Welcome to the MacNN Forums.

If this is your first visit, be sure to check out the FAQ by clicking the link above. You may have to register before you can post: click the register link above to proceed. To start viewing messages, select the forum that you want to visit from the selection below.

You are here: MacNN Forums > Community > Team MacNN > New Altivec-enhanced Seti worker in need of testing

New Altivec-enhanced Seti worker in need of testing (Page 2)
Thread Tools
Mac Elite
Join Date: Jan 2002
Location: Mile High City
Status: Offline
Aug 27, 2005, 02:12 PM
 
The Spice.. er, Credit is finally flowing again on SETI. I've had another 3 alpha client units validated (4 total so far with no errors or invalids), another 72 units to go.

UPDATE: I now have 9 valdiated work units. No errors. Validation is for the orginial 7450, 7400 and G5 clients.
UPDATE #2: Up to 14 Validated units and no errors. Still hasn't gotten to the latest version of the G5 client which was showing some promise before things went down.
( Last edited by Shaktai; Aug 27, 2005 at 07:25 PM. )
     
alexkan  (op)
Forum Regular
Join Date: Aug 2005
Location: Cupertino, CA
Status: Offline
Aug 27, 2005, 07:03 PM
 
Good to see that the validations are starting to arrive in larger numbers now. Anyways, for those of you who thought alpha-2 wasn't fast enough (although I suspect that with the issues SETI BOINC has been having lately, not many of you have crunched that many units with the new client), I've posted alpha-3, which is about 15% faster than alpha-2 on my machine and should be a bit more friendly than the previous versions to processors that have relatively small L2 caches and/or no L3 caches. This one does Gaussian fitting in a slightly different manner from alpha-1 and alpha-2 for speed purposes, and from my tests affects calculated chi-sqr values a tiny bit (but within validation limits).

As usual, I've modified the first post in this thread to point to the new compiles. If for some reason you want to go back to older compiles, I've linked to the alpha-2s as well, and the naming convention should be relatively obvious if you want to go back further. School is starting soon, though, so I'm going to have to move my code to another computer so I can keep working on it. I sure hope that they've upgraded the Macs at school to run Tiger.

I should really stop saying that there isn't much room for improvement whenever I post a new version, so I won't say it this time.

Happy crunching, and be sure to post your work unit times and which client you're running if you're using the new alphas! Otherwise, I won't know if an optimization that helps 7400s actually hurts other architectures.
( Last edited by alexkan; Aug 27, 2005 at 09:06 PM. )
     
Junior Member
Join Date: Jul 2005
Location: USA, Virginia
Status: Offline
Aug 28, 2005, 02:49 AM
 
Here are 77 results from a 2 Ghz G5 iMac running OS 10.4.2 to analyze. I had to take this system out of the mix today, after all I did buy it for my mother-in-law, and now that we have tested it it has been delivered.

Four of the results from this system have been through the validator, but only 3 passed. I can not tell what was wrong with the fourth one it exited normally but for some reason did not pass. At the rate the validator is progressing I would expect to see the first results from the 10.3.9 app to start going through validation late in the day (eastern time) Sunday.

http://setiathome.berkeley.edu/resul...hostid=1297955

Regards
Phil
We must seek intelligent life on other planets as it is increasingly apparent we will not find any on our own.

Link: http://www.boincsynergy.com/images/stats/comb-2033.jpg
     
alexkan  (op)
Forum Regular
Join Date: Aug 2005
Location: Cupertino, CA
Status: Offline
Aug 28, 2005, 03:03 AM
 
Originally Posted by Snake_doctor
Four of the results from this system have been through the validator, but only 3 passed. I can not tell what was wrong with the fourth one it exited normally but for some reason did not pass. At the rate the validator is progressing I would expect to see the first results from the 10.3.9 app to start going through validation late in the day (eastern time) Sunday.
Hmm, that's a bit worrisome. I don't suppose you remember which client you were running for those particular results? It's possible, but I never found out for sure, that the G5 compile of alpha-1 could have been producing incorrect results because the broken precaching was zeroing out data where it wasn't supposed to, which was why I removed the link to it. I should have had the foresight to make each of these clients print their version number to stderr so I could track which compiles are causing issues. I suppose that's a fix for future versions.

If anyone else has failed validations and happens to know which version of the client produced them, please let me know ASAP.
     
Mac Elite
Join Date: Jan 2002
Location: Mile High City
Status: Offline
Aug 28, 2005, 04:08 AM
 
If recollection serves me correctly, my first 4 were 7450 alpha1. Second 4 were 7400 alpha1 and the next 13 were 970 alpha1. So far, so good on all of them that have validated which is about 20 units. The 22nd unit was switched to 7400 alpha2. The last 10-12 that show pending, and the ones not yet able to be returned, were done with 970 alpha2 however.

Here is the URL for my G5's 76 work units. http://setiathome.berkeley.edu/resul...hostid=1287902
They can be easily monitored as they validate or not. As fast as the validators are working, I expect that all will be validated by Sunday Evening Pacific time.

So 20 out of 20 have validated and those involved using all three of the alpha1 apps on my G5. By tomorrow evening, the URL above should give a much bigger picture across a large cross sample of units.

Once upload/download are re-enabled (probably Monday) then I will be able to test the 970 alpha3.
     
Dedicated MacNNer
Join Date: Sep 2004
Location: London
Status: Offline
Aug 28, 2005, 07:37 AM
 
I have a total of 274 wu's on my account including those waiting to upload. So far as I can see, my returned results all seem to have validated ok.

I looked at Snake doctors list ( http://setiathome.berkeley.edu/resul...hostid=1297955)
and could not see the invalid wu. What am I missing seeing?

The results below are from Shaktia's posted online utility for users to check their pending credit and average work unit rate: http://ztb.dyndns.org/pc.php

last wu received 23 Aug 2005 16:34:33 UTC
pending C 1,856.46
granted C 68,885.71
total C 70,742.17
estimated GC 2,346.86
estimated TC 71,232.57
# wus 126
Ø seconds/wu 6,707.01
Ø time/wu 1h 51m 47.011s
Ø CC/wu 14.73
Ø GC/wu 18.63


For comparison here are the previouse results.

last wu received 23 Aug 2005 16:34:33 UTC
pending C 3,172.78
granted C 67,022.49
total C 70,195.27
estimated GC 4,054.44
estimated TC 71,076.93
# wus 226
Ø seconds/wu 6,707.01
Ø time/wu 1h 51m 47.011s
Ø CC/wu 14.04
Ø GC/wu 17.94

No wu's done on V2 moving up to V3.

K.
     
Junior Member
Join Date: Jul 2005
Location: USA, Virginia
Status: Offline
Aug 28, 2005, 11:51 AM
 
Alexkan=
So far I have only one failed result (linked below). My recollection is that it was running the V1 alpha for the first day or so then was switched to v2 but I could be wrong. If I am correct then the result in question would be a V1 result.

The OS 10.3.9 results are pouring in and so far all ok. I had a problem on the G4 dual that that caused results to be sent but but not recorded. as a result there are a loy of "no returns" in the list from the early part of last week. The problem seemed to be related to P@H and not a result of the S@H alpha.

All results from the G5 are from S@H Alpha test.

G5 OS 10.4.2 2GHz- http://setiathome.berkeley.edu/resul...hostid=1297955
G5 Failed result - http://setiathome.berkeley.edu/resul...ltid=101102324

Results after 8/16/05 for both of the systems below are from the S@H alpha.

G4 OS 10.3.9 1.4GHz Powerbook - http://setiathome.berkeley.edu/resul...hostid=1200967

G4 OS 10.3.9 Dual 1.4GHz - http://setiathome.berkeley.edu/resul...hostid=1199152

Regards
Phil
We must seek intelligent life on other planets as it is increasingly apparent we will not find any on our own.

Link: http://www.boincsynergy.com/images/stats/comb-2033.jpg
     
Mac Elite
Join Date: Jan 2002
Location: Mile High City
Status: Offline
Aug 28, 2005, 03:56 PM
 
iMac G5 OS-X 10.4.2 1.6ghz
33 units granted credit with no errors and not showing invalid.
Apps include 7450 alpha1, 7400 alpha1, 970 alpha1 and 7400 alpha2.
Average time per work unit for all apps is 01:43:33 (hh/mm/ss)

Still looking good.

Update: 56 units now validated. No errors.
( Last edited by Shaktai; Aug 29, 2005 at 09:29 AM. )
     
Dedicated MacNNer
Join Date: Aug 2005
Location: Golden Valley, AZ
Status: Offline
Aug 29, 2005, 11:23 AM
 
Just downloaded the 10.3.9 build, now I just have to wait for S@H to go back live. It seems that the validator has caught up with the backload.
     
Bill Michael
Guest
Status:
Aug 30, 2005, 01:54 AM
 
resultid=104339842 - because of the delay in validation, I can't be 100% positive, but I believe this one was with a2. The date returned vs. the modification date on a2 says this one was a1, but it is the last result I returned from the Mini before SETI went down and I was on a2 by then. Once it's back up, I'll have more a2 WUs and will be able to give a better answer, as I _know_ those will all be a2.

The speed difference makes this definitely worthwhile to try... but if _any_ significant percentage are invalid, then it makes the code 'suspect' for use, credits or not. So far though, more seem to be valid than not. Maybe off "just enough" so that timing of results comes into play? SZTAKI had this problem - Linux vs. Windows, whichever returned a result "first", that platform would always validate, the other would not. They had to change the validator criteria for "close enough" to make things work right on that project. I'm not sure how critical absolute "match" is for SETI.
     
alexkan  (op)
Forum Regular
Join Date: Aug 2005
Location: Cupertino, CA
Status: Offline
Aug 30, 2005, 03:18 AM
 
Originally Posted by Bill Michael
resultid=104339842 - because of the delay in validation, I can't be 100% positive, but I believe this one was with a2. The date returned vs. the modification date on a2 says this one was a1, but it is the last result I returned from the Mini before SETI went down and I was on a2 by then. Once it's back up, I'll have more a2 WUs and will be able to give a better answer, as I _know_ those will all be a2.

The speed difference makes this definitely worthwhile to try... but if _any_ significant percentage are invalid, then it makes the code 'suspect' for use, credits or not. So far though, more seem to be valid than not. Maybe off "just enough" so that timing of results comes into play? SZTAKI had this problem - Linux vs. Windows, whichever returned a result "first", that platform would always validate, the other would not. They had to change the validator criteria for "close enough" to make things work right on that project. I'm not sure how critical absolute "match" is for SETI.
This is somewhat worrisome for me, since I was originally worried about G5 alpha-1 failing validation rather than the other compiles. With that in mind, I wanted to get a feel for how you guys would define a significant percentage of failed validations. On one hand, we have two people who have work units that failed to validate, but on the other hand, it seems like the vast majority of work units are going in fine. Also, alpha-2 should yield identical results to alpha-1, and alpha-3 should only vary very slightly in chi-sqr calculations.

It would really be nice if the SETI results database could actually point out why a particular client failed validation on a work unit, so we could focus our attention on which routines have become problematic. This might be worth requesting on their forums.
     
Junior Member
Join Date: Oct 2004
Location: Germany
Status: Offline
Aug 30, 2005, 05:00 PM
 
Til now, all my results are valid (Alpha1), 15 results.

Waiting for new work, then i'm testing Alpha3.
     
Dedicated MacNNer
Join Date: Nov 2004
Location: Virginia
Status: Offline
Aug 31, 2005, 10:02 PM
 
I'm still trying to test out the newest version. My three computers have only been able to download a grand total of two WU in the last week; both of those have finished calculating, and I have not been able to get them to upload.

Oh, well, I'm continuing to crunch on Einstein with one computer - the old iMac can only run SETI, so it's stuck asking for work, and my iBook for some reason is not doing anything; I'm away on travel with my PowerBook, so the iBook will have to wait a couple of days...

beadman
     
Mac Elite
Join Date: Jan 2002
Location: Mile High City
Status: Offline
Aug 31, 2005, 10:48 PM
 
I'm up to 59 validated and still no errors.

The new G5 client appears to be much faster, looks like an estimated 1 hr, 20 min per unit. About 20-25 minutes faster overall.
( Last edited by Shaktai; Aug 31, 2005 at 11:31 PM. )
     
Moderator
Join Date: Sep 2001
Location: Arizona
Status: Offline
Sep 1, 2005, 12:06 AM
 
Okay, I've done this before but I can't seem to accomplish it now;

How do I modify the BOINC Menubar app with the optimized worker?
I like chicken
I like liver
Meow Mix, Meow Mix
Please de-liv-er
     
Mac Elite
Join Date: Jan 2002
Location: Mile High City
Status: Offline
Sep 1, 2005, 12:35 AM
 
Originally Posted by Lateralus
Okay, I've done this before but I can't seem to accomplish it now;

How do I modify the BOINC Menubar app with the optimized worker?
First remember that BOINC is a client that other programs work through. What you need to do is find your BOINC data folder. It should be located at Drive root level > Library > Application Support > BOINC data.

Stop BOINC
Open the BOINC data folder and find projects.
Open the Projects folder and find SETI@Home
Open SETI at Home and place the optimized SETI app and the app_info.xml in that folder. (your don't need to remove the old seti app)
Close everything up.

Restart BOINC. It should now recognize the new app and start crunching work for it.
     
Fresh-Faced Recruit
Join Date: Jun 2005
Location: Nampa, ID
Status: Offline
Sep 2, 2005, 11:02 PM
 
WOW! I have a 1.42GHz Mac Mini running 10.4.2 and just put on the 7450 worker. Since 4:00pm this afternoon, I have completed 2 work units! Times are VERY low now. Just an example:
With new worker:
25356325 = 7244.47 seconds
25356330 = 7279.78 seconds

With JavaLizard worker
25195925 = 19882.30 Seconds
25283354 = 17295.25 seconds
25253724 = 21112.02 seconds
25226993 = 19836.69 seconds

How did you get this much of a performance boost? Was it due to changes in Tiger that were not available in Panther? I have an iBook 1GHz that takes on average 27000 seconds to complete a work unit. I am still using the JavaLizard worker on it and it has Panther. If I use the 7450 worker on the iBook, will I see this huge reduction in computation time?
     
Mac Elite
Join Date: Jan 2002
Location: Mile High City
Status: Offline
Sep 3, 2005, 12:00 AM
 
Originally Posted by electrician1972
If I use the 7450 worker on the iBook, will I see this huge reduction in computation time?
Well, since I don't know if anyone has tested that combination yet, give it a try and let us know the results. Keep in mind, this is still alpha testing. They've done a great job, but there are no guarantees. So far we have seen substantial increases across the board from what I have seen.

Those are impressive times for the Mac Mini.
     
Fresh-Faced Recruit
Join Date: Jun 2005
Location: Nampa, ID
Status: Offline
Sep 3, 2005, 12:17 AM
 
Heh heh heh... yeah kind of surprising, but I have thought highly of the G4 all along and assumed it was capable of some cool stuff. BTW- just finished another work unit: 7272 seconds. Got the worker loaded up on the iBook and it immediately dumped the 3 work units it had previously downloaded when it was running the JavaLizard worker. Now it is attempting to contact the server for more work. I read somewhere in this forum that all you needed to do was stop the boinc process, copy over the worker and .info file into the /projects/setiathome folder, then restart the boinc client and it would continue. Is this not true? I guess it's water under the bridge now...
     
Fresh-Faced Recruit
Join Date: Jun 2005
Location: Nampa, ID
Status: Offline
Sep 3, 2005, 12:48 AM
 
Well, looks like the iBook doesn't like the 7450 worker. It has a 7447 chip, something like a low power version of the 7450. Here's some splat from the error log:
2005-09-02 22:23:48 [SETI@home] Unrecoverable error for result 02se03ab.26535.14560.1022148.9_1 (process got signal 5)
2005-09-02 22:23:48 [SETI@home] Deferring communication with project for 59 seconds
2005-09-02 22:23:50 [SETI@home] Unrecoverable error for result 16oc03ab.22246.6672.278414.109_0 (process got signal 5)
2005-09-02 22:23:50 [SETI@home] Deferring communication with project for 59 seconds
2005-09-02 22:34:59 [SETI@home] No work from project
2005-09-02 22:34:59 [SETI@home] Deferring communication with project for 59 seconds
2005-09-02 22:36:03 [SETI@home] No work from project
2005-09-02 22:36:03 [SETI@home] Deferring communication with project for 59 seconds
2005-09-02 22:37:05 [SETI@home] No work from project
2005-09-02 22:37:05 [SETI@home] Deferring communication with project for 10 minutes and 5 seconds

After eating alive three other work units, it devoured these two, downloaded two more and ate them too. Neato... Going back to JavaLizard worker.
     
Mac Elite
Join Date: Jan 2002
Location: Mile High City
Status: Offline
Sep 3, 2005, 01:03 AM
 
Try the 7400 version. Oh, and it will error out any units already in queue, because the app version is different. So clear the queue first and then install the alpha app.

The difference is the actual design of the altivec unit itself. The 7400 will also run on a G5, or you can run the G5 version. Again, clear the queue first of any units, and then install the optimized app.
     
Dedicated MacNNer
Join Date: Sep 2004
Location: London
Status: Offline
Sep 4, 2005, 03:36 AM
 
The results below are from Shaktia's posted online utility for users to check their pending credit and average work unit rate: http://ztb.dyndns.org/pc.php

using Version 3 sw

last wu received 3 Sep 2005 5:57:20 UTC
pending C 1,626.28
granted C 71,596.66
total C 73,222.94
estimated GC 3,048.72
estimated TC 73,222.94
# wus 160
Ø time/wu 1h 38m 55.867s (5,935.87s)
Ø CC/wu 10.16
Ø GC/wu 19.05


K.
     
Fresh-Faced Recruit
Join Date: Jun 2005
Location: Nampa, ID
Status: Offline
Sep 4, 2005, 10:52 AM
 
Well, I'm still waiting for my iBook to clear out before I change workers. Had some trouble with the Boinc client continuing to download units even though I issued the command -exit_when_idle. It's done this before and doesn't seem to be consistent. Sometimes it ignores the command to just process and exit. Oh well. 2 units to go (about 14 hours). But the numbers for the Mac Mini just keep getting better:
6081.57
6079.53
7151.74
6002.35
6173.78
7045.58
6207.00
Thanks to Alexkan and all involved. This worker is faster on my machine than most of the 3+ GHz pentiums out there. Unreal....
     
Dedicated MacNNer
Join Date: Nov 2004
Location: Virginia
Status: Offline
Sep 4, 2005, 11:06 AM
 
I finally got home from vacation and a business trip and got a chance to get the updated client installed on my iBook. Haven't had a chance yet for validations, but my hopes are high! I was averaging around 22,000 sec. with the 1.3 GHz iBook, OS X 10.3.9, and the first three results with the new client are
18730
11204
12120

Still not sure about my PowerBook, 1 GHz, OS X 10.3.9. It was averaging around 26,000 to 27,000 for each WU using JavaLizard's client, and the first five finished are all still pending.

21107
19829
356 (this one says "success done", and has no error codes...one other return of this WU is about the same time, though)
12265
12250

beadman
     
Fresh-Faced Recruit
Join Date: Sep 2005
Location: Germany
Status: Offline
Sep 4, 2005, 11:37 AM
 
hallo everyone!

is it possible to run this client with osx.2?
when not, is there any other boinc-seti-package for my dual-g4-867 with osx.2.8?

greetz!
     
Junior Member
Join Date: Jul 2005
Location: USA, Virginia
Status: Offline
Sep 4, 2005, 01:07 PM
 
I really hate to come off as the dumbest one here, but could someone please clarify which application is for which model of Mac. No-place on my Powerbook does it say what processor it has in it by chip number. All I can find is that it is a "1GHz powerPC G4." I am running OS 10.3.9. So it looks like there is only one app that is right. But if I upgrade to OS 10.4.x then I have to figure out which of the choices is right.

I also have a 1.4GHz Dual G4, and there is also no info on the system (that I can find) that gives a chip number. As embarrassing as it is to ask a question like this, it might help others (who aren't as willing to look stupid) figure this out as well.

Frankly, I have not really had any need to know the chip-set of the machine since the MacII CX debacle of the 1980's.

Regards
Phil
We must seek intelligent life on other planets as it is increasingly apparent we will not find any on our own.

Link: http://www.boincsynergy.com/images/stats/comb-2033.jpg
     
Junior Member
Join Date: Oct 2004
Location: Germany
Status: Offline
Sep 4, 2005, 01:26 PM
 
Can't see any Powerbook listed on your SETI profil.
Wich Powerbook do you have?

PowerBook3,5; PowerBook5,1 uses the MPC7455
PowerBook5,2; PowerBook6,2 uses the MPC7447

The PowerMac G4 Dual 1.42GHz uses the MPC7455

You have to use the "G4 7450/7455 Version"
     
Mac Elite
Join Date: Jan 2002
Location: Mile High City
Status: Offline
Sep 4, 2005, 02:00 PM
 
Originally Posted by ursmed
hallo everyone!

is it possible to run this client with osx.2?
when not, is there any other boinc-seti-package for my dual-g4-867 with osx.2.8?

greetz!
Greetings ursmed. BOINC (all versions and clients) will not run with OS 10.2.x It requires at least 10.3.x or higher. Several attempts were made to compile versions for 10.2.x by both the developers, and outside third parties, but none were very successful.

Sorry! Your only option is to crunch for a non-BOINC project or to upgrade. 10.4.x is worth the upgrade if you can do it.
     
Administrator
Join Date: Jun 2000
Location: California
Status: Offline
Sep 4, 2005, 02:18 PM
 
Originally Posted by Snake_doctor
... No-place on my Powerbook does it say what processor it has in it by chip number.
...
Open up Terminal. Type "machine" and hit Return. It will tell you the real CPU type.
     
Fresh-Faced Recruit
Join Date: Sep 2005
Location: Germany
Status: Offline
Sep 4, 2005, 02:25 PM
 
Originally Posted by Shaktai
Greetings ursmed. BOINC (all versions and clients) will not run with OS 10.2.x It requires at least 10.3.x or higher. Several attempts were made to compile versions for 10.2.x by both the developers, and outside third parties, but none were very successful.

Sorry! Your only option is to crunch for a non-BOINC project or to upgrade. 10.4.x is worth the upgrade if you can do it.
bad news
i cant believe that this is not possible with osx.2...
do you really upgrade to every osx-version?
for some reasons (nostalgics?, ) i dont want to upgrade this machine.
i have a osx.3.9 one and osx.4 comes only with a new one (hope this year)...

than i have to try a little bit more (i think 4.19 was for osx.2 to) or indeed crunch for seti-classic until the end and than maybe for dnetc with this machine (nice altivec-routines)...
or anybody will fix this and build a jaguar-release...???
     
Administrator
Join Date: Jun 2000
Location: California
Status: Offline
Sep 4, 2005, 02:42 PM
 
Folding@home works nicely with 10.2.8, and they have good altivec. D2OL may still work with Jaguar too, but I have not tested that in awhile.
     
Junior Member
Join Date: Jul 2005
Location: USA, Virginia
Status: Offline
Sep 4, 2005, 02:54 PM
 
Thanks guys
Regards
Phil
We must seek intelligent life on other planets as it is increasingly apparent we will not find any on our own.

Link: http://www.boincsynergy.com/images/stats/comb-2033.jpg
     
Mac Elite
Join Date: Jan 2002
Location: Mile High City
Status: Offline
Sep 4, 2005, 03:12 PM
 
Originally Posted by ursmed
...or anybody will fix this and build a jaguar-release...???
Apparently to get it to work with Jaguar, would require changes to the source code. That means that they would have to write source code just for Jaguar. It really isn't time or cost effective to do that. All the efforts at compiling current source code (including 4.19) have failed. The only successes were a few individuals that compiled it themselves for their specific machine. The compiles wouldn't move over to other Macs cleanly though. Even then, the success was questionable, because there was a very high rate of errors or invalid results.

As reader50 mentioned, Folding@Home is an excellent project for 10.2.x and does some very beneficial scientific work. Some work units there (gromacs) are also altivec enhanced. And you are correct dnetc also provides some excellent performance. Team MacNN has a team for all those projects. You've probably got a couple of more months of SETI Classic though.
     
alexkan  (op)
Forum Regular
Join Date: Aug 2005
Location: Cupertino, CA
Status: Offline
Sep 4, 2005, 07:54 PM
 
electrician1972:
Did you make sure that you're running the version of the worker that's appropriate to your version of OS X? I believe beadman had the same problem with signal 5 errors when the first compiles came out. This is because each compile is linked against system libraries specific to those versions. (This is because libraries like vDSP get faster/better as newer versions of the OS come out, and we want to take full advantage of that.)

It should be pointed out that since the 10.3.9 compile is still based off of alpha-1, it doesn't have a lot of the optimizations that went into alpha-2 and alpha-3--those of you who have tried all the alphas probably saw how much faster the newest alpha is than the first one that we released.

Also, to clarify with regard to compiles for different versions: G4s will run both G4 compiles (but not the G5 compile), and the G5 can run any of the 3 compiles, but the reason there are 3 different compiles is that the different processor architectures behave differently with regard to instruction scheduling, and this could make a difference in terms of performance, but not in terms of the final calculated result.

It's also interesting to me to see that work unit completion time is no longer scaling with clock speed on G4s, but rather mostly by bus speed. I had hoped for much lower times on faster G4s, since a PowerBook at 800 MHz (100 MHz bus) already does the reference work unit in ~9000 seconds, but it seems that we'll have to find some way to make better use of what we have while it's still in fast memory. (It's possible that these optimizations might help SETI@home enhanced when it's done, since it's supposed to do 10x more work per work unit.) It would be interesting, then, to compare the performance of iMac G5s with PowerMac G5s, all other factors being as equal as possible, since iMacs have lower bus speeds for a given processor speed.
     
Fresh-Faced Recruit
Join Date: Jun 2005
Location: Nampa, ID
Status: Offline
Sep 4, 2005, 10:49 PM
 
Originally Posted by alexkan
electrician1972:
Did you make sure that you're running the version of the worker that's appropriate to your version of OS X? I believe beadman had the same problem with signal 5 errors when the first compiles came out. This is because each compile is linked against system libraries specific to those versions. (This is because libraries like vDSP get faster/better as newer versions of the OS come out, and we want to take full advantage of that.)

It should be pointed out that since the 10.3.9 compile is still based off of alpha-1, it doesn't have a lot of the optimizations that went into alpha-2 and alpha-3--those of you who have tried all the alphas probably saw how much faster the newest alpha is than the first one that we released.

Also, to clarify with regard to compiles for different versions: G4s will run both G4 compiles (but not the G5 compile), and the G5 can run any of the 3 compiles, but the reason there are 3 different compiles is that the different processor architectures behave differently with regard to instruction scheduling, and this could make a difference in terms of performance, but not in terms of the final calculated result.

It's also interesting to me to see that work unit completion time is no longer scaling with clock speed on G4s, but rather mostly by bus speed. I had hoped for much lower times on faster G4s, since a PowerBook at 800 MHz (100 MHz bus) already does the reference work unit in ~9000 seconds, but it seems that we'll have to find some way to make better use of what we have while it's still in fast memory. (It's possible that these optimizations might help SETI@home enhanced when it's done, since it's supposed to do 10x more work per work unit.) It would be interesting, then, to compare the performance of iMac G5s with PowerMac G5s, all other factors being as equal as possible, since iMacs have lower bus speeds for a given processor speed.
Oops. Yeah I got a mac mini with tiger and like a dip-stick tried to load the same thing worker on my Panther iBook. Currently about 1/2 hour away from clearing the current work units so I can load the Panther worker. Right now the iBook averages a work unit every 26000 seconds or so. Will get the worker loaded a post times soon. Thanks again for a great job on making the Mac workers so fast...
     
Ola A.
Guest
Status:
Sep 6, 2005, 12:34 PM
 
[Just another stupid question from a beginner]

I downloaded a few optimized seti-programs, how do I run them within the boinc system?

Just tried an evil way by replacing the original ...apple-darwin... file with the optimized binary, (changing name also), but it only reset the whole boinc project.

Ola
     
Fresh-Faced Recruit
Join Date: Jun 2005
Location: Nampa, ID
Status: Offline
Sep 7, 2005, 12:09 AM
 
Originally Posted by Ola A.
[Just another stupid question from a beginner]

I downloaded a few optimized seti-programs, how do I run them within the boinc system?

Just tried an evil way by replacing the original ...apple-darwin... file with the optimized binary, (changing name also), but it only reset the whole boinc project.

Ola
No stupid questions here! Well, it sounds like you have the basics setup. Just make sure you have the optimized seti worker for your system in your /boinc/projects/setiathome.berkeley.edu folder. Startup the boinc client and let it download fresh work units. Sounds like you've already lost the old work units. However, in the future, to avoid such a loss, just stop your boinc client, then restart it with the following:
./{your boinc client name} -exit_when_idle >>boinc.log 2>>error.log & You don't have to redirect output to the log files if you don't want to. In my experience, this command for some reason does not always refrain from downloading new work units, so you may have to stop it and start it again. Hope this helps
     
Fresh-Faced Recruit
Join Date: Jun 2005
Location: Nampa, ID
Status: Offline
Sep 7, 2005, 12:20 AM
 
Finally got the 10.3.9 optimized worker installed on my iBook (1GHz, 768MB ram, High performance) and have processed several units with it. Here is a comparison of the times from the Javalizard worker and the new worker:
Javalizard shortest time: 24980.25 sec.
Javalizard longest time: 27941.18 sec.
Total of 13 results under this worker

New alpha worker:
Shortest time: 358.47 sec.
Longest time: 22518.37 sec.
Total of 10 results turned in so far.

The computer ID is 1341457 if you want to see detailed info.

I am waiting for a new copy of Tiger from the Mother Ship to arrive so I can load it on the iBook - I'll bet that speeds it up a lot with those improved vDSP libs!
     
Ola A.
Guest
Status:
Sep 8, 2005, 03:23 AM
 
Originally Posted by electrician1972
No stupid questions here! Well, it sounds like you have the basics setup. Just make sure you have the optimized seti worker for your system in your /boinc/projects/setiathome.berkeley.edu folder. Startup the boinc client and let it download fresh work units. Sounds like you've already lost the old work units. However, in the future, to avoid such a loss, just stop your boinc client, then restart it with the following:
./{your boinc client name} -exit_when_idle >>boinc.log 2>>error.log & You don't have to redirect output to the log files if you don't want to. In my experience, this command for some reason does not always refrain from downloading new work units, so you may have to stop it and start it again. Hope this helps
Hm. I put the optimized application (boinc_4.44_ppcG4) in the folder ...Projects/Setiathome.berkeley.edu. And did Project reset (or whatever it is called in english). After many hours the boinc manager downloaded some new workunits, but the name (in boinc manager/work) is still setiathome 4.18.
I deleted the old app in the above folder, but maybe there are some more copies in the Slots folder etc.?

What more should I do? Or is it the correct app that is running, only with wrong name?


Ola A.
     
Senior User
Join Date: Feb 2002
Location: Silly Valley, Ca
Status: Offline
Sep 8, 2005, 03:21 PM
 
Originally Posted by Ola A.
Hm. I put the optimized application (boinc_4.44_ppcG4) in the folder ...Projects/Setiathome.berkeley.edu. And did Project reset (or whatever it is called in english). After many hours the boinc manager downloaded some new workunits, but the name (in boinc manager/work) is still setiathome 4.18.
I deleted the old app in the above folder, but maybe there are some more copies in the Slots folder etc.?

What more should I do? Or is it the correct app that is running, only with wrong name?
Ola A.
Are you running the CLUI or a GUI client?
All the info on using the optimized workers and clients is in this thread and several others
Here is an overview with the GUI:
http://forums.macnn.com/showthread.p...48#post2679744

I think we had better make a full write-up and make it a sticky.
     
Junior Member
Join Date: Oct 2004
Location: Germany
Status: Offline
Sep 11, 2005, 12:12 PM
 
So far with alpha3 version:

34 results, all valid
One "complete" wu takes about 12.000 seconds.

Alpha4, bring it on
     
Fresh-Faced Recruit
Join Date: Jun 2005
Location: Nampa, ID
Status: Offline
Sep 11, 2005, 03:36 PM
 
Well here's the scoop on the iBook G4 1GHz: After fighting with Fedex ground dip sticks to get the Tiger package delivered when I was home, I got the thing installed and setup the 7450 Alpha-3 worker. Then impatiently waited for the project to come back up. Hurry up and wait. Currently the system has finished 2 work units:
9415.97 seconds
9281.28 seconds
Amazing that an OS upgrade and optimized worker will take you from 27000+/- seconds to 9000+/- on the same clock speed. I guess there's something to be said for upgrading after all.
     
Fresh-Faced Recruit
Join Date: Aug 2001
Location: UK
Status: Offline
Sep 11, 2005, 07:04 PM
 
Same here 30+ units, all valid, mostly alpha3, taking c. 9000s whereas stock is nearer 27,000s (1GHz eMac 10.4.2, 7450 client). Compares quite well with my AMD 3100+ which takes c. 9000s stock or 4500-5500s with optimised XP client.

Thanks alexkan, may your optimisations get ever quicker
( Last edited by Drash; Sep 11, 2005 at 08:23 PM. )
     
Fresh-Faced Recruit
Join Date: Sep 2005
Status: Offline
Sep 12, 2005, 06:29 AM
 
Originally Posted by mikkyo
Are you running the CLUI or a GUI client?
All the info on using the optimized workers and clients is in this thread and several others
Here is an overview with the GUI:
http://forums.macnn.com/showthread.p...48#post2679744

I think we had better make a full write-up and make it a sticky.
Thanks Mikkyo! Tried to follow the instructions, using the suggested versions.
But first I think I need some basic info about the relations between boinc-manager, boinc-client and seti-worker(?).

My thought was that boinc manager was a system for managing new workunits, different projects etc. and the you added one "client" for each project like seti, folding etc.
But most optimisation seems to be done on the boinc, so clearly I'm more or less wrong.

The message list in boinc-manager (Graphical version) now reported "detected version change 4.43 -> 4.44, recalc cpu benchmark". I don't think the cpu-benchmark improved, rather it is a litter slower, now 710 Wet/ 1907 Dry, on a 1Ghz pb w 256Mb and 10.3.9.

How do I check if speed improves? Time for each wu seems do seems to differ a lot.

Ola A. from Uppsala, Sweden
     
Dedicated MacNNer
Join Date: Nov 2004
Location: Virginia
Status: Offline
Sep 12, 2005, 03:50 PM
 
Let me give it a try, Ola.

BOINC Manager is the application on your Mac that is graphical and lets you see what's going on. BOINC is the client inside of BOINC Manager that actually manages the pausing and resuming of each science client. SETI, Einstein, Climate, etc., are the Science clients that BOINC manages.

Quit BOINC Manager. If you Control-Click on BOINC Manager in your Applications folder, you'll see a pop-up window appear. Select "Show Package Contents" and a finder window will open that shows you a folder. When you open that folder, you'll see all the different files bundled together inside BOINC Manager. One of those files is "boinc" - this is the actual 4.43 BOINC file that would be replaced if you want to use the optimized 4.44 BOINC client. Just drag the old one to the trash, and replace with the 4.44 BOINC file, and then make sure you rename it "boinc" exactly like the previous one was named. This should not affect any work units in your queue...

One thing to check for improving your benchmarks: Open you Systems Preferences, select Energy Saver, then click on Options. Near the bottom of the window will be something called "Processor Performance". The default is Automatic, which allows the processor to slow down. If you set this to "Highest", you'll get maximum performance from your powerbook. CAUTION: The bottom of the powerbook can get hot! I elevate mine powerbook on a couple of small sticks to allow air circulation underneath it; some others have used wire baking racks.

Your Powerbook should get better benchmarks, I think. My old 1GHZ powerbook had around 1000/4000 for it's benchmarks. It had 1 GB RAM, so I don't know how much difference that makes.

beadman
( Last edited by beadman; Sep 12, 2005 at 03:53 PM. Reason: left out info.)
     
Senior User
Join Date: Feb 2002
Location: Silly Valley, Ca
Status: Offline
Sep 12, 2005, 04:57 PM
 
You can find a Boinc Manager or Menubar GUI client with the 4.44 optimized client pre-installed here:
http://members.dslextreme.com/~reade...eam/boinc.html

Then all you need do to use the seti worker from this thread is grab the one for you machine and install it in the projects/setiathome.berkeley.edu folder with the app_info.xml file that comes with it.
     
Bill Michael
Guest
Status:
Sep 13, 2005, 01:19 AM
 
I have two more results from alpha-2 that were validated; no idea what the problem was before. Also now have 10 results validated from alpha-3 and a bunch still waiting for a quorum - no problems. Time on the Mac Mini (G4 1.25GHz) is running right around 7500 seconds. LOVE IT!!!

Now... if only a version could be put together that detects the CPU type, so there was only one download... and handed to UCB to become the standard science application... sigh.
     
Senior User
Join Date: Feb 2002
Location: Silly Valley, Ca
Status: Offline
Sep 13, 2005, 06:06 AM
 
Originally Posted by Bill Michael
Now... if only a version could be put together that detects the CPU type, so there was only one download.
This can be done by alexkan using the -arch flag with a compile for each cpu type and then using lipo to assemble them into one binary like I did with the boinc 4.44 ppcAll client.
     
Carl
Guest
Status:
Sep 13, 2005, 10:26 AM
 
Just wanted to add that I have 29 WU done with the new version and all but 6 have been verified, the others are still pending.

It cut my cpu time from 4-5 hours down to 1.5-2.25 hours. Incredible.

BTW-This is a 1.33 G4 Powerbook, 10.4.2, 1.5GB RAM.

Keep up the good work. Here is a link to my results, the last one I used the Altivec optimized was WU # 25818454. The first one I used the new optimized found on page 1 of this post was 26070600.

http://setiathome.berkeley.edu/resul...?userid=539623

Carl
     
Coroner .no
Guest
Status:
Sep 13, 2005, 12:02 PM
 
Just wanted to say that the alpha3 worker is working like a dream on my PowerMac G4 500Mhz (Sawtooth). I've crunched 19 WU's now and they have all validated properly. Since I am new to optimized workers, the only one I had tried previously was the alpha2 which didn't work, aborting every WU it downloaded.

Compared to the standard worker the crunch times have been cut from 40.000 - 50.000 seconds, to only 10.00 - 12.000 seconds with the alpha3.

Great work on the optimization :-)

Results: http://setiathome.berkeley.edu/resul...userid=8116179


Greetings from Norway
-=Coroner=-
     
 
Thread Tools
Forum Links
Forum Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts
BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are On
Pingbacks are On
Refbacks are On
Top
Privacy Policy
All times are GMT -4. The time now is 01:32 AM.
All contents of these forums © 1995-2014 MacNN. All rights reserved.
Branding + Design: www.gesamtbild.com
vBulletin v.3.8.8 © 2000-2014, Jelsoft Enterprises Ltd., Content Relevant URLs by vBSEO 3.3.2