other things). These benchmarks can be optomised using specific compiler options. What this means is that with an optomised Boinc core client your benchmarks are increased and your claimed credit increases in line. This offsets the much faster crunch time achieved with the optomised science app. Claimed credit rarely matches with granted credit because of the validation process. I personally find my claimed credit is much lower than other machines crunching the same WU.
Overall the result is an amazing increase in work units processed over a given time and a corresponding increase in credit granted after validation.
Installation
Before installing, you need to know some detail about your processor that is not immediately obvious. Download and install CPU-z (about 2/3rds down the page under version 1.31). This will tell you if you have certain cpu extensions (SSE2/SSE3 etc). Don’t worry about the remainder of the detail for now. Just match the CPU config to the application on the websites below.
Be carefull when you install the Science app and follow the instructions on the various websites. There is a small file that needs to be installed alongside the optomised app. This app_info file is critical because it tells the Boinc core client to use the new optomised science app rather than the old stock Berkeley one.
The app_info and setiathome_4.11_windows_intelx86 go into your seti at home project folder (typically found on your hard drive at C:\Program files\Boinc\Projects\Setiathome.berkely.edu), as does the .pdb file if you download that version. The Boinc.exe core client goes into you Boinc directory (typicaly in C:\Program files\Boinc)
Oh, and [u]always[/u] make a backup of you entire Boinc directory before installing.
Currently the fastest aps and clients are found at Crunch3r
Slightly older (slower) versions (but with arguably more help) can be found at Testsuji
Be aware that many people confuse the optomised client with the optomised app. The optomised app does the crunching, while the opomised client re-aligns the benchmarks, but will have little effect on crunch time.
(BTW, big assumption…you are running Windows!)
Thanks to Crunch3r, Trux, Harold, Ned, Tetsuji and others for their fantastic work.:hail:
Wow. What an amazingly clear and concise explaination! top marks I understand now. I will be wanting and optimized core execution module and an optimised boinc to make sure my credit doesent go to hell. I wonder though, why berkerly hasnt done these optimizations themselvs, seems they’re underutilising thier user base quite a bit.
IMO, the above post should be stickified at the top of the boinc forum 
Mods what you think?
Thanks for the comments.
I’ve been running optomised apps since about May/June time, so I have a little knowledge.
But I can’t guarantee the guidelines above are all absolutely correct. If anyone spots something wrong above, feel free to put it right 
In reality the praise should go to the software devs such as Crunch3r, Trux, Harold, Ned, Tetsuji and others who have supported.
I have taken the steps above, and while it didnt seem to have a big effect on the floating point (my pIII-S went from 1200 ish MIPS to 1470 odd), the integer performance is stratospheric. i have a newcastle A64 3000 here that went from 3200 ish to 8300 MIPS. It seems to be crunching pretty fast aswell, its ripping thru this unit here. Only thing is that the screensaver has been removed. Yes i know it takes cycles, but i liked it. Screen blanked after 15 mins anyway.
Actually the only thing i would be concerned about is that the berkerly execution core is version 4 18, while the optimized is 4 11. i would assume that berkerly would only realease new versions to do better science or fix bugs or somthing. I hope that the science is not compromised in this. And if this is the real deal, shouldent berkerly use these cores as official ones?
I dont mean to sound too skeptical, but that is a LOT of integer MIPS :flip: I will continue testing this.
It’s OK, v4.11 is just fine. Validation makes sure the science is not compromised. The result must be statistically “highly similar” to other crunchers on the same WU. If it doesn’t validate, you don’t get credit and the results is discarded to protect the science.
There are standard workunits that are available. All the publicly released optomised apps have been tested against these standard WU’s and compare well to 4.18.
If I recall, the difference between the stock 4.10 and 4.18 came about because of some graphics tweaks in 4.18 to try to make some video cards work. The science is the same. 4.11 Optomised apps are based on 4.10.
I really don’t know why Berkeley don’t adopt. Maybe there is some in-house software quality mechanism that prevents them from deploying third party software. They do link in several places on their site and positively encourage porting and optomisation.
Just now, I think most of the official Berkeley science app development is dedicated to the new science app due for release “shortly”. I haven’t been involved in that one yet, so can’t comment much about it. I know it uses different processing algorithms that increase the sensitivity. On the otherhand, I hear it increases crunch time by between 4 and 10 times on v4.18. I also hear the optomisers have already been hard at work too! 
Cool
Rather than get to a situation where it takes 10 mins for a result, Its sensible to look deeper into the data with the processing power available.
Depressing tho is that there seems to be no optimised stuff for my athlon thunderbird. Prehaps i could use the pentium 1/2 for MMX, but there is no MMX boinc client 
Oh well.
I use the optimised BOINC clients located near the bottom on both my AMDs…seems to work fine.