Welcome to the MacNN Forums.

Knightrider · 03:18 AM

Originally Posted by Andrew F

I'll try making my own wisdom file, if it's advisable...
Alex, how do we run this fft2_test in the terminal?

I put the fft2 file into my applications folder, opened terminal and then drag and dropped on to the terminal window, hit return and it ran ok. Takes a while to finish and then writes the wisdom file to the same folder.

K.

@ Alex. All results so far are valid ones. Times are mostly +/- 2000. The shortist is 58.29, longest is 3,165.97

K.

Fortilan · 05:51 AM

Hi all,

before I run one for myself, does anyone have a bigfft_wisdom file for an iMac G5 they would like to share?

***** Update: I now have a bigfft_wisdom file for an iMac %8 1.8 Ghz if anyone needs one

Todd Madson · 08:02 AM

Fortilan: I think Alex said the file he didn't have was for an iMac G5 so
you will need to run it.

Fortilan · Feb 9, 2006, 03:24 PM

Hi, here are my test 2 results...

these are the results for a 1.8Gh iMac G5

Machine Name: iMac G5
Machine Model: PowerMac8,1
CPU Type: PowerPC G5 (3.0)
Number Of CPUs: 1
CPU Speed: 1.8 GHz
L2 Cache (per CPU): 512 KB
Memory: 768 MB
Bus Speed: 600 MHz
Boot ROM Version: 5.2.2f2

====== results =========
Apple vDSP op / ip
1024 8184.007805 7540.321798
2048 8112.060484 7770.500042
4096 7002.664070 7126.605027
8192 4235.025398 3860.244389
16384 3505.686925 3684.408220
32768 1123.474286 2607.857409
65536 843.473546 1303.084738
131072 809.688210 886.441871
weighted time 17.880000 15.220000 14.090000
FFTW3 interleaved op / ip (this may take a minute)
1024 7895.160471 6100.805818
2048 7381.975040 4101.097244
4096 6882.960410 4549.753492
8192 5486.888252 3421.236204
16384 3981.034305 3152.765423
32768 2324.787436 2308.791193
65536 1609.807832 1597.830095
131072 1580.125607 1615.935819
weighted time 9.200000 9.070000

where do I get the new client/worker to try it out with my bigfft_wisdom file?

amigoivo · Feb 9, 2006, 08:40 PM

Originally Posted by Fortilan

Hi, here are my test 2 results...

.....
where do I get the new client/worker to try it out with my bigfft_wisdom file?

Hello Fortilan,

i was shure that the link was postet on top of page 14 of this thread, the only link that i now found where on the bottom.

This one:
http://inst.eecs.berkeley.edu/~alexk...seti_apple.zip

But the access to this file is permitten.

I read that this one should include the source code too, because of the GPL License.

CU, Ivo

Todd Madson · Feb 10, 2006, 12:56 PM

The link above is for the source.

This link is for the executable:
http://inst.eecs.berkeley.edu/~alexk...lpha-52-g5.tgz

Fortilan · Feb 10, 2006, 05:11 PM

Originally Posted by Todd Madson

The link above is for the source.

This link is for the executable:
http://inst.eecs.berkeley.edu/~alexk...lpha-52-g5.tgz

Thanks Tod,

I have tried the bigfft_wisdon file out with 5.2 but the result initially have been slower than before. I only have a single CPU G5 1.8 Gh iMac so maybe the 'normal' config is better for my set up.

I'm not even sure it is using the 'wisdom' file as the stderr details in the processed units does not mention it and I dont get a load message when my client is starting. I use the 5.2.13 superbench Menubar version.

All my data files are in ~/Library/Application Support/Boinc Data/... and I have placed the 'wisdom' file in /Library/Application Support/Boinc Data/ and in ~/Library/Application Support/Boinc Data/ just to be sure it would find it.

How can I be sure it is finding it and using it properly?

Do I need to try another client? Maybe it does not work well with the superbench Menubar client.

Any help appreciated

Feb 10, 2006, 09:45 PM

huh, I seem to have been getting client errors for the past week, running alpha-5. Probably since I upgraded to 5.2.13. Now with alpha-5.2 and a new bigwisdom, i'm getting about 50% success. I'm not sure what's up. Especially since I can't find a record of SETI even running in my log... all other project activity shows up, but it seems like it's never running SETI.

halimedia · Feb 11, 2006, 06:59 AM

If you're using alpha-5.2, make sure you place bigfft_wisdom in the exact same place as the optimized worker. For folks using the GUI versions of BOINC (Manager, Menubar), this would be:

.../BOINC Data/projects/setiathome.berkeley.edu/

HTH,

Ron

Mark Asiala · Feb 11, 2006, 09:08 AM

Originally Posted by Fortilan

How can I be sure it is finding it and using it properly?

If it's running slower (about 2x slower) it's likely not loading but you can confirm this when you look at a completed workunit (go to mycompters on the s@h site, then look at workunits associated with it), the stderr message associated with your workunit will say either that the wisdom_bigfft file was successfully loaded or not found.

It does not show up in the local log since that reflects stdout and not the stderr message you see on the s@h site. I think it would be a good refinement of the program to write this message to stdout as well so that the user would know it was loaded sucessfully before walking away and assuming that the installation was okay.

Karl Schimanek · Feb 11, 2006, 10:39 AM

Code:
E.T from Tellus  4,289.01   263,501.90  Created  25 Nov 2005 15:51:40 UTC
Bournecruncher  3,572.71   331,163.36  Created  30 Dec 2005 10:58:13 UTC

Regards
Karl

brysonda · Feb 11, 2006, 01:24 PM

I have been using the seti worker A5 for about a week and have been loving the improved performance. Thanks Alex, Rick, and the forum for the work that has gone into building and testing it.

After reading through the forum, I'd like to upgrade to 5.2 to try it out, but have noticed its a little tough to dig through and find the bigfftwisdom files for the various machines. Because of this, I was wondering if it would be desirable to start up a website that could be more easily shared by those involved in the project? I have a reseller web hosting account that I use for some small websites. I am willing to register a domain and create an account for the community if you guys are interested.

If so, how does macseti.org sound?

Thanks again guys!

P.S. - Here are all my current hosts running A5:
PowerMac G5 Dual 1.8
PowerMac G5 Quad
17in PowerBook G4 1.67GHz

Todd Madson · Feb 11, 2006, 02:35 PM

I've noticed (not empirically but just noticing...) that Alpha 5.2 is on average
150-200 seconds slower (still a lot faster than Alpha 5) but the RAC is
significantly higher than Alpha 5 or 6.

When I was running Alpha 5 I'd get a RAC of 800-950 or so at the very most.
Runtimes were in the 3500-4500 seconds range.

Alpha 6 was very fast - 1900-2400 seconds on a DP 2.5 but the RAC would
top out somewhere around 1100 or thereabouts.

Alpha 5.2 is still fast (2100-2550 seconds) but the RAC is in the 1400 range
which is actually better than having a somewhat faster client.

If the speeds of Alpha 6 can be combined with the math accuracy of Alpha 5.2
you would have what I'd deem a perfect client.

Still, anyone who was running Alpha 5 and is now running Alpha 5.2 should be
seeing greatly improved standings and RAC I believe...

Karl Schimanek · Feb 11, 2006, 03:12 PM

I have an invalid result with alpha5.2:
http://setiathome.berkeley.edu/worku...?wuid=54928027

Regards
Karl

Fortilan · Feb 11, 2006, 03:27 PM

Thanks, now I know I had it in the wrong place

I'll give it a go later today

beadman · Feb 11, 2006, 06:53 PM

Originally Posted by brysonda

...
P.S. - Here are all my current hosts running A5:
PowerMac G5 Dual 1.8
PowerMac G5 Quad
17in PowerBook G4 1.67GHz

brysonda: You might consider swithing to the Team MacNN Optimized BOINC. Compare your 17inch PowerBook to my 15inch 1.67GHz PowerBook:

http://setiathome.berkeley.edu/show_...hostid=2041993

Our completion times are the same, but the claimed credit is quite different. You can download the 5.2.13 Team Optimized Boinc her: http://members.dslextreme.com/~reade...boincbeta.html I've found the optimized version to be very stable.

beadman

brysonda · Feb 11, 2006, 11:52 PM

Alex, fyi... had a work unit that failed with a computing error. Looks like it failed to parse it.

http://setiathome.berkeley.edu/resul...ltid=230587261

halimedia · Feb 12, 2006, 07:31 AM

Originally Posted by brysonda

Alex, fyi... had a work unit that failed with a computing error. Looks like it failed to parse it.

Looks like a rogue WU - see here: http://setiathome.berkeley.edu/worku...?wuid=55177857

halimedia · Feb 12, 2006, 07:53 AM

Alex: I just reviewed this week's SETI results of my Quad, and discovered four invalids among a total of over 900 completed results:

http://setiweb.ssl.berkeley.edu/work...?wuid=54881296
http://setiweb.ssl.berkeley.edu/work...?wuid=54822520
http://setiweb.ssl.berkeley.edu/work...?wuid=54053509
http://setiweb.ssl.berkeley.edu/work...?wuid=54041889

NB: I'm still using alpha-5.2 pre-release...

HTH,

Ron

halimedia · Feb 12, 2006, 08:02 AM

Three invalids out of ~240 results from my DP 2.5 GHz:

http://setiweb.ssl.berkeley.edu/work...?wuid=54341220
http://setiweb.ssl.berkeley.edu/work...?wuid=54020098
http://setiweb.ssl.berkeley.edu/work...?wuid=53845541

In contrast, an Xserve DP 2.0 GHz running alpha-4 on Panther has had one invalid during the same period.

Feb 13, 2006, 02:31 AM

Just switch from a5 to a52 on my Quad. WU for a5 seemed to run between 2000 to 2700 sec, a52 seems to be running about 1000 sec slower in the 3000 to 3700 sec range. I ran fft_test2 and put the bigfft_wisdom file in the projects/setiahome directory with the worker. stderr out is:
<core_client_version>5.2.13</core_client_version>
<stderr_txt>
OS X Altivec-optimized vDSP/FFTW S@H application by Rick Berry and Alex Kan
version: alpha-5.2 (public release)
successfully loaded FFTW wisdom from bigfft_wisdom
FFT usage: vDSP out-of-place for FFT sizes < 2^13
FFTW out-of-place for FFT sizes >= 2^13
</stderr_txt>

Knightrider · Feb 13, 2006, 06:05 AM

brysonda · Feb 13, 2006, 09:45 AM

Originally Posted by Bob Palmer

Just switch from a5 to a52 on my Quad. WU for a5 seemed to run between 2000 to 2700 sec, a52 seems to be running about 1000 sec slower in the 3000 to 3700 sec range. I ran fft_test2 and put the bigfft_wisdom file in the projects/setiahome directory with the worker.

Interestingly, I did the same procedure but saw almost the opposite. Before a52, WUs were taking around 3600-3700 sec. After, they went down, 2800-2900 sec. Pretty sweet improvement.

However, I haven't been getting the really impressive numbers like 1800-2000 that some of the Quads have been listing for a52. (bigfft_wisdom loads successfully) I wonder how sensitive the calculation of bigfft_wisdom is...

Feb 13, 2006, 12:52 PM

Happy to report that after the first batch of results that run times are down to the 1800 to 2200 sec range using a52.

Todd Madson · 01:04 PM

Very odd - I've been running A5.2 since the day of its release but my RAC for
my DP 2.5 appears to be increasing - today it is at an all-time high of over
1500.

I've said before that under Alpha 5 it barely cracked 900. This is great!

It is even appearing to show the odd speeding up behavior I saw under
Alpha 6 - I'm seeing a lot of units in the 1900-2300 second time range
now (last few 2137, 2142, 2070, 2437, 2061, 2396, 2355,2393, 2122
seconds).

Here's some odd units my machine worked on during the time of 5.2 that
might bear looking at:

http://setiathome.berkeley.edu/resul...ltid=230903887
http://setiathome.berkeley.edu/resul...ltid=230558551
http://setiathome.berkeley.edu/resul...ltid=230500898

But it appears the problem is actually the data supplied by Seti so it doesn't
appear to be the client.

adream · 10:02 AM

hi there,

here is the bigfft_wisdom file generated by my Dual 2.7 G5 with 3.5GB ram using fft_test2

http://www.zen68343.zen.co.uk/wisdom/

im getting averages of about 2500 sec per unit (alpha 5.2) instead of 4000 secs per unit (alpha 5)

and no errors other than bad workunits so far

i would be up for runnig the alpha 6 as an alpha tester if someone could point me at a download link

many many thanks for everyones time concerning this optimisation process, I for one really appreciate it

regards

adream

Todd Madson · Feb 14, 2006, 12:35 PM

adream - have you tried running with nap mode disabled with performance set to highest?

adream · Feb 14, 2006, 01:08 PM

yes all settings to highest nap disabled, the machine is in constant use for about half the day so the wu times will suffer, cant tell much till it gets left alone for a few days.

i'm impressed with the lack of client errors though pretty good..

wheres the alpha 6

regards

adream

Todd Madson · Feb 14, 2006, 02:02 PM

Alpha 6 didn't work as well as 5.2 as far as RAC was concerned and had a bigger
problem where invalid work units were concerned - Alexkan can provide with full
technical details on why it was scuttled. In a way 5.2 is a better "Alpha 6" than
"Alpha 6" actually was because of the fact that while it is still fast you actually
get credit for the vast majority of work units submitted where this wasn't the case
with Alpha 6.

Feb 14, 2006, 02:10 PM

I'm checking my results since I went from using 4.44 superbench to 5.2.13 (couldn't get 5.2.13 superbench to work) and I've been getting client error results ever since. What does this error message mean? Any suggestions about what I can do to fix this?

<core_client_version>5.2.13</core_client_version>
<message>process got signal 5
</message>
<stderr_txt>
dyld: seti@home-G5-a5 Undefined symbols:
seti@home-G5-a5 undefined reference to _vDSP_maxv expected to be defined in Accelerate
seti@home-G5-a5 undefined reference to _vDSP_maxvi expected to be defined in Accelerate
seti@home-G5-a5 undefined reference to _vDSP_meanv expected to be defined in Accelerate
seti@home-G5-a5 undefined reference to _vDSP_vrsum expected to be defined in Accelerate
seti@home-G5-a5 undefined reference to _vDSP_vsdiv expected to be defined in Accelerate
seti@home-G5-a5 undefined reference to _vDSP_zvmags expected to be defined in Accelerate

</stderr_txt>

Mark Asiala · Feb 14, 2006, 06:58 PM

Originally Posted by Andrew F

I'm checking my results since I went from using 4.44 superbench to 5.2.13 (couldn't get 5.2.13 superbench to work) and I've been getting client error results ever since. What does this error message mean? Any suggestions about what I can do to fix this?

<core_client_version>5.2.13</core_client_version>
<message>process got signal 5
</message>
<stderr_txt>
dyld: seti@home-G5-a5 Undefined symbols:

What operating system are you running? I think that both the optimzed boinc and the worker require MacOS 10.3.9. The error is a dynamic library load error, I believe.

brysonda · Feb 15, 2006, 04:24 PM

Is there anyone that has a Quad bigfft_wisdom file that they would be willing to share? I'm still getting consistently 2400-2900 times for my work units, even though I have seen a number of Quads getting 1800-2000 with a52.

E.T from tellus · 07:16 AM

Todd Madson · 02:32 PM

(Removed link since ET supplied one)

I've been tempted to create a bigfft_wisdom file in single user mode with no GUI.

I had no success before but since it's a long weekend and I could take some time
to look at things while it happens do you think this would benefit or other DP 2.5
owners in any way?

I suppose that would be optimal configuration - I'll try it anyway and see if there
is any improvement.

Update: My farm is four machines, my G5, my G4, an Athlon box I built and
a Pentium M laptop. Over the last few days my RAC has been rising and rising
on my group of machines and with the new Alpha my RAC on the G5 is 1591
and the RAC for my little group is over 2000 for the first time. I'm amazed.

My team is down to 4 people but we have a combined RAC of almost 2500
now.

We'll see what the future holds now.

brysonda · Feb 17, 2006, 10:12 AM

ET, thanks for posting those up.

I can definately confirm that how you create bigfft_wisdom is pretty sensitive... check out the Before and After views. Average time now seems to be around 1750-1800.

E.T from tellus · Feb 17, 2006, 02:03 PM

Thanks Todd, it is shared

brysonda · Feb 17, 2006, 02:27 PM

E.T. - I've just created a bigfft_wisdom file for my PB 1.67 GHz (17in Feb 2005 -- PowerBook5,7 if it matters). Used the processed described by Halimedia this time (restart, leave at login screen, ssh in from remote). If you want to add it to the collection, it can be found here:

http://programerror.com/packages/bigfft_wisdom

After a few work units I'll have an idea as to its performance in a52 over a5 that was previously running.

halimedia · Feb 17, 2006, 02:46 PM

Originally Posted by brysonda

After a few work units I'll have an idea as to its performance in a52 over a5 that was previously running.

My guess is it probably won't be worth it. I tried alpha-5.2 on all sorts of G4s, and the best I found was 5% slower than alpha-5. But I didn't have the chance to test one of the latest PBs. I'm curious to see if the DDR-2 RAM helps any.

FWIW,

Ron

halimedia · Feb 17, 2006, 02:49 PM

BTW, I started a thread to collect Dual-Core G5 sensor data. If you're interested in participating, please <a href="http://forums.macnn.com/showthread.php?t=286086" target="_blank">go there.</a>

Todd Madson · 05:40 PM

Alpha 5.2 invalids on my DP 2.5:
http://setiathome.berkeley.edu/resul...ltid=235434868
http://setiathome.berkeley.edu/resul...ltid=235040343

And:
http://setiathome.berkeley.edu/resul...ltid=230500898
Looks like a bad workunit header.

Not bad out of 21 pages of results:
http://setiathome.berkeley.edu/resul...hostid=1876496

And my G5 is finally over 1600 in RAC which is nice to see.

E.T from tellus · Feb 17, 2006, 06:51 PM

Originally Posted by brysonda

E.T. - I've just created a bigfft_wisdom file for my PB 1.67 GHz (17in Feb 2005 -- PowerBook5,7 if it matters). Used the processed described by Halimedia this time (restart, leave at login screen, ssh in from remote). If you want to add it to the collection, it can be found here:

http://programerror.com/packages/bigfft_wisdom

After a few work units I'll have an idea as to its performance in a52 over a5 that was previously running.

Thanks, it is shared.
Seems to be that i need build website

TiloProbst · Feb 17, 2006, 07:26 PM

Can someone provide a wisdom file for a iMac G5 2,0 GHz?

I have the chance to install 5.2 on such a machine, but no second computer in range to produce a Wisdom file via >console/ssh

Todd Madson · Feb 18, 2006, 12:34 AM

Why don't you use the iMac G5 2.0 ghz to generate the wisdom file?

TiloProbst · 11:38 AM

Well, as I said, I am missing a second computer to execute fft_test2 remotely via ssh.
I can login with >console of course, but the when asked for login/pwd, the usual Admin password does not work, and I don't want to activate the root user (assuming that his password will work). I mean, does it make sense to create the wisdom file locally with GUI still on?

EDIT: I started the fftw_test2 remotely with my sister's PowerBook. but seemingly the test suddenly stopped at
(...)
FFTW3 interleaved op / ip (this may take a minute)
(...)
8192.
the iMac went completely quiet. I then hit the space bar on the iMac's keyboard. the fans spinned up to maximum suddenly, while the stream of air coming out at the top was quiet cool. after that the iMac didn't respond to the keyboard or mouse anymore making it impossible to log out. so I did a force shutdown via the power button.

ok, applications hang up sometimes, that's not my point. my point is why the iMac suddenly went to sleep in the mid of some calculation. this reminds me of my Single 1,8 and Dual 2,0 G5 going quiet all of a sudden in the mid of processing Seti workunits in January with Alpha5 (I guess it does not happen with Alpha5.2 just because I am now here to keep both computers busy).

I searched through all possible logs, but as expected there were no records for the time since the possible sleep and the reboot after the force shutdown.

Lauger · Feb 18, 2006, 10:16 AM

Under Alpha 6, I was getting RAC around 380 on my G5 2.0. Latest RAC with 5.2 is 883! An amazing increase! Thanks Alex!

Todd Madson · Feb 19, 2006, 12:44 PM

I generated mine locally with the GUI still on. It kept hanging when I tried to
generate it using local user mode. Not sure why.

And my RAC went from 800-900 under Alpha 6 to an RAC of 1666.10 so far
and it appears to be increasing. My farm of machines is now an RAC of 2200.
Amazing - my entire team used to have that.

Lateralus · Feb 19, 2006, 10:51 PM

Okay guys. I've been away from the optimized SETI BOINC scene for a while due to a few frustrations with keeping it running properly, but now I'd like to get back into it.

Seems like I've missed a bunch so rather than having to read through half of this thread, would one of you be so kind as to give me a little step by step on installing all of the necessary goodies for BOINC and SETI again?

halimedia · Feb 20, 2006, 06:56 AM

Lateralus: what machine(s) do you have? The current situation is quite different for G4s and G5s...

TiloProbst · Feb 20, 2006, 09:35 AM

Originally Posted by Lateralus

Seems like I've missed a bunch so rather than having to read through half of this thread, would one of you be so kind as to give me a little step by step on installing all of the necessary goodies for BOINC and SETI again?

there is a installation guide on page 7 of this thread.

there will be an installation guide on the resource page in a few days, but I don't think you want to wait that long

in other news my Dual 2.0 Ghz G5 today achieved 1K RAC for the first time ever, and still slowly growing

more info on that computer here. Now for the first time since Summer 2005 I actually achieved my all-time goal to catch up to the high-end PCs.
thx Alex & Rick

Urban Mainframe · Feb 20, 2006, 09:45 AM

Halimedia: It looks like Lateralus is using a DP 2.5 G5 (in his sig). I too am using a DP G5 (2.7) and would also appreciate a step-by-step-guide if at all possible.

I have read through this thread and visited the various websites referenced. I have managed to get the optimised client running and am achieving good results with it. But I haven't been able to make much sense of all the "bigfft_wisdom file" stuff.

I'm sure a step-by-step guide would be useful to many.