Seti@Home BOINC: standard application vs optimized application

[petrusbroder]

I have run a comparison for different processors using an optimized and the standard seti@home-BOINC-applications.
For each comparison I have crunched at least 25 (Celeron) and most 40 (all other processors) WUs,
calculated the mean value and the standard deviation (to compensate for the variation between the WUs).
These are the numbers:


Processor _____ Standard application ____ Optimized application ____ % faster
____________________ seconds/WU __________ seconds/WU
__________________ mean ± std.dev ________ mean ± std.dev _________________________________

A64 3200+ ________ 9 083.69 ± 541.37 ______ 4 912.14 ± 327.27 ________ 84.9 ____ 512 MByte RAM
A64 3000+ ________ 9 876.61 ± 275.95 ______ 4 932.76 ± 495.67 _______ 100.2 ____ 512 MByte RAM
A-XP 2800+ ______ 10 289.61 ± 636.68 ______ 7 944.29 ± 529.50 ________ 29.5 ____ 512 MByte RAM
A-XP 2400+ ______ 11 084.58 ± 549.00 ______ 8 260.94 ± 802.74 ________ 34.2 ____ 768 MByte RAM
A-XP 1900+ ______ 17 607.70 ± 1205.64 ____ 12 899.70 ± 1185.64 _______ 36.5 ____ 1 512 MByte RAM
A-TB 1000 MHz____ 23 935.96 ± 1171.06 ____ 20 648.31 ± 104.60 ________ 15.9 ____ 384 MByte RAM

P4 2.8HT _________ 12 519.47 ± 1 108.76 ____ 6 104.37 ± 675.16 _______ 105.1 ____ 512 MByte RAM, (does 2 Wus)
P III 1270MHz _____ 15 207.38 ± 1 018.74 ____ 11 246.41 ± 646.73 _______ 35.2 ____ 384 MByte RAM (OC'ed from 900 MHz)
Celeron 1.5GHz ___ 24 689.86 ± 400.66 _____ 14 158.07 ± 863.89 _______ 74.4 ____ 384 MByte RAM


MacG5 2 MHz _____ 10 452.08 ± 922.92 ___________ not tested __________________ (does 2 Wus)


I obtained the optimized applications from this page: www.marisan.nl/seti/. I have used the following clients:

Athlon A64: YAOSCW-W-r8.1.zip
Athlon XP : YAOSCW-K-r8.1.zip
Athlon TB : YAOSCW-0-r8.1.zip
Pentium 4 : YAOSCW-N-r8.1.zip
Pentium III: YAOSCW-K-r8.1.zip
Celeron : YAOSCW-K-r8.1.zip
(TB = Thunderbird)

I checked with the utility CPU-z which different flavours of MMX, SSE (1-3) were supported by the processors and chose the application accordingly.
I installed the applications according to the instructrions. Before installing it I set the project to "no more workunits" and waited until all WUs were crunched. When the cache was empty, I reset the project and installed the optimized client. Then I ran a test WU, which always completed its run flawlessly. Then I restarted the project.

 

[RaySun2Be]

Wow, lot of work went into this. Thanks for the info. Definitely some significant gains to be had using optimized clients.

 

[SirUlli]

if your are running on Linux, you have to try this

http://naparst.name/

it is the fastest client in Town

Athlon3.200+ at 45 Minutes, when i am right...

and some times from me, with the optimised Client running on windows

Atlon 3.200+ at 1:10
P4 3.2 with HT 1:30 for 2 WUs

Sir Ulli

 

[networkman]

Of my own fleet of (13) rigs running the Boinc E@H project, the AMD XP processors have a decided advantage over their Intel P4 couterparts, all other factors being equal. How much so? Well, my XP2000+ rigs regularly out-perform the P4 2.66 and P4 2.80 rigs by 15 to 20 percent on average.

And yes, both of the P4s above are the real thing, ie. not Celeron processors.

 

[Dalephi]

Nice chart. Thanks for taking the time to make this.

 

[petrusbroder]

Now, the interesting part (in switching to optimized application) is the following:

Before using the optimized application I crunched an daily (for 5 days) average of 1 355 credits (total: 6 776 credits);
after switching to the optimized applications I crunched an daily average of 1 990 credits (total: 9 947 credits).
That is a 46.8% increase in credits.
Nice!

 

[mrwizer]

I noticed your production jumped quite a bit.

 

[Pokey]

Thanks for this comparison petrusbroder. :thumbsup:

I have been running the optimized client on one of my crunchers and it is clearly faster, but I wasn't sure about the credit because I read some posts on the internet that indicated maybe there was a penality in that regard. I'm glad to see the additional credit is in fact there. That is the whole reason for running it.

 

[SirUlli]

i juist get a new Windows Version of

http://naparst.name/


test it on my Athlon 3.200+

Rumors said that one wu is READY in 45 Minutes

i will tell you the Differences

Sir UIlli

 

[petrusbroder]

Hmmm, I looked for the Windows version, but the page clearly states:

Please note:
1. These programs do not run on Windows; they run on Linux. I may produce a Windows version in the future.
2. These programs rely on Intel's IPP library, which has memory access errors.
3. Happy crunching.

Where did you find the Windows version, SirUlli? :Q

 

[SirUlli]

@petrusbroder

this Version is inhouse testing only

when all goes the right way i will tell the source

first Impression

Athlon64 3.200+ not overclocked Windows SP1 45 Minutes...

Sir Ulli

 

[petrusbroder]

Really nice!

 

[SirUlli]

more Info

my P4 3.2 with HT does 2 WUs in 59 Minutes

give me a few Days to test this, and if these WUs are validatet.

Sir Ulli

 

[Orange Kid]

Originally posted by: petrusbroder
Hmmm, I looked for the Windows version, but the page clearly states:

Please note:
1. These programs do not run on Windows; they run on Linux. I may produce a Windows version in the future.
2. These programs rely on Intel's IPP library, which has memory access errors.
3. Happy crunching.

Where did you find the Windows version, SirUlli? :Q

might try these?
http://pperry.f2s.com/downloads.htm
sry-same ones ur using

 

[petrusbroder]

Thanks for the links. Found also a nice Mac-app and Mac client which I may try later.

 

[SirUlli]

for Info

removed too many Errors, there is an new and better Version, look at my last post

this is the speziall Windows Vsersion

but not out of Errors

it is in Alpha State, but when you will test this

for Info

to use at own Risk

Sir Ulli

 

[petrusbroder]

Thank you SirUlli!

 

[SirUlli]

there is a new and better Version

http://www.guntec.de/Crunch3r/boincx86.html

also use at your own Risk

Sir Ulli

 

[SirUlli]

i knew the most people here like Benchmarks

Athlon64 3.200+ not overclocked

23.11.2005 22:25:40||Benchmark results:
23.11.2005 22:25:40|| Number of CPUs: 1
23.11.2005 22:25:40|| 2727 double precision MIPS (Whetstone) per CPU
23.11.2005 22:25:40|| 8793 integer MIPS (Dhrystone) per CPU
23.11.2005 22:25:40||Finished CPU benchmarks

for Info

sir Ulli

 

[petrusbroder]

Looks terrific. Great thanks SirUlli!

 

[petrusbroder]

OK, TeAm-mates:

I have been running the optimized application now for 7 days (@24/7). The average production is 2120 credits/day compared to 1365 for the 7 days before that.
The interesting part is the following:

For each and very WU I get less credits than before (That was expected, if I crunch for a shorter time (because the application is more efficient) I get less credit).
I crunch more WUs. The decreased credits/WU and the increased number of WUs should cancel (OK, more or less) each other.
There is one more factor though:

The credits are based not only on my claims, but on the credits of those others who crunched the same WU too! Or is it on a reference computer? I do not remember.
Anyhow: my "claimed credit"-values have decreased - as expected. But my "granted credit" has been essentially the same as with the standard application and is in most of my computers higher then the "claimed credit".

Therefore the credits obtained for the WUs are more than I expected, both because I crunch more WUs and because I do not get only the credits I claim, but more: I get the same number of credits as the four crunchers who did the same WU as I did.
That is the reward for having an optimized application.

Of course, if all crunchers in Seti@home - BOINC would use new computers and the optimized application, then the number of "granted credit" would drop because all crunchers would get less ... :Q

 

[salvorhardin]

Credit granted is based on the claimed credit of the other computers. The highest and lowest is taken off and the middle score is the amount that is granted to everyone.

 

[dringdahl]

These really work on all the BOINC projects...

 

[SirUlli]

we are just in the testing of the new cache version,

this Version will be able to run a WU in 45 Minutes on my Athlon64 3.200+

on Windows...

but this is still Beta..........

so i will inform you what is going, be patient...

Sir Ulli

 

[petrusbroder]

Thnks SirUlli, we are looking forward to the better application!