[Todd Madson]

Updates on the G4/400:
More blocks now:
http://setiathome.berkeley.edu/resul...6261&offset=80

9100.21, 9062.28, 9143.59, 9100.72. Unbelievable. This machine has
never consistently produced times below 10k seconds before. Excellent.

Still trying to figure out what the deal is with the G5 - looks like right
before it got the new update it crunched a very long work unit, over an
hour and a half which is extremely rare for it to work so long on a bit
of data. Now everything is pretty consistently 1800-2200 all the time.

See:
http://setiathome.berkeley.edu/resul...496&offset=640

I'm still hoping for a higher RAC - I had a problem with the machine briefly
and my all-time-high RAC has dropped to 1700-something so my machine
went from 104 in the world to 220-something. Ugh. Hopefully this will let
it catch up.

[Gecko_r7]

Originally Posted by adream

navigate to the directory then type ./fft_test2

that should do it

adream

Ran into permission problem, but retyped chmod 755 ./fft_test2 and it worked like a charm! Thanks a bunch for all your help. Learned a few new things in this exercise as well. Excited to see how 5.3 works on my G4. Thanks again!

[amrad9]

Originally Posted by amrad9

Got a wisdom file that I generated for a G4 quicksilver 800 mhz where can I post it.

I have started up the alpha 5.3 version with no problems so far. It started reprocessing a unit that had the G4 alpha5 with no errors when restarting. I'll update you on how it's run throughout the day when I get home from work tonight.

amrad

First 3 work units processed with the alpha 5.3 version came out at 7710, 7787, 8159. Previous work units were from 8950 to 9550 range. Look like I'm processing work units about a 1,000 seconds faster with the alpha5.3 unit.

I looked at the results page and it is saying that it is using alpha 5.2. I opened the 5.3 version in text edit did a search and it also says 5.2 inside the file. Guess that the tag line that prints out in the stderr out file didn't get updated for 5.3.

amrad

[Gecko_r7]

Got 5.3 running on G4 w/ dual 1.33 now. @ 10 WUs have been returned. Times appear about the same as Alpha 5. Some are lower, some about the same, but are within the ranges I'm already used to seeing for A5. I'll see how it averages w/ a full day of production.

Bigfft_wisdom appears to have completed fine (@ 2.5 hours). Question: I generated in single user mode (Thanks adream! ) and saw 2 sets of output blocks, the last being FFTW3 before it finished and then I got the command prompt.
Is this correct?

Everything appears to have loaded fine according to WU output.

CPU time 5634.628477
stderr out <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>

Just want to double check w/ the group as I'm not really seeing an immediate impact like A3, A4 & A5 which were obvious immediately on this machine.
Anything jumping out that I may have done wrong?
Regards.

[adream]

Originally Posted by Gecko_r7

Got 5.3 running on G4 w/ dual 1.33 now. @ 10 WUs have been returned. Times appear about the same as Alpha 5. Some are lower, some about the same, but are within the ranges I'm already used to seeing for A5. I'll see how it averages w/ a full day of production.

Bigfft_wisdom appears to have completed fine (@ 2.5 hours). Question: I generated in single user mode (Thanks adream! ) and saw 2 sets of output blocks, the last being FFTW3 before it finished and then I got the command prompt.
Is this correct?

Everything appears to have loaded fine according to WU output.

CPU time 5634.628477
stderr out <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>

Just want to double check w/ the group as I'm not really seeing an immediate impact like A3, A4 & A5 which were obvious immediately on this machine.
Anything jumping out that I may have done wrong?
Regards.

there isnt such a big jump in performance between alpha 5.2 and 5.3 my dual 2.7 g5 used to average around 2400-500 with 5.2 and now it sits between 2100-200 which is still a very useful increase.

the main area of performance gains is using the fft test files with the g4 client, on my test g4 the numbers have gone from 9500ish to 8000ish which is quite an improvement

looks like the fft_test2 routine you ran worked fine, as the stderr is as it should be

regards

adream

[alexkan]

Originally Posted by Gecko_r7

Got 5.3 running on G4 w/ dual 1.33 now. @ 10 WUs have been returned. Times appear about the same as Alpha 5. Some are lower, some about the same, but are within the ranges I'm already used to seeing for A5. I'll see how it averages w/ a full day of production.

Bigfft_wisdom appears to have completed fine (@ 2.5 hours). Question: I generated in single user mode (Thanks adream! ) and saw 2 sets of output blocks, the last being FFTW3 before it finished and then I got the command prompt.
Is this correct?

Everything appears to have loaded fine according to WU output.

<snip>

Just want to double check w/ the group as I'm not really seeing an immediate impact like A3, A4 & A5 which were obvious immediately on this machine.
Anything jumping out that I may have done wrong?
Regards.

Run fft_test2 with the bigfft_wisdom file already in the directory, and paste the output here. (This skips the wisdom generation step and only does benchmarking, so it shouldn't take long.) The current code blindly switches between FFTW and vDSP at a certain cutoff point without taking into account which is actually faster, but it does this at a point which makes sense for almost all machines I've looked at so far. Your machine might be an exception to this rule, so seeing your benchmark numbers will help me to judge whether or not this is the case.

I was considering augmenting this FFTW/vDSP thing with a program that benchmarks to find the optimal combination of FFTs. Problem is, I'm already at least one release behind on FFTW, which claims to have made minor performance improvements. Also, it seems like generally bad form to be releasing new versions of the cruncher linked against old libraries. However, I'm told that wisdom generated with the version I used isn't usable for later versions. Hence, my dilemma is between a unknown minor performance gain (which I haven't measured yet or anything) versus making you all generate your wisdom again, which I know is time-consuming.

Comments, anyone?

[dwaring]

Can anyone tell me or at least direct me to the source of the significance of these bigfft_wisdom files that would bring to the optimised clients.....?

Thanks....

[Gecko_r7]

Originally Posted by alexkan

Run fft_test2 with the bigfft_wisdom file already in the directory, and paste the output here. (This skips the wisdom generation step and only does benchmarking, so it shouldn't take long.) The current code blindly switches between FFTW and vDSP at a certain cutoff point without taking into account which is actually faster, but it does this at a point which makes sense for almost all machines I've looked at so far. Your machine might be an exception to this rule, so seeing your benchmark numbers will help me to judge whether or not this is the case.

I was considering augmenting this FFTW/vDSP thing with a program that benchmarks to find the optimal combination of FFTs. Problem is, I'm already at least one release behind on FFTW, which claims to have made minor performance improvements. Also, it seems like generally bad form to be releasing new versions of the cruncher linked against old libraries. However, I'm told that wisdom generated with the version I used isn't usable for later versions. Hence, my dilemma is between a unknown minor performance gain (which I haven't measured yet or anything) versus making you all generate your wisdom again, which I know is time-consuming.

Comments, anyone?

Here you go Alex:

Machine Name: Power Mac G4
Machine Model: PowerMac3,6
CPU Type: PowerPC G4 (3.3)
Number Of CPUs: 2
CPU Speed: 1.33 GHz
L2 Cache (per CPU): 256 KB
L3 Cache (per CPU): 2 MB
Memory: 1.5 GB
Bus Speed: 167 MHz

Apple vDSP op / ip

1024 7294.441739 6331.024906
2048 6710.886400 5905.580032
4096 5592.405333 5033.164800
8192 2667.936489 2436.914056
16384 1568.487639 1803.309205
32768 1059.613642 1133.595676
65536 985.988819 1127.880067
131072 577.353587 1034.316127

weighted time 23.180000 14.060000 19.760000

FFTW3 interleaved op / ip (this may take a minute)

1024 5500.726557 4067.203879
2048 4528.819043 2617.721645
4096 4449.206453 2614.631065
8192 4095.846160 2301.887156
16384 2046.893455 1532.665736
32768 1373.305539 1068.612484
65536 1205.097446 964.727605
131072 791.707625 823.124595

weighted time 17.170000 17.340000

BTW, with regards to above 2nd paragraph, the benchmarking program sounds interesting. I also wouldn't mind having to run another wisdom or two if they're based on updated libraries.
Count me in if you need G4 testing! All for the science, right?

[Mark Asiala]

Originally Posted by Elphidieus

That's because I'm running it on an iMac in a tropical country where cooling is a problem.... I don't know about PowerMacs, but on an iMac with a tight enclosure, CPU temp reaches 170°F with 5.3...

My liquid-cooled G5 DP 2.5 running alpha 5.2 checks in at 177F and 154F for CPUs A and B using istat nano. Hardware monitor gives me 187 and 167 with the safe threshold of 190.4. HW monitor shows the drive bay at about 83F which is about 10F over the outside of the box temp.

I'm not really happy with a 10 degree swing between the two programs. Has anyone found a reliable temp monitor? If I had to guess by fan noise, I'd say that iStat nano is more accurate. What do other people use?

[dwaring]

Originally Posted by Mark Asiala

My liquid-cooled G5 DP 2.5 running alpha 5.2 checks in at 177F and 154F for CPUs A and B using istat nano. Hardware monitor gives me 187 and 167 with the safe threshold of 190.4. HW monitor shows the drive bay at about 83F which is about 10F over the outside of the box temp.

I'm not really happy with a 10 degree swing between the two programs. Has anyone found a reliable temp monitor? If I had to guess by fan noise, I'd say that iStat nano is more accurate. What do other people use?

Currently I'm using the same programs to monitor my system, and both have been reading them pretty consistently on all available sensor data. Deviations on both sides are only down to a tenth of a centigrade.

[Todd Madson]

For a performance improvement I have no problem generating another wisdom file.

Oh - looks like my G4 is back to generating 10K blocks again. Temporary improvement?
It generated some 8k blocks early yesterday - two in a row then 10k ever since.

Should the wisdom file be in the projects/setiathome.berkeley.edu folder or out in the
library/application data/boinc/ folder? It's been so long since I've done this I've forgot.

[adream]

Originally Posted by Todd Madson

For a performance improvement I have no problem generating another wisdom file.

Oh - looks like my G4 is back to generating 10K blocks again. Temporary improvement?

Should the wisdom file be in the projects/setiathome.berkeley.edu folder or out in the
library/application data/boinc/ folder? It's been so long since I've done this I've forgot.

the wisdom file, enhanced client and appinfo.xml should all be in projects/setiathome.berkeley.edu folder

remeber that the bigfft_wisdom file must have no extension (ie .txt.)

regards
adream

[Todd Madson]

Re: Temps....177 and 154 jibes with the data I'm receiving. I'm not concerned at all.
It's got nine fans and liquid cooling - it's meant for heavy duty processing. The only
concern some might have is the heat emanating out of the thing.

[Todd Madson]

I'll check the G4 a bit later. I know the G5 is correct but the G4 is acting a bit odd.
More later.

[Gecko_r7]

Shucks! After about 40WU, 5.3 doesn't appear to offer any improvement on my G4 MDD D-1.33 Times are averaging to be the same @ 5500 secs. Temps would always run level at 35C w/ A5, only increased 1C w/ 5.3 since I've changed. Ambient environment is pretty constant where I'm at so I'm sure the 1C increase is due to 5.3.

Another observation is that WU time clustering is having greater variation with more times ranging between 5000-6000, whereas w/ A5 they would mostly be in the 5400-5700 range, the occasional exceptions being a shorter or longer one. Not sure if this is relevant but some WUs are showing sterr repeated 2-3 or 4 times in Result ID. No validation errors so far.

[amrad9]

I have gained about 1500 seconds with 5.3 on my Quicksilver 800 mhz. WUs are averaging 7700-7900 seconds a wu some a little higher and some lower. With 5.0 I was averaging 9300-9500 seconds a wu.

The only temp sensor that is able to be read is from one of my HDs and it is at 40C. No other sensors avialable.

amrad9

[lepetitmartien]

Gone to the 5.3 this afternoon on my singe G5 1.8, units have come from about 2400-2500" with 5.2 to 2000-2100" with 5.3 for the first ones in.

Thumbs up!

Best rac to date: 624, and getting better.

[liebsmaschine]

I've been using alpha 5 for a couple of months now with no problems. In the spirit of helping the science :-) I came back to this thread to see if there was anything new. (There's a lot of catch-up reading to do for a newbie to this thread!)

Anyway, I found alpha 5.3 and downloaded the zipped file with the app_info.xml file included. I also downloaded the bigfft_wisdom file for the PB 1.67GHz that someone had posted. I chmodded the file to ugo+rwx.

I restarted BOINC and left it for a couple of hours while I ran about my business. When I came back, I had a string of "process exited with code 2" errors and a bunch of failed WUs.

I redownloaded the zip file thinking there had been some corruption (however unlikely) and decided to try to make my own bigfft file. So, after spending another hour reading through the thread (Alex, can you update your first post with some of the key things from the rest of the pages, like the fft_test2 app link and the link to the new alpha?), I found the fft_test2 app, downloaded it, and then ran it in single-user mode.

That's where my other problem happens: the computer freezes just as it's running the last benchmark in the second group of benchmarks, and no bigfft_wisdom file is outputted. (I know it freezes because the caps-lock key light becomes non-functional.) I'm in single-user mode with only two other processes running, so I'm not sure what it could be.

So, my question is: 1) anyone know what the cause of the error 2 is (or where exactly to get more info on what's causing it, and 2) why the fft_test2 executable is freezing my computer and/or if the wisdom file I got from the thread is valid enough to use (or if it could be causing the error 2s)?

Appreciate any help!
Chris

[Todd Madson]

I've been watching the G5 and what has happened has been very interesting -
with Alpha 5.2 I'd get a lot of work units varying all over the place - some of
them as long as 2500 seconds, some as little as 1800 but generally on the
high side.

Look at this:
http://setiathome.berkeley.edu/resul...496&offset=720

Eight work units all right around the 2100 mark. The next page, all 2100
or 2000 seconds.

It seems like Alpha 5.3 makes my machine consistently churn out work
units at a faster, more predictable pace.

As far as the G4 goes - I'm definetely noticing an improvement in that the
daily "Results" tally which used to be 80-95 is now in the 110-120 range
now. Nice. Even if the work units aren't consistently 9k anymore. I'll
look into it more in the coming days.

[amrad9]

Originally Posted by jackal

I've been using alpha 5 for a couple of months now with no problems. In the spirit of helping the science :-) I came back to this thread to see if there was anything new. (There's a lot of catch-up reading to do for a newbie to this thread!)

Anyway, I found alpha 5.3 and downloaded the zipped file with the app_info.xml file included. I also downloaded the bigfft_wisdom file for the PB 1.67GHz that someone had posted. I chmodded the file to ugo+rwx.

I restarted BOINC and left it for a couple of hours while I ran about my business. When I came back, I had a string of "process exited with code 2" errors and a bunch of failed WUs.

I redownloaded the zip file thinking there had been some corruption (however unlikely) and decided to try to make my own bigfft file. So, after spending another hour reading through the thread (Alex, can you update your first post with some of the key things from the rest of the pages, like the fft_test2 app link and the link to the new alpha?), I found the fft_test2 app, downloaded it, and then ran it in single-user mode.

That's where my other problem happens: the computer freezes just as it's running the last benchmark in the second group of benchmarks, and no bigfft_wisdom file is outputted. (I know it freezes because the caps-lock key light becomes non-functional.) I'm in single-user mode with only two other processes running, so I'm not sure what it could be.

So, my question is: 1) anyone know what the cause of the error 2 is (or where exactly to get more info on what's causing it, and 2) why the fft_test2 executable is freezing my computer and/or if the wisdom file I got from the thread is valid enough to use (or if it could be causing the error 2s)?

Appreciate any help!
Chris

I was having problems when trying to get it going also. Try these things.
1. chmod +x the alpha 5.3 worker. (this was my main problem, none of the other optomizations had required this)
2. make sure that the permissions are correct on the boinc folder and apply to the enclosing items. The follwing can be used in terminal: sudo chown -R root:wheel /Applications/Boinc
You need to change the /Applications/Boinc to navigate to your specific boinc folder

It takes a while to generate the wisdom file. It might not be locked up just calculating the test. It took about 2 hours to generate one on a G5 1.8 and over 5 hours to make one on my G4 800 mhz. I did mine right in terminal not in single user mode.

amrad9

[liebsmaschine]

Originally Posted by amrad9

I was having problems when trying to get it going also. Try these things.
1. chmod +x the alpha 5.3 worker. (this was my main problem, none of the other optomizations had required this)
2. make sure that the permissions are correct on the boinc folder and apply to the enclosing items. The follwing can be used in terminal: sudo chown -R root:wheel /Applications/Boinc
You need to change the /Applications/Boinc to navigate to your specific boinc folder

Although I think the permissions were workable before, I did chown -R root:wheel the entire /boinc directory just in case and verified the worker was ugo+x (which it was). In any case, no more quitting on error 2--it's working fine now. (Perhaps just restarting the computer fixed it--I was pretty sure I had restarted boinc, but maybe I had forgotten to quit something.)

I watched the first unit go to 50% done in about 6 minutes but then had to leave. Looking at the result online, it says 5788 seconds, so perhaps it slowed down significantly after I left. No info in the log to indicate it restarted the unit or anything. It's on the next unit now, and it's not nearly as speedy--in fact, this one's 30% done at 1 hr, so that looks to be about 3 hrs for this unit. This is actually slower than the previous alpha 5 that I was running, which would get them done in 1-1.5 hrs. Perhaps this is just a larger unit or something--I'll leave it running overnight and see what happens after this.

Also, in that link to the result online, I see the following under the stderr out line:

FFT usage: vDSP out-of-place for FFT sizes < 2^13
FFTW out-of-place for FFT sizes >= 2^13

That's not reflected in my stderrdae.txt file in the BOINC directory. In any case, is this a normal output or is this indicative of something wrong (say, an inconsistency with the bigfft_wisdom file)?

Originally Posted by amrad9

It takes a while to generate the wisdom file. It might not be locked up just calculating the test. It took about 2 hours to generate one on a G5 1.8 and over 5 hours to make one on my G4 800 mhz. I did mine right in terminal not in single user mode.

I'm pretty sure that it actually froze. When I ran it the second time, I hit the caps-lock key every half-hour or so all throughout the benchmarks, and it responded. Then, when it hit the one where it froze before (I'm sorry I can't remember exactly which one it was, but it was the last one under the FFTW3 tests, I think), the caps-lock key worked for a while and then quit responding and didn't work even after I left it another hour.

I'm using the one for the PB 1.67 GHz provided by E.T. from Tellus here. Should I really be concerned about using my own? Also, I'm not using a customized boinc client; is that something worth looking into? (I also run most of the other Mac-compatible boinc projects, so I don't want to mess those results up if I don't have to.)

[amrad9]

Originally Posted by jackal

Also, in that link to the result online, I see the following under the stderr out line:

That's not reflected in my stderrdae.txt file in the BOINC directory. In any case, is this a normal output or is this indicative of something wrong (say, an inconsistency with the bigfft_wisdom file)?

Good that you got it running. These show up in all of my results with the wisdom file so, its normal. You can look at other results from the top computers which are macs and see the same result. Guess this message board does not like quotes inside of quotes cause the FFT usage quote didn't show up when replying.

Originally Posted by jackal

I'm using the one for the PB 1.67 GHz provided by E.T. from Tellus here. Should I really be concerned about using my own? Also, I'm not using a customized boinc client; is that something worth looking into? (I also run most of the other Mac-compatible boinc projects, so I don't want to mess those results up if I don't have to.)

I have several machines at work that I have setup to run bionc. They are different speeds of G5s and I used the wisdom files from ET. They all seem to be running those fine. I had to make one for my G4 because there was not a file avialable to download. The 5.2.13 client should be fine for you to use with the multiple projects. I use the 5.2.13 beta superbench client on most of the machines. I am just running seti at work. The G5s there process units in 40-50 mins. So, having the superbench client allows the machine to keep enough work on hand.

I looked at the workunits you had processed and looks like a couple were in the 5700 range. The others are averaging about 4850 seconds a unit. 80 minutes each for the powerbook doesn't seem to bad. Looks like your claimed credits seem a bit low. Try running the cpu benchmarks again to see if they improve and your claimed credit goes up.

amrad9

[rick]

Originally Posted by jackal

That's not reflected in my stderrdae.txt file in the BOINC directory. In any case, is this a normal output or is this indicative of something wrong (say, an inconsistency with the bigfft_wisdom file)?

You may have to grill Alex on the specifics but I think I can summarise:

This is perfectly normal. In the newer clients there are actually two different sets of routines that are used to calculate the FFTs (Fast Fourier transforms): vDSP (from Apple) and FFTW.

The reason that we use two is because the FFTW routines are faster than vDSP for larger FFTs. Seeing as most of the time is spent doing FFTs then it's actually worth the hassle of using two different libraries of routines (and this is the main reason for the performance in the alpha 5 clients, as far as I know).

Originally Posted by jackal

I'm using the one for the PB 1.67 GHz provided by E.T. from Tellus here. Should I really be concerned about using my own? Also, I'm not using a customized boinc client; is that something worth looking into? (I also run most of the other Mac-compatible boinc projects, so I don't want to mess those results up if I don't have to.)

As long as the wisdom file was generated in a "good" way (i.e. it was the only thing running apart from background system processes) then it will work fine on all machines with the same processor. If there is a sufficient difference between the processor types then you might want to generate your own wisdom file (e.g. running wisdom for a 1.67 GHz PowerBook on and older 800 MHz dual G4 PowerMac). In practice, the differences are likely to be minimal or zero.

Also, the BOINC client that you use will have no effect on the worker's wisdom file. The whole FFTW / wisdom file thing is custom code that Alex has written for the SETI worker and nothing else has any influence on it or is even aware of it.

The only thing that the BOINC client affects is how your credit is calculated which is based on how fast you processed a work unit (i.e. entirely based on how fast the worker is) and also an internal "benchmark" that the BOINC clients use.

I say "benchmark" because it really does not accurately represent performance on more modern CPU architectures (the benchmark routines were written in the late 80s / early 90s if I recall). As far as I'm aware, the whole custom BOINC scene is an effort to implement more accurate benchmarks that match up to reality.

[rhettmaxwell]

Just to make sure I don't run into problems unnecessarily, Alpha5.3 is for Tiger only like all A5 versions, right?

I'm asking this because I have a few remote machines on Panther which I can't upgrade via Timbuktu. Or is there a way to do so?

[amrad9]

Rick

Thanks for giving Jackal and myself a bit more detailed explanation. I am basically a unix newbie but will try to help when I can.

Is there some reason as to why the alpha 5.2 and 5.3 have become more picky about having all the details set correctly. When I started using the optomizied workers I wasn't having to make sure the permissions were correct. I just replaced the original files restarted boinc and they seemed to start running ok. I also have not had to chmod the worker until the 5.3 one. I had a couple of machines that I couldn't get 5.2 to start running without errrors. I got these running when I saw the post about chmodding the worker for 5.3. Is this because of the need for the wisdom file that the worker needs to be chmodded?

I'm pretty good at figuring things out technically but, without the knowledge gained from the people posting in here I don't think I could have gotten the 5.3 optomized client running.

amrad9

[jedimstr]

Originally Posted by amrad9

... I got these running when I saw the post about chmodding the worker for 5.3. Is this because of the need for the wisdom file that the worker needs to be chmodded?

I'm pretty good at figuring things out technically but, without the knowledge gained from the people posting in here I don't think I could have gotten the 5.3 optomized client running.

amrad9

I didn't have to chmod the files in 5.3 to have them work correctly.

If you just downloaded the originally posted seti worker file and just edited your app_info.xml by hand, then Safari doesn't preserve permissions....therefore it necessesitates using chmod to reset them locally.

If however you downloaded the tarball (tar.gz) version which contains both the worker file and the appropriate app_info.xml provided on alexkan's site later on pg 18 of the thread: http://inst.eecs.berkeley.edu/~alexkan/seti/a53/ then you don't have to chmod because permissions are preserved in the tarball package just like all the other previous optimised client tarballs.

[liebsmaschine]

Just thought I'd update: everything's working fine now--client is processing WUs properly, etc. Average WU time is around 5000 seconds (anywhere from 4600 to 5700)--which seems to be roughly what it was before, although there may be a 10% increase in speed, perhaps. (My results history doesn't go back far enough to show the WUs I processed before upgrading clients--is there any other way to check my work history? I have a bunch of failed WUs in my results taking up all of the space due to the issues I had before getting it to work properly.) As I indicated before, I'm on a 1.67GHz PB 17" (low-res) with 1GB of RAM, so just wondering how that compares to some of the others with that machine (Alex, I believe you have a hi-res version of the same machine, but I think the core specs are pretty much the same--the RAM is a bit faster on yours, though, and that may really boost your S@h performance compared to mine.)

So far, I've processed 34 WUs since upgrading workers, and I've had 3 errors. So, there's an 8-10% failure rate. Just wanted to pass that on, since others have posted that.

One other thing--I haven't had the chance to keep a detailed eye on my BOINC Manager progress as it's working, but, for example, right now I have a S@h WU that's been going on for 1 hour and 4 minutes, and it's marked as being at 32.18%. Extrapolating that out would lead to at least a 2 and probably a 3 hour processing time on a WU (I know that certain parts of the FFT calculations take longer than others). I've seen that several times; however, a look through all of my results indicates that there are no out-of-the-ordinary processing times.

I wish I could take a video screen capture of my Work tab and see what happens as it processes that unit--will it really take 3 hours, or does it shoot from 30% to 90% in 10 minutes? I've got stuff to do and can't spend the time watching it, but has anyone else seen this happening?

[amrad9]

Originally Posted by jedimstr

I didn't have to chmod the files in 5.3 to have them work correctly.

If you just downloaded the originally posted seti worker file and just edited your app_info.xml by hand, then Safari doesn't preserve permissions....therefore it necessesitates using chmod to reset them locally.

If however you downloaded the tarball (tar.gz) version which contains both the worker file and the appropriate app_info.xml provided on alexkan's site later on pg 18 of the thread: http://inst.eecs.berkeley.edu/~alexkan/seti/a53/ then you don't have to chmod because permissions are preserved in the tarball package just like all the other previous optimised client tarballs.

Ok thanks for that info. At work I have downloaded the files on my computer and decompressed the packages. Then I copied the uncompresseed worker folder to a storage folder I have on our server. I have been copying the uncompressed worker folder off of the server rather than decompressing the tarball on the computer that I'm installing it on. So I'm guessing that copying the files on and off the server must be screwing up the permissions.

amrad9

[beadman]

Alex and Rick;
I finally got around to trying the 5.3 - installed in my PowerBook G4 http://setiathome.berkeley.edu/resul...hostid=2041993 and out of 21 completed so far, have had two invalid WU. In both cases, mine is the only invlaid result. I managed to download the two WU in question, but for some reason I can't access my FTP site to post them. Send me a private message with your email address and I'll forward both WU to you.

beadman

[gorbag]

Well, I guess I'm the first here to note: the new enhanced worker is out (as of May 3rd), and slowly being distributed to seti@home participants. See here: http://setiathome.berkeley.edu/sah_enhanced.php

So I guess we're back to square 2 (I would hope some of the enhancements made in the G5 worker made it into the baseline).

[beadman]

Jackal:
I too have a 1.67 GHz PB, 1 GB RAM, 15inch, running the 5.3 with BOINC Manager SuperBench - I'm averaging around 5000 seconds per WU now, and was getting around 5600 or so using 5.2.13. Computer is http://setiathome.berkeley.edu/show_...hostid=2041993. Out of around 24 or so finished WU, I've had two invalid.

beadman

[jedimstr]

Is it too early to hope for an Altivec optimized SETI Enhanced (SETI II) worker based on the official release (does the official release for the Mac have Altivec enhancements already just like the Windows/Linux versions have SSE enhancements built in)? And when it does become available, will there be a packaged tarball with both 5.3 and the new worker with a combined app_info.xml that has entries for both?

[Knightrider]

Originally Posted by gorbag

Well, I guess I'm the first here to note: the new enhanced worker is out (as of May 3rd), and slowly being distributed to seti@home participants. See here: http://setiathome.berkeley.edu/sah_enhanced.php

So I guess we're back to square 2 (I would hope some of the enhancements made in the G5 worker made it into the baseline).

Well spotted. It looks like they are getting stricter with the optimizations, which is ok as it will emphasise that there is nothing illigal or any cheating going on.

Originally Posted by Seti At Home

Will optimized versions be available?
Yes. The source code is available under the terms of the GPL, so people who want specifically optimized versions can build them and distribute them. Because SETI@home links with other libraries released under the GPL, we need to be more strict in how we enforce the terms of the license. People who distribute optimized versions must include copies of the files README, COPYING, COPYRIGHT, and AUTHORS with the binaries they distribute. If the distributors of optimized versions modify the source code of SETI@home, they must make the modified source code available as well. If you set up SETI@home Enhanced using the BOINC anonymous platform mechanism it is important that you use "setiathome_enhanced" as the application name.

K.

[tony.escobar]

I've been keeping tabs on the Correct Way to Migrate to Seti@Home Enhanced thread on the Berkeley forum. Anyone have thoughts on how we manage this process with Alex & Rick's app?

Or should we start a new topic for this?


TIA
QS

[Thanar]

So;

I take it that the new Seti@Home Enhanced clients for Mac OS X include some optimizations over the previous S@h official versions? Any common tricks between the new S@h-ench. clients and Alex' / Rick's highly optimized clients?

[lepetitmartien]

As It's a different beast altogether, I think a new thread would do us good. I'm wondering about the transition for us with optimized clients… :-/

[liebsmaschine]

Originally Posted by lepetitmartien

As It's a different beast altogether, I think a new thread would do us good. I'm wondering about the transition for us with optimized clients… :-/

Agree. This thread is becoming extremely unwieldy, too; starting over from scratch will do us well.

I vote to let Alex start the new thread. That way, he can edit the first post as needed to eventually post download links to the new client.

Also, it would be nice for someone knowledgeable to (as lepetitmartien said) go over how to smoothly transition for those of us with optimized workers. I'm hearing that we'll need to delete app_info.xml and the optimized executables and let BOINC download the stock worker. As it may be upwards of a month before everyone gets the new, enhanced app, that WILL most likely mean there will be a bit of crunching on the [slow] stock client.

The jury's still out for me, though, on whether BOINC will end up being able to download the enhanced app while app_info.xml specifies the app required for standard WUs. Again--someone out there knowledgeable in this?

Looking forward to an optimized enhanced app. It's going to be difficult (or at least different) to compare performance, though: I participate in S@h beta (which uses the same app as is now being distributed as "setiathome enhanced", so technically I already have it), and my result times go anywhere from 9,000 seconds to 270,000 seconds on a PowerBook 17" G4 1.67GHz (the WUs are not all the same size). But we'll let Alex and Rick handle that...

References:
Thread discussing how to upgrade optimized applications
My Seti@Home Beta results

[Thanar]

I have a bigfft_wisdom file from my iBook G4 @ 1.2 Ghz right here, if anyone is interested. That's for the alpha 5.3 worker, of course.

Didn't have the chance to send more than just one WU since installing the wisdom file, but it looks like I've dropped from 9100s to well under 7000 per WU. Interesting enough! You could take a look if you want to...

[Todd Madson]

My router hung this morning so I got a pretty good view of what my 2.5 was doing
as far as time devoted per block:

http://pod.ath.cx/alpha5.3.jpg

As you can see, lots of 34-35 minute work units. Nice. And this is with running
other software applications at the same time.

And yes, this thread is so long and unwieldy now I'm waiting for it to disintegrate.

As far as the enhanced client goes, I'm going to run this as long as I possibly can
before switching.

[[email protected]]

Hi, if I understand things correctly, I can now use the G4-a53 worker with a G4 and with a bigfft_wisdom file?

I ran the fft2_test to produce this:

projects/setiathome.berkeley.edu/fft_test2
Apple vDSP op / ip
1024 7456.540444 6452.775385
2048 7030.452419 6203.340370
4096 5878.148672 5333.154755
8192 2743.444126 2471.431252
16384 2125.620127 2539.254314
32768 889.251731 1170.503442
65536 555.479474 659.546575
131072 548.222339 492.382688
weighted time 26.430000 27.850000 20.810000
FFTW3 interleaved op / ip (this may take a minute)
1024 5785.246897 4329.604129
2048 4793.490286 2775.178586
4096 4628.197517 2720.629622
8192 4340.374289 2430.125994
16384 3020.977801 1990.517153
32768 1155.720964 994.696601
65536 853.530862 746.172220
131072 705.971960 692.684085
weighted time 19.900000 20.790000

How do I get a wisdom file from this now? I have a Mac mini G4 1.42Ghz

Thanks in advance

[Thanar]

The wisdom file was created by the fft2_test application to the folder the command was run from in the Terminal. If you did the test under single user mode, you should have mounted a volume first (see previous page about that).

Alternatively, do a spotlight rearch for "bigfft_wisdom", it should be there somewhere, and place it in the same folder as the a53 worker.

[liebsmaschine]

If Spotlight does not nab it (since its database is not necessarily instantly updated), try the following commands in Terminal:

sudo /usr/libexec/locate.updatedb
(wait a few minutes)
locate bigfft_wisdom

locate is unix's filename find program. The locate.updatedb executable updates the database that locate works from (it automatically updates weekly via a cron job, but we want to update it now to be sure, since the file is new). That will probably catch some things that Spotlight misses.

Note that locate only finds file NAMES, not contents like Spotlight does. For that, I've found an interesting script someone wrote calledgrepfind. (You'll have to download that as a text file, change the permissions so that it's executable, and then follow the instructions in the text file for how to use it.)

Supposedly bigfft_wisdom should be created when fft2_test runs. When I ran fft2_test under single-user mode, it locked up on me and never created the wisdom file. I did have some problems getting the drive to mount as writable--the prompt said to type "mount -uw," but that produced an error. After reading man mount, I tried mount -aw, and that made the drive writable, but fft2_test still locked up and didn't output a wisdom file.

I did not try running it in multiple-user mode, as it's now a moot point with S@h enhanced coming out. I'll look into it more after we get into that.

[lepetitmartien]

About the Enhanced, I'm trying at the moment to mod the app_info.xml file to use both the optimised 53 and the official Enhanced client. Right now it's not rocket science. I'll know this monday if it works or not as it does on other platforms.

(how to screw his rac big time lol

[lepetitmartien]

And running the moded xml now. Seems ok.

[rhettmaxwell]

Originally Posted by lepetitmartien

And running the moded xml now. Seems ok.

Please share!

[jedimstr]

Originally Posted by Bad to the bone

Please share!

I've got one in this thread:
http://forums.macnn.com/showthread.php?t=294782

[rhettmaxwell]

Originally Posted by jedimstr

I've got one in this thread:
http://forums.macnn.com/showthread.php?t=294782

Thanks much!

[lepetitmartien]

Sorry girls and boys, but due to my "little" problem, I couldn't share my xml… I didn't even have time to see if it worked in full… :-/ boohooo cry cry

[pubmkfly747]

Hi,

This thread is getting too long to review! Can someone tell me if there is a SETI client optimized for the G4 7410 that still displays graphics? I know it's corny, but I'd like to see the graphics. If not, I'll take a link to the latest optimized client without graphics. Thanks.

[Todd Madson]

See:
http://setiathome.berkeley.edu/resul...496&offset=620

That's from this morning. A stunningly unprecedented 15 blocks below 2000 seconds.

Then from late last night and early this morning:
http://setiathome.berkeley.edu/resul...496&offset=640

Another 11 all under 2000 seconds.

Alpha 5.3 definetely is fast on a G5 2.5 dual.

Too bad we're going to enhanced now. Sigh.