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 17)
Thread Tools
lepetitmartien
Junior Member
Join Date: Feb 2006
Location: Paris, France, Europe, Earth, Sol
Status: Offline
Mar 13, 2006, 12:58 AM
 
Originally Posted by TiloProbst
mk1 = ?
it's the september 2003 model. Mk2 is the buggy bag one.
uhm, I have 900 Mhz FSB, maybe that explains the performance differences?

I once achieved ~440 RAC with my Single 1,8 .. Can't give a more precise prediction since this machine runs maybe 14 hours a day, often not even that.
I have the same (PCI-x, 900MHz FSB), right now I'm at 340 and it's still going up, I'll tell when it's leveling, as it runs 24/24 7/7 I may go even higher. I run einstein and predictor too but they have only 5% each of the available time, mainly in case of [email protected] outages as the MacMusic team is SETI only.

The great thing is I jumped today as the 2nd cruncher/day of my team today (was third for a loooong time)
MacMusic.Org says "Hi all!" :)
G5 desktop 1.8, 900 MHz frontbus (2003 model)
Latest wisdom file for it on demand, just PM me :)
     
phantomac
Forum Regular
Join Date: Oct 2003
Location: Germany
Status: Offline
Mar 13, 2006, 03:05 PM
 
Todd: I still wonder why my Dual 2.7 is slower than your Dual 2.5:

http://setiathome.berkeley.edu/show_...hostid=2206613

(I use the wisdom file provided by E.T.)
     
Todd Madson
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status: Offline
Mar 14, 2006, 08:01 PM
 
Weird.

You should be going faster than my machine - looking at the stats,
they seem comparable although you seem to be getting more blocks
in the 2500 seconds range and mine seem to be more 1900-2300.

-You've got more ram than I do (I have only 2.5 GB).

-My ram is listed as PC3200U-30330 in the system profiler.
Not the fastest ram but not the slowest either - but maybe one thing that
is helping - two of the sets were matched pairs according to the manufacturer.

-Your ram may have different timings. Remember the G5 will dictate the system
timing based on the slowest pair even if you have a set that is faster.

-In Chud Tools I run with nap disabled.

-In Energy Saver under the Options tab I have CPU performance set to highest.

-I actually created my own BigFFT Wisdom file from Alex's application, it took
about two hours. I'm not sure if making your own would give you an edge or
not - it might be worth a try.

-When I am not at my machine I run no other applications but Seti.

-Version of Boinc Manager I am running is 5.2.13 that I got from the Team Macnn page.

-I am running the seti stuff on my boot disc (Sata 150 gig, nothing special) if that helps.

-My DP 2.5 has the stock Radeon 9600XT card, I only attempt overclocking the card
if I'm running X-Plane or something.

-Some other thoughts - I recently put it so that my largest ram chips were in the
first slot in the system. The 256 meg chips that came with the machine were
originally in the machine. Don't know if that helps or but but that's the way it
is now.

-I recently set it up to cache several days of blocks to crunch because I kept
running into situations where we would have an outage for two to five days
and my machine was sitting idle most of that time. Try setting it up to
cache 4-5 days of production and then look at your stats.

Some ideas: anyone know if a ramdisk exists for OS X? Might it be possible
to run Seti out of a ramdisk rather than slow actual disk?
     
brysonda
Fresh-Faced Recruit
Join Date: Jan 2006
Status: Offline
Mar 15, 2006, 12:04 PM
 
Ok, so I have some real data about the effects of disabling nap on a Quad G5. I took a sample of runtime data from two time periods, one several days before disabling nap, one several days after.

Using a sample of 20 run times (one results page, all points inclusive), the results indicate a 4.18% improvement. However, in this case, the confidence intervals are on top of each other making the results not significant.

From the set of 20, I selected 10 of the 'normal' values (excluding the short outliers at 1700 and the long 2200 ones). In this case the advantage of disabling nap shrinks to 2.13%, but there is a clear separation of 19.5 seconds between the confidence intervals.

Some notes:
- Quad G5 - Purchased January 2006
- 2.5 GB RAM, Stock 250 GB Harddrive, 10.4.5
- Processor Peformance set to Highest
- BOINC MenuBar, MacNN Superbench version
- a52, using ET's bigfft_wisdom file
- The G5 was not rebooted at any point during the test

Thoughts:
Nap does seem to make a difference, albeit not earth shattering. It would be interesting to see what how much of a difference there is in terms of power consumption. Does it also only increase 2-4%, or is it more detrimental? Unfortunately I don't have a meter for measuring this... though I wonder if the built in sensors could be used to take some data? Hmm...
     
Todd Madson
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status: Offline
Mar 15, 2006, 05:17 PM
 
Brysonda:
We used a meter called the Power Angel on my machine where there was a
100 watt jump in power consumption when Nap was disabled.

G5 2.5 DP:
With Nap Enabled: 160 watts idle
With Nap Disabled: 260 watts idle

Not sure if it's worth it but I'm in a race for my life now (a competitor is
literally buying machines to bury me) so every little bit helps.

I believe it is cumulative - every little bit helps.
     
Mark Asiala
Fresh-Faced Recruit
Join Date: Jan 2006
Location: Maryland
Status: Offline
Mar 16, 2006, 11:53 AM
 
Originally Posted by Todd Madson
G5 2.5 DP:
With Nap Enabled: 160 watts idle
With Nap Disabled: 260 watts idle
There is the key, "idle". What I would be curious about is what is the power draw while running [email protected] on both processors with Nap enabled and with Nap disabled. I would bet money that the difference is a lot less than 100 watts and maybe only 5-10 Watts (I'm *not* putting money on the exact figure!).

The 2-4% gain may be about right. To really test it I would set up a loop to run the reference work unit 20 times per processor under each scenario after a fresh reboot and doing nothing else on the machine. Then again, I would subsequently spend a month trying to regain my lost productivity during the testing period.

Mark
     
Todd Madson
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status: Offline
Mar 16, 2006, 05:31 PM
 
I'll see if I can borrow my buddies power angel again (I've been meaning to get one myself).

It's sort of funny to see how much power things consume.
     
Todd Madson
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status: Offline
Mar 16, 2006, 06:18 PM
 
Ideas for better performance:

I just had a thought:
-has anyone run this in non-gui mode? i.e. no OS X gui just text mode?
-if so, has anyone verified performance increases if any?
-I will test this weekend if I can.

I saw something interesting in the top 120 computers list today:
-Machines 1-4 on top computers list G5 quads.
-Machines 6-7 on top computers list G5 quads.
-Machines 8-116 are Dual Opterons, Dual Pentium 4s, and Dual Athlon 64s.
-Machine 117 is a DC 2.3 (E.T. from Tellus)
-Machine 120 is my DP 2.5

I think there's a way to improve.

So far it appears my dual 2.5 is doing better than any other dual
2.5 in the world participating in the project.

Indeed, it appears to be doing better than the dual 2.7s - in fact,
I see NO dual 2.7s on the top 120. I ponder why this is?
     
Knightrider
Dedicated MacNNer
Join Date: Sep 2004
Location: London
Status: Offline
Mar 17, 2006, 05:39 AM
 
[email protected] Enhanced.

This new application is currently in beta-test. The BOINC core client will automatically update [email protected] to the new enhanced version once it is made available. Since there is more computing involved, users will notice a greatly increased turnaround time per work unit. Since BOINC credits by computation and not by workunit, this will not change the rate at which you receive credit. We will also extend the return deadlines since clients will be crunching on workunits for longer periods of time.
This is interesting reading. Will it make Alex and Ricks version redundant?

http://setiathome.berkeley.edu/sah_plans.php

K,
     
phantomac
Forum Regular
Join Date: Oct 2003
Location: Germany
Status: Offline
Mar 17, 2006, 07:32 AM
 
Originally Posted by Todd Madson
Ideas for better performance:

I just had a thought:
-has anyone run this in non-gui mode? i.e. no OS X gui just text mode?
-if so, has anyone verified performance increases if any?
-I will test this weekend if I can.
I'm running the graphical client on the Mac and I don't expect the non-gui mode to gain much performance. The GUI client is set to the project tab and hidden. That way the other tabs don't get redrawn all the time and don't consume CPU time.

-Machine 120 is my DP 2.5

Indeed, it appears to be doing better than the dual 2.7s - in fact,
I see NO dual 2.7s on the top 120. I ponder why this is?
Your machine was the one I had in mind to overtake. I have no idea what's wrong with the Dual 2.7. Next week I'll try to create a new wisdom file for it and see if it improves anything.

I won't disable nap as I think there's another problem out there keeping my machine from outperforming yours.
     
Todd Madson
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status: Offline
Mar 17, 2006, 10:55 AM
 
Hmmm. I have been hiding the graphical client but not viewing the project tab
but rather the work tab. I'll try the project tab to avoid redraws like that.

Thanks for the tip.

As far as Knightriders post - I've heard the next version of the project will be
one where work units take forever to process again. Maybe that version can
be optimized so it takes less time?
     
Mark Asiala
Fresh-Faced Recruit
Join Date: Jan 2006
Location: Maryland
Status: Offline
Mar 17, 2006, 06:36 PM
 
Originally Posted by Todd Madson
Hmmm. I have been hiding the graphical client but not viewing the project tab
but rather the work tab. I'll try the project tab to avoid redraws like that.
My method is to run the non-gui client and just periodically check in on using BOINCManager, then quit the app after I am done. I do a 'chmod 000 boinc' on the boinc which is embedded in the gui client.

What I found before with a simple view of top is that if you click the close box on BOINCManager so that it auto-hides to the background the usage goes to 0.0% but if I simply choose "Hide BOINCManager" from the BOINCManager application menu, it will continue to eat up a fraction of 1% or so.

Personally, I like my set up with a non-gui client running on all of my machines and just running BOINCManager when I have something to manage. I was the highest rated DP 2.5 on [email protected] with this set up and had been in the Top 10-14 RAC until I started splitting time with [email protected] and also because of the higher number of Quads coming in.

Mark
     
Mark Asiala
Fresh-Faced Recruit
Join Date: Jan 2006
Location: Maryland
Status: Offline
Mar 17, 2006, 06:51 PM
 
Originally Posted by Knightrider
[email protected] Enhanced.
This is interesting reading. Will it make Alex and Ricks version redundant?
http://setiathome.berkeley.edu/sah_plans.php
K,
I've seen postings on other forums that they think the stock enhanced clients may be more optimized than the current [email protected] clients out of the box. That said, I've also seen that crunch3r is working on an optimized beta client to be ready around the time that [email protected] enhanced goes live. I would think that the optimization need will continue as it has now but people are expecting perhaps less of a % gain than is currently realized.
     
rick
Fresh-Faced Recruit
Join Date: Sep 2005
Status: Offline
Mar 18, 2006, 06:56 AM
 
Originally Posted by Mark Asialia
I've seen postings on other forums that they think the stock enhanced clients may be more optimized than the current [email protected] clients out of the box. That said, I've also seen that crunch3r is working on an optimized beta client to be ready around the time that [email protected] enhanced goes live. I would think that the optimization need will continue as it has now but people are expecting perhaps less of a % gain than is currently realized.
When I had a look at the [email protected] enhanced source code back in December, the main changes were that they'd finally switched to use FFTW by default instead of the slower "Ooura" routines that they were using before.

There are various other small things that have changed, but in general the code is pretty similar so most (probably all) of our optimisations should port across just fine.

To be honest, at the moment I'm more frustrated that our "solution" isn't exactly user friendly. At the moment, someone wanting to use our optimised client has to f**k about with various installation files in random directories. Not cool.

If we want world domination (why settle for anything less?) we need to make it much easier for people to install. I'm currently thinking of ways to bundle this into a single installation package (i.e. fire and forget). I'll look into this once the x86 version is out the door.
     
nikoniko
Fresh-Faced Recruit
Join Date: Mar 2006
Location: Makedonia, Hellas
Status: Offline
Mar 18, 2006, 07:36 AM
 
I downloaded and installed the [email protected] at
/Library/Application Support/BOINC Data/projects/setiathome.berkeley.edu/
as well as the app_info.xml and one of the bigfft_wisdom for a 5 Quad that I found on this thread. I have noticed that the WU are processed about 2-3x times faster but all the results have the following error

<core_client_version>5.2.13</core_client_version>
<stderr_txt>
OS X Altivec-optimized vDSP/FFTW [email protected] application by Rick Berry and Alex Kan
version: alpha-5.2 (public release)
failed to load FFTW wisdom from bigfft_wisdom
using FFTW_ESTIMATE plan generation
FFT usage: vDSP out-of-place for FFT sizes < 2^13
FFTW out-of-place for FFT sizes >= 2^13

</stderr_txt>

You can check the results for the host
http://setiathome.berkeley.edu/resul...hostid=2248990

I have also installed the opitimized Boinc Manager (not superbench)

1) What do I have to do to fix this.

2) I also read that someone could generate the bigfft_wisdom file but I couldn't find any reference to the application that generates the wisdom file. Is there any link to download the specific application;

3) I have noticed that some people are gettting much higher values from benchmark test than mines what do I have to do to get similar results. Is it because of the above error;
Measured floating point speed 2210.07 million ops/sec
Measured integer speed 5196.43 million ops/sec

I have a Quad with 2GBRam and 7800 as Videocard
     
Todd Madson
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status: Offline
Mar 18, 2006, 10:57 AM
 
My machine is machine 108 today. Unreal. We'll see how long that can continue.

Thanks Mark for your suggestion, I'll give that a shot although if I quit out of
Boinc Manager altogether my seti clients stop crunching.

Ideas on how to only bring up the manager if needed?
     
nikoniko
Fresh-Faced Recruit
Join Date: Mar 2006
Location: Makedonia, Hellas
Status: Offline
Mar 19, 2006, 04:24 PM
 
ok I have solved 3) by replacing boinc manager with superbench manager
1 and 2 are still open if someone has a solution to offfer
     
Mark Asiala
Fresh-Faced Recruit
Join Date: Jan 2006
Location: Maryland
Status: Offline
Mar 19, 2006, 10:04 PM
 
Originally Posted by Todd Madson
Thanks Mark for your suggestion, I'll give that a shot although if I quit out of
Boinc Manager altogether my seti clients stop crunching.

Ideas on how to only bring up the manager if needed?
I set up the the superbench, gui-compatible boinc CLI client as a startup service so that boinc starts up when the computer starts up, regardless if someone logs in or not. It also works well because I like to logout completely when I'm not using my computer to free up as much as I can to run boinc and whatever workers (seti or einstein). Good instructions on how to do this can be found at http://stegic.net/index.php?p=2.

I had some trouble starting up BOINCManager without it trying to startup another copy of boinc and then complaining. To prevent this, I simply changed the permissions of the boinc client in "./BOINCManager.app/Contents/Resources/boinc" to 000 (i.e, no read, write or execute permissions). Technically, however, you should be fine with a "chmod a-x".

I also set up a gui password in "/Library/Application Support/BOINC Data/gui_rpc_auth.cfg" but found that BOINCManager always prompted me for the password unless I put a copy of the file in "~/Library/Application Support/BOINC Data/gui_rpc_auth.cfg". That way when I startup, it tries to connect to localhost and grabs the password automatically.

While this may sound like a lot, its pretty easy once you know what you're doing. This also works very well for one computer that does nothing but crunch seti and also for the kids' computer where I do not want them to have access to being able to turn my [email protected] crunching off by mistake or on purpose.

Mark
     
thunderaudio
Fresh-Faced Recruit
Join Date: Aug 2004
Status: Offline
Mar 20, 2006, 12:31 AM
 
I just downloaded the BOINC Manager 5.2.13 superbench clients from the beta page and I set it up according to the instructions in the forum. But I noticed when in the work tab it says Application setiathome 4.18. Is this correct? The work units say

http://setiathome.berkeley.edu/resul...ltid=266464064

<core_client_version>5.2.13</core_client_version>
<stderr_txt>
OS X Altivec-optimized vDSP/FFTW [email protected] application by Rick Berry and Alex Kan
version: alpha-5.2 (public release)
failed to load FFTW wisdom from bigfft_wisdom
using FFTW_ESTIMATE plan generation
FFT usage: vDSP out-of-place for FFT sizes < 2^13
FFTW out-of-place for FFT sizes >= 2^13

</stderr_txt>


Hmmm... What Am I doing wrong? I removed the setiathome_4.18_powerpc-apple-darwin file and the
app_info.xml
bigfft_wisdom.txt
[email protected]
are in the setiathome.berkeley.edufolder

I have 4 more G5 2.7's that could be running but I want to make sure I have this set up correctly

James.
     
lepetitmartien
Junior Member
Join Date: Feb 2006
Location: Paris, France, Europe, Earth, Sol
Status: Offline
Mar 20, 2006, 01:15 AM
 
Don't bother about the 4.18, it's normal… It's explained somewhere in the thread… somewhere…

The only thing wrong you have (and it's not your fault) is the .txt at the end of the wisdom file, it should not have any extension. Remove it and you're fine (restart BOINC of course).
MacMusic.Org says "Hi all!" :)
G5 desktop 1.8, 900 MHz frontbus (2003 model)
Latest wisdom file for it on demand, just PM me :)
     
thunderaudio
Fresh-Faced Recruit
Join Date: Aug 2004
Status: Offline
Mar 21, 2006, 12:50 AM
 
Originally Posted by lepetitmartien
Don't bother about the 4.18, it's normal… It's explained somewhere in the thread… somewhere…

The only thing wrong you have (and it's not your fault) is the .txt at the end of the wisdom file, it should not have any extension. Remove it and you're fine (restart BOINC of course).
Ah that fixed it. My 2.7 is much better now. 2200 seconds per WU. Thanks.

http://setiathome.berkeley.edu/resul...ltid=267667513

But how does ET's quad get 1700 second WU's? Mine gets only 2900

http://setiathome.berkeley.edu/resul...ltid=267698408

James.
     
Bob Palmer
Guest
Status:
Mar 21, 2006, 04:12 AM
 
I run in CLI mode, and last I checked my Quad is #3. I have several apps (Mail, Safari, Calendar, Addres Book, & Terminal) up and running all the time, but when I leave the machine, Finder is the active app.
     
Todd Madson
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status: Offline
Mar 21, 2006, 04:04 PM
 
ET gets 1700 seconds per WU because he is using a quad which is part
of the dual core family.

The dual core processors have a larger cache and essentially scream
doing Seti with the optimized client.

The original DP machines (1.8, 2.0, 2.3, 2.5, 2.7) do well although not
as well as the DCs (2.0, 2.3, 2.7) because of the larger cache which
helps seti. The faster memory can't hurt either.

My DP 2.5 does pretty well:
http://setiathome.berkeley.edu/resul...496&offset=440
(You'll see 1900-2000 second work units periodically)...

But ET's 2.3 does better even though the CPUs are technically "slower".
     
brysonda
Fresh-Faced Recruit
Join Date: Jan 2006
Status: Offline
Mar 23, 2006, 08:37 PM
 
Is the source code for a5 available? I have a couple of G4 systems running 10.3.9 Server that I'd like to get running a5, but it seems to be 10.4 only.
     
lepetitmartien
Junior Member
Join Date: Feb 2006
Location: Paris, France, Europe, Earth, Sol
Status: Offline
Mar 23, 2006, 11:23 PM
 
A little update on my G5 mono 1.8 (the evil 900 MHz front bus one)

Max RAC to date: 494
Best WU computing time 2138", the medium WU is in the 2400-2500.

It's not the maximum as 10% are devoted to other projects, and also, I use the computer. I've had some slowdown in the RAC the last days (down to 460, some units failed) and I will use the chance I've got no more units on the other project to put [email protected] as 100%.

Let us see how far can it go (well can it do the 550?



[edit]And all this the day [email protected] has a problem uploading units and doesn't validate… %)
( Last edited by lepetitmartien; Mar 24, 2006 at 01:07 PM. )
MacMusic.Org says "Hi all!" :)
G5 desktop 1.8, 900 MHz frontbus (2003 model)
Latest wisdom file for it on demand, just PM me :)
     
Todd Madson
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status: Offline
Mar 28, 2006, 03:36 PM
 
I attempted running alpha 5.2 with the Trux calibrating client that has done
so well on the PC platforms (I can vouch for this on the Pentium M laptop
and x86 Linux machine I have on my farm) but when it came to the G5 it
was not doing so well. My friend added this on all of his PCs and he really
trounced me good - if you've got a PC this client will improve your RAC in
an exponential manner.

The person who ported this to the Macintosh indicated the following when
it came to the alpha 5.2 "I just looked up your G5, I see why your getting
a performance "hit", your machine is actually overclaiming on alot of wu's
with the client your using, getting "overinflated" results.....if you use the
same optimized seti client from here at macnn, then use the trux boinc,
you will still be crunching work units just as fast as before, just not
overclaiming." I believe he's referring to alpha 4 which I haven't tried.

I also tried Alpha 5 with the Trux calibrating client and had similar results....

In short, what trux does is give you a different Boinc Manager and it will
optimize the reporting data on work units to provide a more favorable
credit. See the separate thread on that.

I might try Alpha 4 with it just for fun but I was hoping to get 2K rac with
Alpha 5.2 and I'm very close but no cigar so far.
     
BTBlomberg
Forum Regular
Join Date: Sep 2005
Location: Chicago Suburbs
Status: Offline
Mar 28, 2006, 06:03 PM
 
I tried it to on my PowerBook G4 500 (Overclocked to 550Mhz) and saw the same results. I have this on all my PCs (Linux and XP) and it is great there. Gave each a nice Claimed Credit boost which in turn effects the averaging Granted Credit. Some PCs Doubled their RAC.

With the Mac Trux Boing I went from where I got 25 to 32 Credits Claimed per WUs I got 9 or 10 and that was after several dozen WUs (You have to wait through a couple dozen WUs for this Boinc to calibrate the benchmarks for the correction). My RAC Dropped dramatically on that machine. I switched back after 3 or 4 days to the straight Boinc for Mac. Now it's OK again.

Now what the compiler of the Mac version said is definately wrong adn this is becasue the whole point of the Calibrated BOINC Client was to make sure each WU claims around 32 Credits. This is what a Boinc SETI Client is suposed to get for each WU. With Optimized Clients they have droped as the Benchmarks have not been adjusted for the change in the client code.

Since I saw that before switching and after going back I have been getting around 32 Claimed Credits per WU I figure it goes to show that Alex and Rick must have their calibration right. I am using the Alpha 5 for G4 client.
     
sdubz
Junior Member
Join Date: Jun 2006
Status: Offline
Mar 28, 2006, 07:06 PM
 
Originally Posted by BTBlomberg
With the Mac Trux Boing I went from where I got 25 to 32 Credits Claimed per WUs I got 9 or 10 and that was after several dozen WUs (You have to wait through a couple dozen WUs for this Boinc to calibrate the benchmarks for the correction). My RAC Dropped dramatically on that machine. I switched back after 3 or 4 days to the straight Boinc for Mac. Now it's OK again.

If you don't put the truxoft_prefs.xml in the correct place (BOINC data folder) it won't calibrate, it will act mostly like a "normal" boinc client.

Been using trux, along with the altivec from here and my rac on a G4 mac mini is over 240. http://setiathome.berkeley.edu/show_...hostid=2250966
     
BTBlomberg
Forum Regular
Join Date: Sep 2005
Location: Chicago Suburbs
Status: Offline
Mar 28, 2006, 07:34 PM
 
Boog,

That is where I have it and it did calibrate only negatively. I would check in the Messages area of the Boinc Manager and it would have a Line for the Calabration where it would say things like:
CC Calibration: 31.40 >> 18.37 (time: 9643s >> 12138s /...
Slowly I guess it did work up to WU CCs of 15 - 20 looking back, but still not the 32 to be expected and as i said it rean for 3 or 4 days ad had done . Took my Powerbook down from a RAC of 160ish to the 90s and is now about 136 while it recovers.

Also a RAC of 240 on an G4 Mini I guess must be guaged by your CPU speed. CPUs with smaller caches will perform a little slower that one my think if you compare raw CPU MHz. It does appear to be working for you or you are just lucky to have the right setup that it hits just right.

I am up for trying again but I still have the truxoft_prefs.xml right where it belongs. I just suspect an error in the compile myself. Could be related to the Benchmarking the CPU, which factors into the claibration.
     
sdubz
Junior Member
Join Date: Jun 2006
Status: Offline
Mar 28, 2006, 10:44 PM
 
My mini is a 1.5ghz (I overclocked it).

When I first started it on the trux version it only claimed like 4 per wu, set some optimizations (-O3, unroll loops, etc), and got it up a little higher. I'm far from being a programmer, just a bit of a hack.

I combined the directions that come with the standard boinc source and combined the directions (somewhat) from trux's bsd directions to get this build.

If someone else with more knowledge would like to try to build a better benchmarking version they are more than welcome to, because that would save waiting so long for the calibration to get back in line, but it will get back up close to 32 (mine claims from 29 to 31 normally).
     
mikkyo
Senior User
Join Date: Feb 2002
Location: Silly Valley, Ca
Status: Offline
Mar 29, 2006, 03:38 AM
 
I think the only reason the RAC goes up with the trux client is because it returns the results immediately, instead of at the next WU download time.
Every time you complete a WU, it sends it back, the completed ones don't stack up so that several are sent at once.
     
phantomac
Forum Regular
Join Date: Oct 2003
Location: Germany
Status: Offline
Mar 29, 2006, 05:04 AM
 
Dual 2.7 G5 again:

Ok, I've created my own wisdom file (still in single user mode, without the GUI loaded, no other processes of note running). No speed gain compared to E.T.s wisdom file.

I've swapped the RAM DIMMs' order so that now the four Crucial 1 GB DIMMs are first and the original two Apple 512 MB DIMMs are last. No speed gain, instead the machine got even slower (about 3-4 Minutes per WU).

I've disabled NAP. No speed gain.

With regard to the thing that happened when I switched the DIMMs' order, I now believe even more that the problem is the bridge between the RAM and the 1,35 GHz Bus. It's too much out of sync speedwise, when compared to the "slower" G5 CPUs available. There're probably a lot of clock cycles wasted there.
     
Todd Madson
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status: Offline
Mar 29, 2006, 04:31 PM
 
Phantomac: Check the ram speed with the Apple memory removed.

If the Apple memory is slower than the Crucial, the machine will slow
the entire system to that speed.

If the Crucial ram is faster, give that a try. Just pull the Apple ram out
temporarily.

Also, try caching 5-6 days of blocks just to see what happens.

I found the biggest problem of them all is that the G5 frequently outruns
the ability of the Seti project to feed it blocks. I set it up for 5.5 days for
a short while and then later 8.5. In my "machines" page it claimed I'd
processed 850-915 blocks for the most part (that recent number).

Today it is saying just under 500 as there's not enough blocks to go around.
So the machine is really idling most of the time with a ton of blocks at 100%.

Some of the really bad results I've been getting lately is because
the Seti/Boinc folks are having problems with their servers.

My best RAC was just under 2000, I keep hovering in the mid 1900s now.

One more thought: you might want to try the Trux optimization to see
how that works for you.
( Last edited by Todd Madson; Mar 29, 2006 at 04:52 PM. )
     
brysonda
Fresh-Faced Recruit
Join Date: Jan 2006
Status: Offline
Mar 29, 2006, 05:27 PM
 
Phantomac: Another thing to try is using the command line version (can be gui compatible). I noticed that my Quad would cease computing with relative frequency while using the BOINC MenuBar, even though it had plenty of work units in its cache. It would pretty much always do so during 'outages' (similar to what I've read on this or other forums about it), but would happen other times as well. After switching to a CLI superbench client, it has been computing continuosly on all cached WUs and fared well during the weekend outage.

I meant to use the pure CLI version, but grabbed the GUI-compatible version instead. So its not the core client thats the problem. Must be something that results from the combination of the GUI and the core.
     
Todd Madson
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status: Offline
Mar 30, 2006, 12:18 PM
 
I've wondered about this quite a lot and the fact that you mentioned it
has pretty much confirmed it. For reasons completely unknown the
Boinc menubar does have periods where it just stops the whole works
for no particular reason.

I checked and had well in excess of 120 work units to be crunched and
yet my output is nearly half what it usually is. I can't watch the cpu all
day so I will likely have to retool my setup.

Okay, question: how do I go from the Superbench menubar client to the
CLI version - what do I need to download? How is it configured?
     
brysonda
Fresh-Faced Recruit
Join Date: Jan 2006
Status: Offline
Mar 30, 2006, 12:53 PM
 
I believe I grabbed the CLI 5.2.13 Gui-compatible superbench binary for the G5. Just unpack it and move the file to the boinc directory. For example, in a Terminal:

mv {boinc cli binary} ~/Library/Application\ Support/BOINC\ Data/

Fortunately the directory setup is the same, so no reconfiguration is necessary. Simply exectute the binary within a Terminal window while in the boinc directory. For example:

cd ~/Library/Application\ Support/BOINC\ Data/
./{boinc cli binary}

It should start up and spit out text that looks like the Log window from the Menubar. All you existing work units and seti binary setup should work the same. To quick boinc, press ctrl-c.


If you don't want to keep the terminal window floating around, there are various options for running it in the background. One, is to simply tell the shell to run it as a detached process (add an ampersand after the command... ./{boinc cli binary} &). This is easy, but will make it more difficult to check on the status of things. To kill it you will need to run "killall boinc" from a Terminal, or find the process in Activity Monitor.

Alternatively, you can use a nice shell utility call 'screen'. Open a Terminal, type screen and hit enter. Then do the steps above to run boinc. Once boinc is running ok, you can hit ctrl-a, followed by ctrl-d. This will cause screen to detach from the current shell, saying "[detached]". Now you can close the terminal and boinc will stay running in the background. To get back to boinc to check on it, you can re-attach to the screen session by opening a terminal and running screen with the argument -x (screen -x). Use ctrl-a, ctrl-d to detach again. To completely exit screen, simply exit out of the shell that it is running. I use this method on a number of Xserves and Power Mac servers that I run headless. Screen has many other nifty features, but those are the basics.
     
Shaktai
Mac Elite
Join Date: Jan 2002
Location: Mile High City
Status: Offline
Mar 31, 2006, 12:52 AM
 
Does anyone have a wisdom file for the original G5 iMac 1.6 ghz?

UPDATE: Never mind. I found that the wisodm file for the Original 1.8ghz G5 works just fine. Cut my processing time by more then half.
( Last edited by Shaktai; Mar 31, 2006 at 09:52 PM. )
     
phantomac
Forum Regular
Join Date: Oct 2003
Location: Germany
Status: Offline
Mar 31, 2006, 01:33 AM
 
Originally Posted by brysonda
Phantomac: Another thing to try is using the command line version (can be gui compatible). I noticed that my Quad would cease computing with relative frequency while using the BOINC MenuBar, even though it had plenty of work units in its cache. It would pretty much always do so during 'outages' (similar to what I've read on this or other forums about it), but would happen other times as well. After switching to a CLI superbench client, it has been computing continuosly on all cached WUs and fared well during the weekend outage.

I meant to use the pure CLI version, but grabbed the GUI-compatible version instead. So its not the core client thats the problem. Must be something that results from the combination of the GUI and the core.
Ok, I just switched to the CLI client. I never had the GUI client stop on me, though. Let's see if it's worth anything. :-)
     
Todd Madson
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status: Offline
Mar 31, 2006, 01:19 PM
 
Let me know - I haven't had time to mess with it and I'm finding that my RAC is
getting slowly worse and worse but I believe it has a lot to do with the recent
seti outages due to their equipment issues.
     
gulliver
Fresh-Faced Recruit
Join Date: Nov 2005
Location: Europe
Status: Offline
Apr 2, 2006, 08:57 AM
 
Originally Posted by phantomac
Dual 2.7 G5 again:

Ok, I've created my own wisdom file (still in single user mode, without the GUI loaded, no other processes of note running). No speed gain compared to E.T.s wisdom file.

I've swapped the RAM DIMMs' order so that now the four Crucial 1 GB DIMMs are first and the original two Apple 512 MB DIMMs are last. No speed gain, instead the machine got even slower (about 3-4 Minutes per WU).

I've disabled NAP. No speed gain.

With regard to the thing that happened when I switched the DIMMs' order, I now believe even more that the problem is the bridge between the RAM and the 1,35 GHz Bus. It's too much out of sync speedwise, when compared to the "slower" G5 CPUs available. There're probably a lot of clock cycles wasted there.
Hi phantomac,

I see exactly the same. I have a G5/Dual 2.5GHz machine that needs an average of 2,300 seconds/WU and a G5 Dual 2.7GHz machine that takes an average of 2,700 seconds/WU. Both machines have 1 GB RAM and exactly the same setup. Both run Alpha 52. The 2.5 has the stock ATI 9600XT card, the 2.7 runs a ATI X800XT.

Weird, isn't it?
     
phantomac
Forum Regular
Join Date: Oct 2003
Location: Germany
Status: Offline
Apr 2, 2006, 09:17 AM
 
Switching to the CLI client didn't make any difference, btw. Same speed as before.

I wonder if I should try replacing the X800XT with the stock 9650 graphics card. I just can't see how that would make a difference.
     
mikkyo
Senior User
Join Date: Feb 2002
Location: Silly Valley, Ca
Status: Offline
Apr 2, 2006, 04:04 PM
 
What brand and speed of RAM are in the similar machines?

If you want even more speed, quit the Finder ( you can always relaunch it from the Dock).

Make an AppleScript with

tell application "Finder"
quit
end tell

save as an Application and drag to your Dock so you can quit it whenever you want.

Or via command line:
osascript -e 'tell application "Finder"' -e 'quit' -e 'end tell'

Click the Finder Icon in the Dock to launch it again when you need it, or via command line
open Finder

For even more speed, don't log in on the machine and set up a cron job to run at boot.

For the most speed possible(if you want a dedicated cruncher), turn off all services, (web sharing, file sharing, etc) and set the machine to boot to console (no GUI)
( Last edited by mikkyo; Apr 2, 2006 at 04:49 PM. )
     
Gecko_r7
Forum Regular
Join Date: Oct 2005
Location: Las Vegas, NV
Status: Offline
Apr 2, 2006, 10:32 PM
 
Originally Posted by mikkyo
What brand and speed of RAM are in the similar machines?

If you want even more speed, quit the Finder ( you can always relaunch it from the Dock).

Make an AppleScript with

tell application "Finder"
quit
end tell

save as an Application and drag to your Dock so you can quit it whenever you want.

Or via command line:
osascript -e 'tell application "Finder"' -e 'quit' -e 'end tell'

Click the Finder Icon in the Dock to launch it again when you need it, or via command line
open Finder

For even more speed, don't log in on the machine and set up a cron job to run at boot.

For the most speed possible(if you want a dedicated cruncher), turn off all services, (web sharing, file sharing, etc) and set the machine to boot to console (no GUI)
Mikkyo: I set-up the script as you explained as an application and when I run it while the script editor is open, it works great. When I close the script editor however and run my newly created application by itself, Finder quits and then Finder re-launches automatically. Any ideas? I noticed that if I go back into the script, select "save-as" and check the box "stay open", it works and Finder remains de-activated, but the script application must remain open. Defeats the purpose this way, doesn't it; for Finder to stay closed, another must stay open? I must be doing something wrong. Sorry. I'm a NB with these kind of things. If anything, this forum is helping me to get better familiar w/ the many neat features of OSX. Thanks!
     
mikkyo
Senior User
Join Date: Feb 2002
Location: Silly Valley, Ca
Status: Offline
Apr 3, 2006, 03:50 AM
 
Try saving the script as Run Only.
     
Fireman Sam
Guest
Status:
Apr 4, 2006, 03:44 AM
 
Just a little help needed installing this optimised client...... I read the post way up but not sure If I've got everything where it needs to be? I'm running a G5 2x2GHz with 10.4.6

I installed the original client from the seti website, then downloaded the "[email protected] (Apple vDSP) for G5 (Tiger)" folder.

I then dragged the "setiathome-4.18-applevdsp-tiger-G5-alpha-4" and "app_info.xml" files into the Library>application support>BOINC data>projects>setiathome.berkeley.edu folder.

Is this all correct, or have I missed something somewhere? All the files in the setiathome.berkely.edu folder are now:

31mr02aa.9452.20800.165912.1.62_2_0
07oc02aa.6989.1361.573572.1.68_0_0
31mr02aa.9452.20800.165912.1.62
07oc02aa.6989.1361.573572.1.68
app_info.xml
setiathome-4.18-applevdsp-tiger-G5-alpha-4

Thanks for any help.......
     
sdubz
Junior Member
Join Date: Jun 2006
Status: Offline
Apr 5, 2006, 06:26 AM
 
Originally Posted by Fireman Sam
J

31mr02aa.9452.20800.165912.1.62_2_0
07oc02aa.6989.1361.573572.1.68_0_0
31mr02aa.9452.20800.165912.1.62
07oc02aa.6989.1361.573572.1.68
app_info.xml
setiathome-4.18-applevdsp-tiger-G5-alpha-4

Thanks for any help.......
That looks correct to me!
     
halimedia
Dedicated MacNNer
Join Date: Oct 2005
Location: Switzerland
Status: Offline
Apr 6, 2006, 05:54 AM
 
Originally Posted by boog
That looks correct to me!
Well, it does - but you don't want to use alpha-4 on Tiger. Check <a href="http://forums.macnn.com/showthread.php?t=266339 target="_blank">the first post of this thread</a> for more info on how to get alpha-5.2 running. It's well worth it on a DP 2.0 G5!

HTH,

Ron
     
Todd Madson
Mac Elite
Join Date: Apr 2000
Location: Minneapolis, MN USA
Status: Offline
Apr 13, 2006, 12:42 PM
 
Well, it's taking me bloody ages to get back to the kind of RAC I had prior
to the trux-test. World position dropped to the 185-ish spot. Geez.

The primary adage: if it isn't broken, don't fix it!
     
Brazuca
Mac Elite
Join Date: Oct 2000
Status: Offline
Apr 14, 2006, 12:29 PM
 
Man, this is a pain.

Can someone point me to the "simple" instructions on how to set up Boinc to attach to both Seti and to WorldCommunityGrid?

Do I need seperate clients or something? I downloaded the optimized Seti client recommended here, but I'm not sure how to run it (double-clicking on the file results in an "exit" and "logout" in Terminal. Do I need something else?

Why in the world is this so complicated??


edit: Ok, so I got the latest client for my dual 2.0 G5 (alpha-52-g5) as indicated in the first post of this thread. I also found the bigfft_wisdom file posted on pg 15. What now? Where is the "worker"?

edit2: So, the "worker" is the official BOINC client that I get from the official site? Or is there a MacNN version? How do I get the optimized "client" to run with the official "worker"? Should I download the CLI or the GUI version (If I get the CLI version, can I get a GUI to control it)?
( Last edited by Brazuca; Apr 14, 2006 at 12:46 PM. )
"It's about time trees did something good insted of just standing there LIKE JERKS!" :)
     
Brazuca
Mac Elite
Join Date: Oct 2000
Status: Offline
Apr 14, 2006, 12:56 PM
 
edit 3: found the instructions on page 7. Will try now.

That really should be linked on the first post or on the webpage.
"It's about time trees did something good insted of just standing there LIKE JERKS!" :)
     
 
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
Top
Privacy Policy
All times are GMT -4. The time now is 07:08 PM.
All contents of these forums © 1995-2017 MacNN. All rights reserved.
Branding + Design: www.gesamtbild.com
vBulletin v.3.8.8 © 2000-2017, Jelsoft Enterprises Ltd.,