Thanks everyone :o)

Pepperammi, 4 GHz on my P-D 805 seems to run okay...haven't ventured further yet, I'm doing a benchmark and Prime95 run to see whether it's stable.
.......
Oh yeah, I updated the test package so kwsn-test-xW.exe does not ask for a dll file anymore.

Regards,
Simon.

Thats one hell of an overclock.
I'll grab hold of the new pakage. :)

So far so good. My P4 3.2@3.6GHz HT with 512MB PC3500 DDR running XP Pro.

11950 seconds for a claim of 50.00 = 15 credits/cpu hr (AR=0.56343). So if it gets what it claims then this will result in a RAC of 720...now to sit and wait for the granted to come in.

Compare this to an unoptimised 25050 seconds for a claim and granted of 64.55 = 9.28 credits/cpu hr (AR=0.42298).

That's a 60% improvement in credits/cpu hr. But this was a sample of 2 versus 1. I'll crunch a few more to get some more statistically meaningful results.

Live long and crunch.

Oh very sweet. One of them has been granted 50.00. FANTASTIC! And it outperforms the stock application on a 3.8GHz machine by 5820 seconds!

All hale Chicken of Angnor! Ni!

Simon, How keen the automatic subscription facility! I just got through trying to download your Windows, SSE3 numbercruncher into my ...berkeley.edu folder. I had to do everything manually (and guess what to do) because I didn't see anything work automatically. All your files must be in that folder but with all the other past Boinc and Seti files and crunchers, also, possibly even Crunch3r's from months ago. Could you please tell me exactly what files should remain in that folder and which come out, if any. Or could you tell me the exact procedure to have your program overwrite what's in my ...berkeley.edu folder? I still see the beautiful screensaver graph tray fill with waves, etc, and, with your program, I take it that I'm not supposed to see that thing. Thanks a lot. Clyde

---

here you can see all my results with the new optimized Win client
----------------------------------------------------------------
It works really fine

Simon, How keen the automatic subscription facility! I just got through trying to download your Windows, SSE3 numbercruncher into my ...berkeley.edu folder. I had to do everything manually (and guess what to do) because I didn't see anything work automatically. All your files must be in that folder but with all the other past Boinc and Seti files and crunchers, also, possibly even Crunch3r's from months ago. Could you please tell me exactly what files should remain in that folder and which come out, if any. Or could you tell me the exact procedure to have your program overwrite what's in my ...berkeley.edu folder? I still see the beautiful screensaver graph tray fill with waves, etc, and, with your program, I take it that I'm not supposed to see that thing. Thanks a lot. Clyde

Clyde,

currently none of my clients include graphics. That will change in future versions.

These files definitely need to stay in your projects/setiathome.berkeley.edu folder:

app_info.xml - the one that came with my client
all files starting with "setiathome-5.15-kwsn"
all files starting with a number - these are your WUs. Take extra care not to delete any!

That was the bunch.

HTH,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

Well mi first wu is OK i got 60 some credit for it...so its OK

Thank you for the optimization Simon ,im goig to test te MMX on mi P II soon.

---

Scorpions - Wind Of Change

Hi Simon, here a few detailed results from my host.

As you can see from the comparison, I think it should be possible to squeeze some more optimisation from the apps. I hope so !!

Intel P4 2.8 GHz
Northwood
L2Cache 512KB
533MHz FSB
512MB RAM
WinXP Pro SP2


Testbench Time Results

Test WU 1
xB= P-M SSE2 app - 13mins 54secs
xN= P4 SSE2 app - 13mins 31secs
xK= SSE app - 14mins 5secs
Default Client= 17mins 33secs

Test WU 4
xB= P-M SSE2 app - 3mins 0secs
xN= P4 SSE2 app - 2mins 58secs
xK= SSE app - 4mins 17secs
Default Client= 4mins 52secs

Test WU 5
xB= P-M SSE2 app - 16mins 45secs
xN= P4 SSE2 app - 16mins 11secs
xK= SSE app - 16mins 57secs
Default Client= 21mins 14secs

Test WU 7
xB= P-M SSE2 app - 7mins 54secs
xN= P4 SSE2 app - 7mins 23secs
xK= SSE app - 7mins 51secs
Default Client= 12mins 10secs


KWSN xN - P4 SSE2 app

AR= 0.426462 Time= 19810.91 Credit= 62.40

AR= 0.429483 Time= 19753.84 Credit= 61.77

AR= 0.430764 Time= 19759.17 Credit= 61.60

AR= 0.612825 Time= 15532.47 Credit= 48.63


Crunch3r SSE2 opt app

AR= 0.34092 Time= 16326.36 Correct Credit= 58.70

AR= 0.32799 Time= 16351.88 Correct Credit= 58.69

AR= 0.393934 Time= 15926.86 Correct Credit= 67.23

AR= 0.393941 Time= 16372.16 Correct Credit= 67.23

AR= 0.426010 Time= 15385.61 Correct Credit= 62.45

AR= 0.426465 Time= 15348.84 Correct Credit= 62.40

AR= 0.438664 Time= 14819.67 Correct Credit= 61.27

AR= 0.440447 Time= 15038.00 Correct Credit= 61.03

AR= 0.528260 Time= 12606.70 Correct Credit= 50.33

AR= 0.600282 Time= 12768.14 Correct Credit= 49.19

AR= 0.677314 Time= 10810.80 Correct Credit= 34.45

AR= 0.767817 Time= 10139.16 Correct Credit= 32.02

AR= 0.954573 Time= 9008.41 Correct Credit= 28.30

---

Simon, thanks for all your effort!

Here's my first result using your PentiumM-SSE2.

Pentium M 740 1.73 GHz
768MB x 266MHz DDR2
WinXP Pro SP2

Some more stats from this machine:

Based on an average of 4 KWSN WUs,
9,761.07 seconds / 51.09 granted credits = 193.1931 seconds per credit

Based on an average of 5 Crunch3r WU,
11,009.38 seconds / 53.76 granted credits = 211.7573 seconds per credit

That's an 8.76% speed-up over Crunch3r's app!

All hail Chicken of Angnor!

352630588 84486295 11 Jul 2006 6:08:14 UTC 11 Jul 2006 21:47:18 UTC Over Success Done 37,801.00 26.50 pending
352628185 84485715 11 Jul 2006 5:58:30 UTC 3 Aug 2006 20:50:47 UTC In Progress
352428506 84437477 10 Jul 2006 17:11:35 UTC 11 Jul 2006 22:53:43 UTC Over Success Done 72,936.00 62.35 pending
352425929 84436891 10 Jul 2006 17:02:55 UTC 11 Jul 2006 14:15:24 UTC Over Success Done 75,673.00 62.45 pending
352424964 84436666 10 Jul 2006 16:59:30 UTC 11 Jul 2006 22:50:06 UTC Over Success Done 101,335.83 58.19 58.19
352422743 84436146 10 Jul 2006 16:52:37 UTC 4 Aug 2006 14:24:08 UTC In Progress Unknown New --- --- ---
352421774 84435923 10 Jul 2006 16:49:14 UTC 12 Jul 2006 2:46:07 UTC Over Success Done 110,398.97 62.45 62.45
352416580 84434705 10 Jul 2006 16:31:30 UTC 11 Jul 2006 10:12:10 UTC Over Success Done 54,954.00 62.46 62.46

---

Mike Bader
BOINC V7.16.5
http://setiathome.berkeley.edu/team_join_form.php?id=5 - Join Our International Team
[img]http://boinc.mundayweb.com/one/stats.php?

Simon & everybody, couple of questions. Is there really any point in compiling an SSE3 optimized client, seeing that SSE2 crunchs faster or in the same time than SSE3?

In what OS / CPU is SSE3 faster ?

Are any of the SSE3 13 new instructions even used? If they are they don't seem to speed up anything.
At least not for my P4 Prescott, and the funny thing is SSE3 instruction set code name is Prescott New Instructions (PNI). :p Go figure.
Anyway, I know for a fact that SSE2 is faster in AMDs too.

/Regards.

---

/Mav

We have lingered long enough on the shores of the cosmic ocean.
We are ready at last to set sail for the stars.
(Carl Sagan)

Yeah I tried sse3 code once and found It was slower than sse2 code, I don't know why though, Simon If You could make an sse3 version that was faster than sse2 ok, Otherwise It would be a waste of cpu time to use sse3.

---

The T1 Trust, PRR T1 Class 4-4-4-4 #5550, 1 of America's First HST's

Well my more recent re-run test benchmark results show SSE3 to be few seconds faster or the same speed as SSE2. Occasionally SSE2 is a few seconds faster. When I re-run the results I seem to get the same picture so I guess it depends some on the AR of the units.
I see the same with general-SSE2 and P4specific-SSE2. P4specific generally few seconds faster but occasionally slower.
This on my Pentium D 830 3.2Ghz(Smithfield) and Pentium 4 HT 3.4Ghz(Northwood). Both Windows XP.

Erm.. got another question. Simon & everybody. ;)

Why is it that the result stderr out displays this msg...

Optimized Windows SETI@Home Enhanced application
Version info: Windows P4 SSE2 32-bit V5.15 'Chicken Good!' (R-1.2|+fversion|xN)
Compiled by Simon Zadra (KWSN - Chicken of Angnor) - Member of the Knights who say Ni! (http://www.kwsn.net)
Download Updates at: http://www.zadra.org/seti_enhanced/

Work Unit Info
True angle range: 0.447784

...eight times instead of just one? Here's an example.

It usually happened with Crunch3r's optimized 5.12 app, too.
Has it got anything to do with the app_info.xml file?

/Regards

---

/Mav

We have lingered long enough on the shores of the cosmic ocean.
We are ready at last to set sail for the stars.
(Carl Sagan)

That's easy to answer - each time the application starts crunching, it will write a part of the stderr. I'm actually working on making it output things only at the end so that doesn't happen anymore.

It happens with the default client, too, only that just outputs one line a run.

So when it gets preempted or you reboot or you just switch it off manually, it will repeat what it put into stderr.txt.

HTH,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

That's easy to answer - each time the application starts crunching, it will write a part of the stderr. I'm actually working on making it output things only at the end so that doesn't happen anymore.

It happens with the default client, too, only that just outputs one line a run.

So when it gets preempted or you reboot or you just switch it off manually, it will repeat what it put into stderr.txt.

HTH,
Simon.

It can actually be handy to see.

I have one that I started with Crunch3r's application before I went on vacation, restarted after I got back from vacation, started again with the new optimized application, and finally restarted a last time when I adjusted my cache size.

[edit]
One other thing to consider - if the program bombs out when will the information about the program version be written?
[/edit]

---

That's easy to answer - each time the application starts crunching, it will write a part of the stderr. I'm actually working on making it output things only at the end so that doesn't happen anymore.

It happens with the default client, too, only that just outputs one line a run.

So when it gets preempted or you reboot or you just switch it off manually, it will repeat what it put into stderr.txt.

HTH,
Simon.

Also, I think the duplicate log entery occurs if you have the prefs set to remove from memory at the tak switch, if you leave it in memory it just picks up where left off with no log entry. I think I'll test that though. :-)

Alinator

<Update> Yep, I just set one of my boxes to leave in memory with a short switch interval, and no log entry when SAH resumed at the last switch.

Simon & everybody, couple of questions. Is there really any point in compiling an SSE3 optimized client, seeing that SSE2 crunchs faster or in the same time than SSE3?

In what OS / CPU is SSE3 faster ?

Are any of the SSE3 13 new instructions even used? If they are they don't seem to speed up anything.
At least not for my P4 Prescott, and the funny thing is SSE3 instruction set code name is Prescott New Instructions (PNI). :p Go figure.
Anyway, I know for a fact that SSE2 is faster in AMDs too.

/Regards.

Diego, did you take a look at the comparison tables?

There are some host/AR combinations where the SSE3 client is markedly faster. And no, early Prescotts are not usually among them - the later models have larger caches and those might help, as well.

In any case, the comparison tables should give you a better idea what's what on various hosts and ARs.

[edit]
One other thing to consider - if the program bombs out when will the information about the program version be written?
[/edit]

I've thought about it, when the program exits abnormally (for whatever reason), a different routine runs that outputs the error message. All that'd need to happen is for me to put the version output in both routines.

Regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

Thanks, Simon, for your instructions (and the use of your cruncher). Since yesterday was Tuesday and it was afternoon here the forums were closed so I never had a chance to see them. However (and I should have done this several weeks ago to reap better results from Crunch3r up 'til now) I decided to grab the bull by the horns and logically figure things out. When I turned off Boinc, downloaded and unzipped your files into my ...berkeley.edu folder I saw that they were buried two folders deep in kwsn-sse-p4-designated folders. I went down those two tiers, moved those individual files onto the desktop, reset the folder windows to ...berkeley.edu and put each file (including the cruncher) there. I removed all other (old) executables and app_infos to the recycling bin. I'll take care of the old text files later. I saw your instructions textfile before doing this. I had to wait several hours until the lines cleared after the maintenance shutdown to see the fruits of my labor. It looked like it worked. I first saw the cache full of about six crunched and 24 uncrunched workunits before and after the exchange, but, in the early evening, nothing in the cache. It looked like everything was fine by 2300 hours GMT when I shut off the monitor. At about 1000 and 1115 GMT today I checked and saw that your cruncher was working fine. The latest check at about 1800 hours revealed that the long-angle-range units were being crunched about 10 to 30 percent faster, but the intermediate ones (like 55- and 62-credit ones) were being crunched about 70-to-80 percent faster (taking 45 percent less time). This makes me feel good. I haven't seen any very-low-angle-range units yet. The (six) units in the cache previously crunched by the default cruncher were saved and credited; the (24) uncrunched units were tossed and redlined on my results list. I have hope and confidence that your cruncher will very significantly outdo what I've been using. Now if I can get the graphics fixed I'll be able to install it on my subordinate computer, also. What or where is Angnor? I couldn't find it on the map of Austria in my encyclopedia.

---

Lol.

It was only a matter of time until someone asked -
Angnor is, indeed, an imaginary place.

As you may have noticed, the KWSN (Knights who say Ni!) are a silly bunch. So, a silly name to go along with that never hurts. Oh yeah, and I have a silly walk I'd like to develop....

Well, if you haven't noticed by now, all this is heavily influenced by the "Monty Python's Flying Circus" crew and all their ventures, chief among them "The Search for the Holy Grail".

My nickname is an obscure reference to a line in that movie. Sir Robin's minstrels keep singing about his glorious deeds.

You can find lots of quotes from the movie here or many other places.

Incidentally, the lines about Robin include "who personally wet himself at the Battle with the Chicken of Angnor".

So there! Any errors in spelling were introduced by me, to befuddle you, the would-be investigator. Ni!

Regards,
Simon.

---

Donate to SETI@Home via PayPal!

Optimized SETI@Home apps + Information

Lol.

Incidentally, the lines about Robin include "who personally wet himself at the Battle with the Chicken of Angnor".

So there! Any errors in spelling were introduced by me, to befuddle you, the would-be investigator. Ni!

Regards,
Simon.

Well, I'm befuddled. I thought it was the Dragon of Angnor and the Chicken of Bristol, and best not to discuss the Battle of Badon Hill. But maybe it was different in other versions. I guess even dragons start small & cuddly. Mike