Seti@Home optimized science apps and information
Optimized Seti@Home apps => Windows => Topic started by: Simon on 08 Jul 2006, 07:08:22 pm
-
Hi everyone,
with the help of a lot of people who dedicated their time to testing, provided feedback and donated for the compiler/library licenses, here is what you've been waiting for:
Windows optimized clients (static executables, 32/64-bit compatible)
There are various different optimized clients. Please use a program like CPU-Z (http://www.cpuid.org/cpuz.php) to find out exactly what your system supports.
Choose the first, in order, your system supports among the two lists (AMD/Intel or Intel only).
Intel-only clients (do not run these on AMD systems!)
Pentium M SSE2-optimized client (http://lunatics.at/index.php?module=Downloads;sa=dlview;id=15)
Only use this client on Pentium M systems like Dothan, Banias or Core Solo/Duo. Use the next ones for all other Intel CPUs.
Pentium 4 SSE3-optimized client (http://lunatics.at/index.php?module=Downloads;sa=dlview;id=14)
Pentium 4 SSE2-optimized client (http://lunatics.at/index.php?module=Downloads;sa=dlview;id=13)
AMD and Intel-compatible clients
SSE2-optimized client (http://lunatics.at/index.php?module=Downloads;sa=dlview;id=12)
SSE-optimized client (http://lunatics.at/index.php?module=Downloads;sa=dlview;id=11)
MMX-optimized client (http://lunatics.at/index.php?module=Downloads;sa=dlview;id=10)
You can find statistics and information about these clients on the Result comparison table (http://lunatics.at/index.php?page=wincomp). Host information like CPU type, RAM, Cache size etc. is included. Please take a look before you download, you may find that for your specific CPU, a different client is quickest.
Please read Instructions.txt for Installation instructions.
These clients include the checkpoint() fix mentioned here (http://setiathome.berkeley.edu/forum_thread.php?id=32193#358828).
Also, please post your results (http://lunatics.at/index.php?action=post;topic=16.0)!
Regards,
Simon.
-
Hi Simon,
I'm sure I followed the very easy directions in the info file and I restarted boinc. I'm only getting client errors and don't know why. If I delete all files in my Seti folder will I be able to download the original files again from the Seti site?
Thank you!
Peter
-
Yes you will. Simply stop BOINC, delete the app_info.xml and .exe file, then restart it. It should redownload the standard app for you.
Did you use CPU-Z to check what your computer supports? Most likely the reason that it's erroring out is you're using a version not designed for your machine.
Also, could you please supply a link to those results? I would be very interested in them.
Regards,
Simon.
-
Yes, I did use CPU-Z. I've got a P4 HT 2.6 and downloaded the SS3 version. I had used the SS3 on one of AKOS files before the new version of SETI. I appreciate what you have done very much and once I am crunching again I will watch the boards to see if anyone else has the same problem.
Thanks again!
Peter
Here is the link to my results.
http://setiathome.berkeley.edu/results.php?userid=8370477
-
No P4 2.6 with HT I know supports SSE3 - maybe I'm wrong. To my knowledge, the right version for you would be the P4 SSE2-optimized version. You can easily see whether a version runs on your computer - simply open a dos box, navigate to where you extracted the file, and run it. It will tell you about problems then and give you the chance to fix things before your WUs on BOINC go wrong.
Regards,
Simon.
P.S. Someone else was adamant his P3 supported SSE2 (they don't). He had the same trouble you did.
-
I'll delete the app_info_xml and the .exe file and download your P4 SSE2-optimized version and let you know how it works!
Peter
-
Hi again,
the link you posted does not work for me :) Please go to the "your hosts" page and post either the link on the left (computer id) or the one to the results for that computer (on the right).
Good luck with the SSE2 version! It should definitely run on your system.
Regards,
Simon.
-
Results from the SSE3
ID 2268317
http://setiathome.berkeley.edu/hosts_user.php
Now running happily on SSE2 just as you said!
Cheers Simon! I'm very grateful!
Peter
-
It's in "Your Account" then "Computers on this account" on the setiathome home page.
HTH,
Simon.
-
I've got three diverse systems running your download. No troubles after I fixed my head-fade about the level of SSE supported by my P3.
The Pentium M WinXP SP2 pro system has returned two validated results
ID351705901 (http://setiathome.berkeley.edu/result.php?resultid=351705901)
ID351688186 (http://setiathome.berkeley.edu/result.php?resultid=351688186)
The Gallatin WinXP Pro SP2(Northwood-descended P4 Extreme Edition) host running hyperthreaded with an Einstein S5 on the other side has a validated result ID351257359 (http://setiathome.berkeley.edu/result.php?resultid=351257359)
The Coppermine Pentium III Win98 SE host has validated on a mixed old/new result
(edited again to add validated result from yet another host)
-
Host ID 1774683 (http://setiathome.berkeley.edu/show_host_detail.php?hostid=1774683) Pentium D 830 3.2Ghz, Dual channel DDR2, Win XP
xP Release Result ID 351696436 (http://setiathome.berkeley.edu/result.php?resultid=351696436) Validated
xP Release Result ID 351696383 (http://setiathome.berkeley.edu/result.php?resultid=351696383) Validated
xP Release Result ID 351696452 (http://setiathome.berkeley.edu/result.php?resultid=351696452) Validated
xP Release Result ID 351696484 (http://setiathome.berkeley.edu/result.php?resultid=351696484) Validated
xP Release Result ID 351696465 (http://setiathome.berkeley.edu/result.php?resultid=351696465) Validated
xP Release Result ID 351696413 (http://setiathome.berkeley.edu/result.php?resultid=351696413) Validated
xP Release Result ID 351696500 (http://setiathome.berkeley.edu/result.php?resultid=351696500) Validated
xP Release Result ID 351696385 (http://setiathome.berkeley.edu/result.php?resultid=351696385) Validated
xP Release Result ID 351696367 (http://setiathome.berkeley.edu/result.php?resultid=351696367) Validated
xP Release Result ID 351696422 (http://setiathome.berkeley.edu/result.php?resultid=351696422) Validated
xP Release Result ID 351696488 (http://setiathome.berkeley.edu/result.php?resultid=351696488) Validated
xP Release Result ID 351696480 (http://setiathome.berkeley.edu/result.php?resultid=351696480) pending
xP Release Result ID 351696498 (http://setiathome.berkeley.edu/result.php?resultid=351696498) pending
xP Release Result ID 351696440 (http://setiathome.berkeley.edu/result.php?resultid=351696440) pending
All AR=1.090487
[EDIT]
Added 2 results and updated couple to Validated
[EDIT2]
Added another 2 results and updated another to Validated
[EDIT3]
Added 4 more and updated 1 to Validated
-
Hello!
Since I have your "setiathome-5.15-kwsm-mmx.exe" on my AMD K7 600 MHz I have a few problems with my PC.
After a little time working on my PC, open and closed a few times the Internet Explorer, it comes a error message like:
"Internet Explorer have become an error, it must be closed! Would you like to send the error message to Microsoft?"
And I cannot open other windows.
After a long time I come back and switched on the display, there was an error message:
preupd.exe
Die Ausnahme "unknown software exception" (0xc000001d) ist in der Anwendung an der Stelle 0x77c0f928 aufgetreten.
Klicken Sie auf "OK", um das Programm zu beenden.
And a lot more error messages comes too.
Now, this is because of the new app?
On my AMD K8 3200+, he is running 24/7 without working other programs, no problems.
A True angle range: 0.051058
CPU time 14432.40625 with your V5.15
CPU time 13914.15625 with Cruch3rs V5.12
So yours is 518,25 seconds slower.
Only one experience. :-)
Greetings!
:-)
-
Host ID 2268317 Pentium(R) 4 CPU 2.60GHz, XP, SSE2-optimized 5.15 client
Result ID 351725476 Validated
Result ID 351791673 Validated
Result ID 351911138 Pending
Result ID 351964959 Pending
40-45% speed up!
Thank you so much Simon!!
Peter
-
Hi Dirk,
yeah, I noticed that his apps run more quickly on AMD systems, mostly. Shows me there is still room for improvement :)
I don't know what kind of trouble the MMX app is making on your slower Athlon - did you run Crunch3rs MMX app on it before without trouble?
Speicher testen mit Memtest - http://www.memtest.org z.B. - klingt für mich so, als ob Windows keinen virtuellen Speicher (Auslagerungsdatei) mehr frei hat und deswegen die anderen Programme abkacheln.
Regards,
Simon.
-
Hi Simon,
Your SSE version seems to be runing well on:
AuthenticAMD AMD Athlon(tm) XP 2000+
Microsoft Windows XP Home Edition, Service Pack 2, (05.01.2600.00)
No results yet to report.
I am sorry to ask for more, but could you please find a way to make the cliets self-update? I'm sure many others also run multiple machines in multiple locations. Manual updating is quite tedious.
Thank you for your time and effort.
http://setiathome.berkeley.edu/show_user.php?userid=41362
Roger Bannon, Lawton, Oklahoma, USA
-
You do realize you're asking a lot there, right?
Auto-updating is a feature I've thought some about, it would be pretty work-intensive to make from scratch. Nevertheless, it's on my To-do list, and will be announced if I manage to do it.
Regards,
Simon.
-
You do realize you're asking a lot there, right?
Auto-updating is a feature I've thought some about, it would be pretty work-intensive to make from scratch. Nevertheless, it's on my To-do list, and will be announced if I manage to do it.
Regards,
Simon.
I don't know much about this but would it be easyer to modify Boinc instead to look at the file versions on your site for updates periodically instead of putting it in the app? If someone wanted to get really fancy sometime in the distant future i hear theres a new boinc beta that detects cpu instruction abilities
-
I am sorry to ask for more, but could you please find a way to make the cliets self-update? I'm sure many others also run multiple machines in multiple locations. Manual updating is quite tedious.
Hi Roger
You can have a look at BoincStudio, by DocMaboul.
A damned good free tool that can do "one-shot updating" on multiple machines, and much more.
Still in beta version, but I use it on a ~ 40 rigs farm, with good satisfactions :)
International Technical Support Forum : http://forum.boincstudio.boinc.fr/boincstudio/support-international/liste_sujet-1.htm
Website : http://boinc-studio.nikolai.be/index.php?lang=en
Bets regards, Franck.
-
Hi Dirk,
yeah, I noticed that his apps run more quickly on AMD systems, mostly. Shows me there is still room for improvement :)
I don't know what kind of trouble the MMX app is making on your slower Athlon - did you run Crunch3rs MMX app on it before without trouble? ...
---CUT---
...Regards,
Simon.
Hello Simon!
Thanks for your work!
Before I had your Windows MMX 32-bit V5.15 'Chicken Good!' (R-1.2|+fversion|mmx) on my AMD K7 600 MHz my PC worked fine.
Now back with Windows MMX V5.12 by Crunch3r it´s O.K. again...
It was only my experience maybe there are more people who have MMX PCs too to tell us their experiences...
And I changed the app too on my AMD K8 3200+ back to Cruch3rs app... because his is faster... ;)
No no I don´t want to reduce your work!
No it´s only like this, that I would like to have the fastest app for my PC. :)
Maybe you can ask Crunch3r how he had done it with his apps for the AMDs...?
Or he can have a look to your apps...?
Or all the good guys who made the old optimized apps come together (of course you too, because you are a good guy too! :) ) and you all together make super special apps...
I saw on the community side of S@H that Sir Ulli have the same CPU like me, maybe you can have a look, or we send you our ARs and times...?
Maybe this help you to optimize your app?
My AMD K8 3200+ with Windows SSE2 V5.12 by Crunch3r
http://setiathome.berkeley.edu/results.php?hostid=2176124
Sir Ullis AMD K8 3200+ with your Windows SSE2 32-bit V5.15 'Chicken Good!' (R-1.2|+fversion|xW)
http://setiathome.berkeley.edu/results.php?hostid=2488963
I would like to support you too!
Greetings!
:)
EDIT:[/color]
But I don´t know the Version of Sir Ullis K8 3200+ (754 or 939)
My is the 754, and I have 256MB more RAM (3x256MB)
-
Dirk,
Ich versteh Dich schon ;) Ich verwende selber auch noch auf einigen Hosts Crunch3rs Programme, weil meine zwar fast, aber eben nicht ganz so schnell sind auf AMDs.
Anyway - there is always room for improvement :) I've been in contact with a lot of the "old guard" of optimizers. Some hints need to be investigated, yet, but I really don't think Crunch3r is coming back to S@H, because I asked him for exactly that (or maybe some hints or his sources), and he said no (which I fully understand and accept).
Regards,
Simon.
-
Hi Simon,
richtig, da ist immer Raum für Verbesserungen. Wäre ich momentan nicht so im Stress (Diplomarbeit), würde ich mich da sehr gerne einklinken und weiterhelfen. Ich hab' in den paar freien Minuten, die mir so bleiben, als "Fingerübung" den Seti-Client (basierend auf deinen Sources) speziell für meinen Pentium D optimiert. Man glaubt gar nicht, wieviel Luft da noch ist. Aber falls Du Detailfragen oder Tipps hast, nur zu! Ich arbeite nun schon eine gewisse Zeit mit IPP/ICC.
Grüße,
Michael (aka TheGreatCornholio)
P.S.: Entschuldige meinen "Rückfall ins Deutsche", geht so deutlich schneller...
-
Da ich auch einen Pentium-D habe, bin ich sehr interessiert! Gibt's ne Chance, an ein Archiv Deiner veränderten Sourcen zu kommen?
Und: ich kann leider noch immer kein C/C++ selber...nur Code lesen und einfache Syntax errors erkennen. Daher wäre Mitarbeit von jemandem der da mehr kann wirklich sehr hilfreich!
Diplomarbeit geht vor, da stimme ich Dir zu :) Jede Hilfe ist willkommen.
Danke & schönen Gruss,
Simon.
-
Dirk,
Ich versteh Dich schon ;) Ich verwende selber auch noch auf einigen Hosts Crunch3rs Programme, weil meine zwar fast, aber eben nicht ganz so schnell sind auf AMDs.
Anyway - there is always room for improvement :) I've been in contact with a lot of the "old guard" of optimizers. Some hints need to be investigated, yet, but I really don't think Crunch3r is coming back to S@H, because I asked him for exactly that (or maybe some hints or his sources), and he said no (which I fully understand and accept).
Regards,
Simon.
Hello!
I don´t want to urge or kick you!
Only a small question:
When you have a faster app for AMDs, what is then the name, maybe V5.16?
That I know, that I can come back to you! ;)
Greetings!
:)
-
Da ich auch einen Pentium-D habe, bin ich sehr interessiert! Gibt's ne Chance, an ein Archiv Deiner veränderten Sourcen zu kommen?
Hi Simon,
sobald ich wieder etwas Zeit gefunden habe, werde ich die letzten Änderungen noch komplettieren und Du kannst dann gerne alles haben - so quasi als Anschauungsmaterial und zum "Spicken", falls Du etwas verwenden willst. Ich muss Dich aber gleich vorwarnen, ich habe einiges an Code 'rausgeschmissen, was sich nicht auf IPP bzw. SSE bezieht. Vorab kann ich Dir aber schonam den Tipp geben (der ist aber nicht neu), dass es viel bringt, die Sin/Cos Berechnung zu cachen (also vorab in einem Array ablegen). Bei IPP gibt's dazu ja z.B. die Funktion ippsSinCos_64f, die Dir ein Array damit füllen kann. Ein weiterer Tipp ist auch, die v_GetPowerSpectrum zu ersetzen, die gibt es fast 1:1 auch in IPP. Sämtliche Mittelwertberechnungen und Standardabweichungen kann IPP auch sehr zügig durchführen. Schau Dir auch mal das EXP-Macro an, das ist eher suboptimal... Einige Sachen, die ich hier angeführt habe, sind zwar nicht wirklich entscheidend für die Gesamtperformance, allerdings sind sie quasi ein Hinweis darauf, dass man auch an kritischen Stellen noch Verbesserungsmöglichkeiten findet.
Ich hab' schon gelesen, dass Du deinen D auf 4GHz getaktet hast... Ich hoffe, Du hast ein gutes Board! Mich hat dieser Prozessor schon zwei Boards gekostet (naja, waren billig). Damit kann man auf jedenfall einiges Rechnen ;D
Das bringt mich noch zu einem weiteren Vorschlag: bau' doch auch in deinen Client die Anzeige der tatsächlichen Prozessorfrequenz ein, das schafft auch anderen Leuten einen besseren Überblick über die Leistungsfähigkeit des Prozessors, wenn man Ergebnisse vergleichen will (ippGetCpuFreqMhz()). Sonst wundern sich die Leute noch, wie ein 2.66GHz Pentium D schneller sein kann als ein 965er ;D
Grüße,
Michael
-
Ahh, sehr gute Idee das mit der MHz-Anzeige (und danke für die Nennung der relevanten IPP Funktion ;) ). Vor allem, weil BOINC leider nur die CPU ID ausliest (und damit den Original-Speed) und dazu Benchmarks macht - aber keinerlei MHz-Informationen, was mich schon geärgert hat.
Die anderen Dinge schau ich mir gerne mal an, diese Sache gibt mir genug Anstoß, nun endlich mal ernsthaft C/C++ zu lernen.
Die 4 GHz liefen zwar unter Prime95 stabil, ich hab aber immer wieder "Machine Check Exception" Single-Bit (korrigierbare) ECC-Fehler bekommen. Bin jetzt in mehreren kleinen Schritten bis auf 3.86 GHz runter, nun läuft er schön stabil ohne irgendwelches Gemurre. Scheint so, als ob der Cache auf CPU1 nicht ganz so flott laufen will wie auf 2 (die Fehler waren alle von der ersten CPU).
Mein Board ist ein Asus P5ND2-SLI und hat stolze 76 Euro gekostet. Overclockt allerdings sehr fein (ich könnt noch wesentlich mehr Volt reinbraten, aber das muß ja nicht sein...Luftkühlung).
Das Board davor (AsRock PT-880 Dual, ~63 Euro) hat's auch gebraten - ein bisserl Elektroschnee...irgendwas in der Stromversorgung hat sich durchs Gehäuse verteillt. Da ich das nicht nochmal wollte, hab ich dem Verkäufer mal bisschen Gas gegeben und ihm alles mögliche von wegen Feuergefahr und unverantwortlich erzählt, weil ich ein besseres Board wollte...hat funktioniert ;) Aufpreis gezahlt und danke.
Danke für die Hinweise, und ich warte auf Deine Sourcen :)
Gruß,
Simon.
-
Hallo nochmal,
ein kleiner Vorgeschmack (sicher nicht 100% repräsentativ): http://setiathome.berkeley.edu/workunit.php?wuid=83843919. Das oberste Ergebnis ist von mir, das drunter mit dem originalen Crunch3r Client. Beides sind P4 mit 3.4GHz.
4 GHz waren bei mir auch nicht stabil zu bekommen, zumal bei diesen Außentemperaturen jede Wärmequelle im inneren mich noch mehr ins Schwitzen bringt. Die zwei abgerauchten Boards waren Asus P5P800SE, jetzt mit dem ASrock geht alles erheblich stabiler - auch bei "nur" 3.4GHz.
Grüße,
Michael
-
:o :o :o :o
Wow?
Na da steckt ja wirklich noch massiv viel drin...ich dachte mir eigentlich, daß Caching jetzt schon in den Standard-Sourcen aktiviert wäre (das war eigentlich der massivste Speedup bei den prä-enhanced clients gegenüber default).
Ich bin grob erstaunt - und noch neugieriger :)
Das mit der Hitze ist bei mir auch nicht besser...derzeitige Zimmertemperatur tagsüber ~29C, nachts 23-25C. Wenigstens ist das Wetter mal schön!
Simon.
-
Hi,
das Ergebnis ist natürlich schon seeehr vorteilhaft, andere waren knapper ;D. Caching ist aktiviert, nur wird das im Standard-Enhanced-Client in einer Schleife, die eine Million mal durchlaufen wird, jedes einzelne mal abgefragt, ob Caching oder nicht. Man kann sich die Vorteile auch mutwillig zunichte machen :D. Ich hab dieses Caching grundsätzlich aktiviert, weil mein Standpunkt ist, dass so viel Speicher wirklich in jedem halbwegs aktuellen Rechner übrig sein müsste (16MB). Dann erübrigt sich auch die permanente Prüfung, ob Caching - ja oder nein. Zwei wichtige Schleifen (beim Chirping) hab' ich noch per Hand mit Assembler und SSE2/3 ein bisschen auf Leistung getrimmt. Das dürfte theoretisch für Leute interessant sein, deren Compiler nicht sooo gut bei der automatischen Vektorisierung ist (gnu).
Naja, und sind da halt noch die vielen, vielen kleinen Änderungen, die mir so eingefallen sind.
Grüße,
Michael
-
Na dann kann ich nur noch hoffen, daß Du kein "echter Programmierer" bist, weil die ja bekanntlich nie Kommentare schreiben...*kicher*
Das mit dem Caching klingt sehr einleuchtend...komisch, daß das noch keinem von den Codern selber aufgefallen ist, die sind eigentlich nicht auf den Kopf gefallen.
Gruß,
Simon.
-
Na dann kann ich nur noch hoffen, daß Du kein "echter Programmierer" bist, weil die ja bekanntlich nie Kommentare schreiben...*kicher*
:o erwischt! Keine Angst, ich schreib schon noch ein paar Kommentare dazu... Darf mich ja bald Dipl. Inf. schimpfen, aber das heißt wirklich garnix.
Ich werde sicher auch haufenweise potenzielle Verbesserungen übersehen... VTune hilft dabei aber schon, dass das nicht passiert. Das Caching ist ja auch eine sehr gute Idee und auch gut gemacht, da gibt's nix dran auszusetzen.
Ich denke mal, dass man als Entwickler auch nicht permanent Zeit hat, auf's letzte Prozent die Geschwindigkeit zu optimieren. Schließlich "lebt" das Projekt und es gibt dauernd Änderungen/Verbesserungen. In gewissem Maß macht so eine Optimierung den Code halt einfach schlechter lesbar/wartbar.
Grüße,
Michael
P.S. Ich sehe gerade, ich bin "befördert" worden! Jetz' hab ich rote Bäckchen...
-
Ja, so sehe ich das auch - der Entwicklungsprozeß ist eher organisch als fix reglementiert - CVS eben.
VTune muß ich erst verwenden lernen, bisher hat mir die Analyse ganz genau überhaupt nix gesagt. Schon mal probiert, aber einfach keine Ahnung. Muß mir mal die ganzen Dokumente von Intel dazu ansehen, die Doku ist eh sehr umfassend und gut.
Gruß,
Simon.
-
For all non german speakers - Monza_MH26 is a capable coder and has further optimized the sources I put online to make a potentially much quicker binary. So I've been pestering him to send his sources to me and he agreed - and we've been waxing lyrical about that in german.
Plus a lot of helpful hints, some of which I've already put into my sources (like the CPU real frequency...really good idea).
So things are moving forward, and please forgive our german conversation - it just makes things much easier, and I'll translate the important bits anyway.
Regards,
Simon.
-
For all non german speakers - Monza_MH26 is a capable coder and has further optimized the sources I put online to make a potentially much quicker binary. So I've been pestering him to send his sources to me and he agreed - and we've been waxing lyrical about that in german.
Plus a lot of helpful hints, some of which I've already put into my sources (like the CPU real frequency...really good idea).
So things are moving forward, and please forgive our german conversation - it just makes things much easier, and I'll translate the important bits anyway.
Regards,
Simon.
Thats fantastic news. :) Thanks Monza_MH26
Simon, I notice your Linux clients downloads have reached over 600 and your Windows clients downloads over 1500. WOW.
-
Yeah, I'm amazed at the response generated, myself :)
Trying not to let my ego swell too much, but it does feel good. Also, 235 source packages have been grabbed, which I never really expected at all.
Regards,
Simon.
-
Have just finished all 7 kwsn-test benchmark WU's and would like to thank you for providing them - no probs with any installation - no corrupt files - preserved the old directory for the new optimised ones - everything running fine.
But... can you point me to which file in the kwsn-test benchmark directory is the result one to send you? Do i post them here (how) or email it/them to you (where)? Sorry for asking but i just couldn't find the answer here.
-
Thanks!
Basically, you could zip up the testData/ directory and attach it to a post here. Also, you could post your times using the various WUs and clients here.
Right now, I'm working on something that will make returning results much easier, but it'll still take a bit of time.
Regards,
Simon.
-
testData results for an
Intel P4 640 3.20GHz
Intel D865GSA
2048Mb DDR-SDRAM
[attachment deleted by admin]
-
Thanks!
Basically, you could zip up the testData/ directory and attach it to a post here. Also, you could post your times using the various WUs and clients here.
Right now, I'm working on something that will make returning results much easier, but it'll still take a bit of time.
Regards,
Simon.
Okay, I am without clue. :o
How do you run the test work units and get a testData directory?
-Tim
edit: Never mind, I found how to run the benchmarks
http://lunatics.at/index.php/topic,22.msg227/topicseen.html#msg227
-
Okay, I'm done running the benchmarks.
For my:
CPU AMD Athlon 64 X2 3800+
mobo ECS nForce4-A939
RAM 1024MB DDR CAS3
[attachment deleted by admin]
-
Simon,
I have been running your apps on two of my machine and I noticed a major difference (20-40%) in average turnaround time. This is just an estimate that I took from a set of 10 wus with the same AR both before and after the switch. I was planning on running the test WUs but was unsuccessful...but that is not to big a deal. However if I could get it to run I will glady post the results.
I just downloaded the MMX app for an Athlon 900 MHz system I was running Crunch3r's app (5.12) on the same machine and when I switched over to your app the "message" section of BOINC still says that I am running "S@H_enhanced version 512". I checked the app_info.xml file and it says that it is looking for the 5.15 app....
I would appreciate your input.
Thanks
citroja
ALSO - any chance of an SSE3 for AMD cpus???
-
SSE3 for AMD CPUs is slower than SSE2 - for this reason, I won't be releasing an SSE3 version.
It's normal that it still says "5.12" as version for WUs you started out with Crunch3r's app - the app_info.xml I included specifically resumes any WU from v 5.12 to v 5.15. You could open your task manager, it will tell you what app is really running (his has a different file name).
HTH,
Simon.
-
SSE3 for AMD CPUs is slower than SSE2 - for this reason, I won't be releasing an SSE3 version.
It's normal that it still says "5.12" as version for WUs you started out with Crunch3r's app - the app_info.xml I included specifically resumes any WU from v 5.12 to v 5.15. You could open your task manager, it will tell you what app is really running (his has a different file name).
thanks on both items...still working on the test WUs.....
Citroja
-
Greetings.
Tried the Pentium M version on the IBM T42 laptop tonight:
-1.7 ghz Pentium M
-1.0 gigabyte ram
-40 gig internal drive
-Windows XP Pro SP2
Looks about 15627 seconds for the first work unit it encountered,
I need to see more to get an idea of how fast it truly is over
24 hours or so.
Here is the workunit: http://setiathome.berkeley.edu/result.php?resultid=364924312
Not validated yet, but I expect soon.
More data as I get it...
-
Update - it appears, although I cannot be completely sure that the speed
of this optimized client is as fast or slightly faster than Cruncher's Pentium M
client but with the added bonus of slightly better RAC.
So, good show so far!
-
Update - it appears, although I cannot be completely sure that the speed
of this optimized client is as fast or slightly faster than Cruncher's Pentium M
client but with the added bonus of slightly better RAC.
So, good show so far!
I can let you know sometime this weekend if you want another opinion...I have the same system setup as you except I have the 1.6 GHz chip. I believe that it is running crunch3r's app but i haven't looked at the computer in about a month (i sold it to my cousin).
citroja
-
i just head that Crunch3r is working on a 64 Bit App that is 20 Percent faster than the old one,,
did someone also working on this...
Greetings from Germany
Sir Ulli
-
Good for him, maybe this time the source code will be (and stay) available ;)
Yes, in fact, people are working on 64 Bit versions.
Regards,
Simon.
-
Good for him, maybe this time the source code will be (and stay) available ;)
Yes, in fact, people are working on 64 Bit versions.
Regards,
Simon.
for Info
did you also are working on an 64 bit Version, and when it will be able to download...
only for Info
Greetings from Germany NRW
Sir Ulli
-
Wow.
Machine K7S5A Motherboard with AMD Athlon 2400 XP processor
running HoneyX bios from two years ago, now running new modified
HoneyX OC bios + SSE enabling. Processor clocked to 2.16 ghz.
I had my linux box running the stock client:
http://setiathome.berkeley.edu/results.php?hostid=2502491
See results in the 35,000 second range.
THEN I updated the bios to allow for SSE operation and formatted
a drive in that machine to run windows and added the SSE client at
this site for windows and got results in the:
http://setiathome.berkeley.edu/results.php?hostid=2650492
9300-9400 range. UNBELIEVABLE improvement. Exponential improvement!
Part of it is SSE but the other part is the client.
Thanks. Hugely.
-
i just head that Crunch3r is working on a 64 Bit App that is 20 Percent faster than the old one,,
did someone also working on this...
Greetings from Germany
Sir Ulli
Well isn't this sweet, i've said many times that converting to 64bit should improve the speed due to more registers to work with etc.. This seem to be the proof if Crunch3r have optmized app .. Pity that the executables won't show up in the community but i surely understand why it never will after the whole mess from the community..
As soon as there will show up 64 bit executables i'll start to dust of my x64 em64t pro eval and install it.. :)
Kind Regards Vyper
-
Proof will be in the pudding. Sir Uli may have heard it is 20% faster...but until we see it...
-
Will I for one have given up on clearing the errors and getting a 64bit compiled. I've been trying to weeks with no success. So last week I just packed it in.
I can't do :'( sniff sniff.
-
Is there a timeline for releasing the x64 app?
-
Has there been any thought to a Core 2 Duo (SSE3, SSE4, EM64T) specific build? I don't know if the SS4 instructions will provide any gain.
-
What ever happened to Babak Delkhoon (DeNitro) 64bit apt?
Any updates? News? :o
-
Simon and a couple other very talented guys have been working on optimizations to the client and I've been working with them incorporating the changes into EM64T builds to test. The progress on the whole range of clients is looking great.
--Babak Delkhoon
-
Thanks for the update DeNitro.
-
OK guys here is part of yesterdays boinc_opt list:
Crunch3r wrote:
Date: Thu, 19 Oct 2006 17:15:17 +0200
From: "Crunch3r" <setiathome@gmx.net>
Subject: Re: [Boinc_opt] Is there a 64bit version available?
Take a look at http://calbe.dw70.de
There's the archive containing the 64 boinc client and manager.
Note:
1. install the 32 bit 5.4.11 boinc first
2. then you need to replace all files in your boinc directory with the ones
in the archive.
End of quote
I am running Linux only, but maybe some of you guys want to try and report how much faster this client really is? ;)
-
Hi Chboss,
he only has a 64 bit BOINC client online, not a S@H science app. So that still may help because according to him, it sets CPU affinity and does a few other tricks - AFAIK those are all in the 5.7.xx codebase and will eventually be available as a "stable" version (5.7.xx is the development tree).
But, it won't change a lot for S@H crunching. The good thing about it is that now you can easily run a 64-bit Windows science app via BOINC, which previously didn't work for everyone (for the record, it did for me).
Regards,
Simon.
-
Hi Simon
You are right I misread and misunderstood his mail... ::)
However maybe the first step to a 64bit S@H science application?
In any case I am not too interested in the Win version, I have only 64bit Linux and Mac boxes left in my farm. ;)
Thanks for your update.
-
Well I am interested in a 64bit app and Boinc install(client) for Windows, I'm using DeNitro's Test2 SSE2 app and Crunch3r's 64bit files(compiled from 5.7.0 source code overlaid on top of some of the 5.4.11 Boinc files, Then It's 5.7.64. I also had to redo all the app info files as they were causing errors(parsing problems).
http://setiathome.berkeley.edu/forum_thread.php?id=34997
executable and main_program was like this
<executable/> and <main_program/> Every time I did a benchmark the app would crash, Done like below and no more crash or erasing of files by Boinc.
<executable> and <main_program>
</executable> and </main_program>
The 1st pair was missing from each statement, Complete examples are in one of My posts at setiathome, The incorrect examples are found in all of the Chicken's And DeNitro's app info files and need to be changed.
Note: You need XP x64 to use an XP64 program(For those that don't know already).
-
The auto installer appears to get stuck on my system whilst its running the benchmark tests, even if I try the short option nothing happens after 10 - 15 mins, is that normal or does it usually take alot longer to run its tests?
-
That does seem a bit long.
What sort of system do you have, and when you run task manager, how much CPU usage does it show while benchmarking?
Regards,
Simon.
-
Managed to get it to run on the short option took 18 mins, It a AMD Sempron 3000+, 1GB RAM, Win XP, CPU showed around 97% one thing it didn't seem to do is stop running the tests after it told be it had finished them, had to kill the process.
Going to try the long option now see how loing that takes :P
-
Ok well it took approx 50 mins to do it on the long test, but its now all worked ang installed and this time it also stopped all the processes correctly 8)