[beadman]

Alex:
How machine-specific are the wisdom files? For example, I have a 1.33 GHz iBook G4 - can I use a wisdom file created for a 1.67 GHZ G4 PowerBook?

beadman

[BTBlomberg]

I take it the wisdom is tuned to the processor and it's capabilities. I fhte iBook G4 and PB G4 are using the exact same CPU it may be OK. But keep in mind that just because the CPU model number may be the same that does not mean the processor model has not had revisions over the time between the two models came out. CPUs often get fixed along the way as they find glitches in the core and fabrication, so...

It would likely be best to just download the FFT_TEST3 and run it on each machine. That way you know for certain you have the best one of each machine.

[BTBlomberg]

OK, not this observation is not related to the client but to SETI itself:

I have been checking my results on SETI's site as most of us here do, but I have notice quite a few complete and valid results disappearing early from some of my computers results lists. Normally they are there for a couple weeks but some pop up fine just after the result is submitted and then a short time later you refresh and they are gone.

My IBM P3 550 has been cranking along at a good and gladly slower pace than my PB G4 500 for the last couple of weeks and has returned a couple dozen results at least. I looked at it last night and it only shows a result that was returned the day before, my last SETI (pre-enhanced result), my outstanding WUs in progress and the WUs I lost when changing to crunch3rs SETI client and BOINC 5.5.0. No other recent units that have been completed show.

Then on my PB G4 I have the same problem. I just complete at least 8 to 10 4 hour WUs (12.5-14.5 credits) and now my list only shows 3??? I also think some other longer ones vanished too.

Anyone else notice this?

It's not like they are too old. Older WUs from the end of may are showing still. Just wierd.

[Thanar]

Updated my online wisdom file using the latest test executable... wisdom file for iBook G4 @ 1.2.

The worker (v5, G4 opt.) works like a charm, btw... Keep up the good work, everybody!

[Gecko_r7]

Alex: I ran fft_test3, but need you to look at the results. It looks slower to me than fft_test_shell in most places unless I'm reading it wrong.
Higher numbers = faster, correct? Except weighted summary at bottom, lower is better?

http://homepage.mac.com/gecko_r7/.cv...t3.txt-zip.zip

I can't post the results here (from txt file) in a readable format. The boad is jamming everything togther.

Thanks!

[sdubz]

Originally Posted by BTBlomberg

OK, not this observation is not related to the client but to SETI itself:

I have been checking my results on SETI's site as most of us here do, but I have notice quite a few complete and valid results disappearing early from some of my computers results lists. Normally they are there for a couple weeks but some pop up fine just after the result is submitted and then a short time later you refresh and they are gone.

My IBM P3 550 has been cranking along at a good and gladly slower pace than my PB G4 500 for the last couple of weeks and has returned a couple dozen results at least. I looked at it last night and it only shows a result that was returned the day before, my last SETI (pre-enhanced result), my outstanding WUs in progress and the WUs I lost when changing to crunch3rs SETI client and BOINC 5.5.0. No other recent units that have been completed show.

Then on my PB G4 I have the same problem. I just complete at least 8 to 10 4 hour WUs (12.5-14.5 credits) and now my list only shows 3??? I also think some other longer ones vanished too.

Anyone else notice this?

It's not like they are too old. Older WUs from the end of may are showing still. Just wierd.

I think they are validating and "completing" their end faster now.

I had a g3 350mhz imac crunching for a while, and it would only show the last one completed because it was so slow!

[Todd Madson]

Latest update: with the latest wisdom generated and v5 I'm getting presently
two work units that top out around 1 hr and 43 minutes.

I haven't had a chance to try the "reference" work unit but we'll see how the
next two or three days goes, should be very interesting.

[alexkan]

Originally Posted by Gecko_r7

Alex: I ran fft_test3, but need you to look at the results. It looks slower to me than fft_test_shell in most places unless I'm reading it wrong.
Higher numbers = faster, correct? Except weighted summary at bottom, lower is better?

I see what you're talking about, and I have some idea why.

When I compiled FFTW for fft_test2 and the old optimized SETI clients, I used GCC 4, since I was making no effort to maintain backwards compatibility. However, the official SETI build system pays more attention to this, and uses GCC 3.3 when compiling for PPC to make sure binaries run on versions of OS X older than 10.3.9. Apparently, using GCC 3.3 instead of GCC 4 produces a noticeable speed difference. (Perhaps boog can attest to this.)

However, BOINC only runs on OS X 10.3 and above, so all this trouble seems to be for nothing, since there's no reason that anyone running 10.3 couldn't upgrade to 10.3.9, as far as I can tell. I'm going to recompile fft_test3 and the optimized client with GCC 4 and see if that makes a difference.

If you are in any way affected by this change, please speak up now. For now, rest assured that the weighted time is the only number that ultimately matters, and if you're looking at the MFLOP numbers, the only ones that make an appreciable difference are the ones where the first column (FFT size) are 16384 or more.

[Todd Madson]

Alex:

http://setiathome.berkeley.edu/resul...hostid=1876496

Last six: 6591.64, 6228.24, 2476.88, 2464.56, 2859.86, 9554.82.

My feeling is the most recent the 6591 and 6228 were done with the "new"
wisdom generated. I'm running 10.4 so no issues of compatibility.

[arkayn]

I generated my new wisdom file this afternoon so it will be at least a day before I see if there is any positive results.

[fran]

Originally Posted by Todd Madson

...
I haven't had a chance to try the "reference" work unit but we'll see how the
next two or three days goes, should be very interesting.

How does one try the reference unit? I would really like to give it a shot since everyone one of my boxes is different, and I want to see an apples to apples comparison.

TIA!

[Gecko_r7]

Originally Posted by alexkan

If you are in any way affected by this change, please speak up now. For now, rest assured that the weighted time is the only number that ultimately matters, and if you're looking at the MFLOP numbers, the only ones that make an appreciable difference are the ones where the first column (FFT size) are 16384 or more.

Thanks Alex!
I'm running 10.4, so good to go.
Thanks for the sanity check!
Looking forward to trying/testing the revision.

[halimedia]

Originally Posted by alexkan

I'm going to recompile fft_test3 and the optimized client with GCC 4 and see if that makes a difference.

I have not yet had time to try your universal compile of fft_test3, and will therefore wait for your GCC 4 compile. While on the GCC-subject, do you know if the stock intel mac worker was compiled w/ GCC 3 or 4? If the former, compiling it with GCC 4 might be an easy way to boost intel Mac performance by a little bit. Just a thought...

Cheers,

Ron

[brysonda]

Originally Posted by halimedia

While on the GCC-subject, do you know if the stock intel mac worker was compiled w/ GCC 3 or 4? If the former, compiling it with GCC 4 might be an easy way to boost intel Mac performance by a little bit. Just a thought...

I believe that the only version >= 4 of gcc can generate intel mac binaries.

[halimedia]

Weee! Forums up again!

I suppose you're right - not many reasons why Apple would have adapted GCC 3 to the Intel platform (except for highly backward compatible universal binaries, maybe)...

[gulliver]

A good prove how good the .v5 really is: My G5 Dual 2.7 GHz is 1/3 faster then a P4/3.4 GHz running the Crunch3r-app and the optimized 5.5.0 client!

Here is the result: http://setiweb.ssl.berkeley.edu/work...?wuid=81618404
340728696 633575 12 Jun 2006 1:45:32 UTC 14 Jun 2006 17:00:00 UTC Over Success Done 10,708.61 61.43 pending
340728697 48090 12 Jun 2006 1:46:14 UTC 12 Jun 2006 22:04:29 UTC Over Success Done 15,536.33 61.78 pending

Well done, alexkan!

[alexkan]

Originally Posted by gulliver

A good prove how good the .v5 really is: My G5 Dual 2.7 GHz is 1/3 faster then a P4/3.4 GHz running the Crunch3r-app and the optimized 5.5.0 client!

Here is the result: http://setiweb.ssl.berkeley.edu/work...?wuid=81618404
340728696 633575 12 Jun 2006 1:45:32 UTC 14 Jun 2006 17:00:00 UTC Over Success Done 10,708.61 61.43 pending
340728697 48090 12 Jun 2006 1:46:14 UTC 12 Jun 2006 22:04:29 UTC Over Success Done 15,536.33 61.78 pending

Well done, alexkan!

Let's not get carried away--that's not a fair comparison. That P4 3.4 GHz has HT, so that core is running 2 WUs simultaneously at about 15000 secs each. If you want to talk about overall throughput, then yeah, you win, but on a per-core basis, the P4 still has us beat.

[arkayn]

I figure that I average about 2 units a day on my eMac, but I seem to missing quite a few results for it.

The 2 dells do not seem to have that problem though.

[arkayn]

Maybe they are missing some as I have a gap between the 6th and the 9th.

[lepetitmartien]

Hi all, I'm unfortunately out of the game for 2 weeks as my G5 decided to have a permanent issue instead of an erratic one. It's at the repair shop and the G3 has already enough to do with me already

Seemingly my erratic freezes and now permanent ones are motherboard based, I'm still waiting for the diagnosis…

First time a desktop mac fails me and it's less than 3 years old.
/rant

I've just been able to recover a proper user account with keychains etc. Optimize this thing good, I'll be back to make it scream (well… as much as my G5 can) as soon as possible. Or maybe it'll be a macbook… or a mini… let's see…
/OT

[arkayn]

That definately sucks, at least I have 3 computers to keep up my crunching.

[Gecko_r7]

Alex: Is the fft_test3 link on your 6-11 post the same, or is this the recompiled version since then?

[alexkan]

Originally Posted by Gecko_r7

Alex: Is the fft_test3 link on your 6-11 post the same, or is this the recompiled version since then?

As of now, it's recompiled. I'm tired of compiling FFTW over and over with different GCC versions and compiler flags. Anybody want to buy me a faster machine to compile code on?

At this point in time, GCC 3.3 is producing faster code than GCC 4.0 where it counts (on the long FFTs), so the PowerPC side of things hasn't changed much, besides compiler flags. The binary is still Universal, though, and I see no real changes that need to be made to the x86 side of things. Post your benchmarks if you can get them to show up properly in the forums, so I can see how much of a difference the compile made, if any.

[Gecko_r7]

Originally Posted by alexkan

As of now, it's recompiled. I'm tired of compiling FFTW over and over with different GCC versions and compiler flags. Anybody want to buy me a faster machine to compile code on?

At this point in time, GCC 3.3 is producing faster code than GCC 4.0 where it counts (on the long FFTs), so the PowerPC side of things hasn't changed much, besides compiler flags. The binary is still Universal, though, and I see no real changes that need to be made to the x86 side of things. Post your benchmarks if you can get them to show up properly in the forums, so I can see how much of a difference the compile made, if any.

Thanks Alex. I'll run the same comparison tomorrow AM and post a link to the file. Wish I could help on that new machine. Our new baby is going to keep me on the G4 a bit longer as well.

[nikoniko]

For about two months the seti didn't ssend me any WU. By reading this thread I found that there is a new version.
I have installed the v5 and generated the sah file but I have noticed something
http://setiathome.berkeley.edu/show_...hostid=2248990
Measured floating point speed 1353.49 million ops/sec
Measured integer speed 4546.16 million ops/sec

I have a quad
In the previous version the floating point and integer speeds where x4, that is it was showing the total of the four cpus. Previously I was claiming aprox 65-95 credits for every wu that was doen in about 70 minutes. Now the WUs take much longer and the claim is 4-6 times less.
Can it be that because the machine is reported to handle 4 times less than it does that the claim is 4 times less;

br
Nikolas

[Knightrider]

Alex, here are the results of my fft 3 test.

Last login: Sat Jun 17 10:47:04 on console
/Applications/fft_test3; exit
Welcome to Darwin!
Knightrider:~ knightrider$ /Applications/fft_test3; exit
Apple vDSP op / ip / ipm
8 2581.110154 2720.629622 1750.666017
16 3890.368928 3947.580235 2767.375835
32 6847.843265 6847.843265 5592.405333
64 9151.208727 9151.208727 7895.160471
128 11184.810667 11457.610927 9994.937191
256 10956.549224 10956.549224 9761.289309
512 12326.117878 12582.912000 10981.450473
1024 11001.453115 11374.383729 10168.009697
2048 11184.810667 12511.822102 10397.147944
4096 10324.440615 10882.518486 9820.809366
8192 6100.805818 6231.537371 5894.697514
16384 6435.096548 6479.476524 6140.680366
32768 5470.831304 6252.378634 5655.241348
65536 2209.345317 6429.591760 4814.985758
131072 1491.308089 3034.177362 2535.223751
weighted time 0.008789 0.004199 0.005098
FFTW3 interleaved op / ip / ipm (this may take a minute)
8 2796.202667 2684.354560 1881.556935
16 2581.110154 2606.169476 2418.337441
32 5785.246897 2504.062090 2362.988169
64 7895.160471 3694.065908 3412.315119
128 9586.980571 5946.355038 5526.612329
256 10526.880627 7895.160471 7158.278827
512 8628.282514 7105.644424 6564.997565
1024 9192.995068 7989.150476 7294.441739
2048 9002.408585 6050.799213 5592.405333
4096 8753.330087 5921.370353 5631.513063
8192 6815.744000 3929.798342 3809.673502
16384 5659.783711 3997.974877 3819.203642
32768 4839.581538 2969.418761 2811.823911
65536 3076.624138 2753.184164 2491.280334
131072 2902.927959 2152.548468 1940.222259
weighted time 0.004844 0.006377 0.007061
logout
[Process completed]

K.

[Knightrider]

Originally Posted by nikoniko

For about two months the seti didn't ssend me any WU. By reading this thread I found that there is a new version.
I have installed the v5 and generated the sah file but I have noticed something
http://setiathome.berkeley.edu/show_...hostid=2248990
Measured floating point speed 1353.49 million ops/sec
Measured integer speed 4546.16 million ops/sec

I have a quad
In the previous version the floating point and integer speeds where x4, that is it was showing the total of the four cpus. Previously I was claiming aprox 65-95 credits for every wu that was doen in about 70 minutes. Now the WUs take much longer and the claim is 4-6 times less.
Can it be that because the machine is reported to handle 4 times less than it does that the claim is 4 times less;

br
Nikolas

The system for allocating credit has changed and now uses FLOP for calculation not mfpi and mis.

It is still a hot subject. Read this thread. There are probably other threads as well if you look around.

K.

[Gecko_r7]

Hi Alex,

Here 's the most recent fft_test3 test. Looks like @ 1.5%-2% difference vs. last one on G4.
Tests on the other two builds are also included for your reference.
All three tests done in SU mode.
How is does this stack vs. your tests/expectations?

http://homepage.mac.com/gecko_r7/.cv...06.txt-zip.zip

Cheers!

[arkayn]

I just set my eMac to 66% SETI and 33% Einstein, the PC will remain at 50% for both.

[Tyre MacAdmin]

What optimization would be best for a Dual 1.8GHz G4 7447A?

[Gecko_r7]

Originally Posted by Tyler McAdams

What optimization would be best for a Dual 1.8GHz G4 7447A?

From Alex Kan's 6.8.06 post 1 page back in this thread.

http://tbp.berkeley.edu/~alexkan/set...-ppc-v5-g4.zip

fft_test3 Wisdom from Alex's 6.11.06 post

http://tbp.berkeley.edu/~alexkan/seti/fft_test3.zip

Read this thread back 1 to 2 pages will explain how/what to do w/ the wisdom file.
For the optimized application:

Suspend Seti & close BOINC.
Go to your SYSTEM Library/Application Support/BOINC Data/projects/setiathome.berkeley.edu folder. TRASH or rename and move elsewhere, the standard Seti Enhanced 5.13 application and app_info.xml file.
Unzip the optimized ap and move contents into the same folder above.
Restart BOINC and S@H.
Crunch happy!
BTW, it should be @ 90-95% faster than the standard ap on your G4.

Alex does a heck of a job for us! Be sure to let him know if you like it.
Good luck!

[Tyre MacAdmin]

Originally Posted by Gecko_r7

From Alex Kan's 6.8.06 post 1 page back in this thread.

http://tbp.berkeley.edu/~alexkan/set...-ppc-v5-g4.zip

fft_test3 Wisdom from Alex's 6.11.06 post

http://tbp.berkeley.edu/~alexkan/seti/fft_test3.zip

Read this thread back 1 to 2 pages will explain how/what to do w/ the wisdom file.
For the optimized application:

Suspend Seti & close BOINC.
Go to your SYSTEM Library/Application Support/BOINC Data/projects/setiathome.berkeley.edu folder. TRASH or rename and move elsewhere, the standard Seti Enhanced 5.13 application and app_info.xml file.
Unzip the optimized ap and move contents into the same folder above.
Restart BOINC and S@H.
Crunch happy!
BTW, it should be @ 90-95% faster than the standard ap on your G4.

Alex does a heck of a job for us! Be sure to let him know if you like it.
Good luck!

I see where to put the new app and wisdom file but I do not see anyhting on how to create the .sah file from the wisdom file.

[Gecko_r7]

Originally Posted by Tyler McAdams

I see where to put the new app and wisdom file but I do not see anyhting on how to create the .sah file from the wisdom file.

No problem. This thread can be pretty wieldy and a more like a puzzle sometimes.
Just went through same processs on the main seti board recently. Try this:

http://setiathome.berkeley.edu/forum...d=31234#337018

Another way, potentially quicker.
Place fft_test3 in Applications directory.
Open Terminal, at the command prompt type "cd Applications" (w/o quotes), hit enter
Drag/drop fft_test3 into Terminal window and hit enter.
If you get a permission error, at the command prompt type: "chmod 755 fft_test3" & hit enter
At the command prompt, type: "./fft_test3"
Once complete (takes a bit and will return to command prompt when done), exit Terminal and Click on Applications directory.
Wisdom.sah should be there. Move to same folder as your opt. v5-g4 application.
All set.

[Tyre MacAdmin]

Originally Posted by Gecko_r7

No problem. This thread can be pretty wieldy and a more like a puzzle sometimes.
Just went through same processs on the main seti board recently. Try this:

http://setiathome.berkeley.edu/forum...d=31234#337018

Another way, potentially quicker.
Place fft_test3 in Applications directory.
Open Terminal, at the command prompt type "cd Applications" (w/o quotes), hit enter
Drag/drop fft_test3 into Terminal window and hit enter.
If you get a permission error, at the command prompt type: "chmod 755 fft_test3" & hit enter
At the command prompt, type: "./fft_test3"
Once complete (takes a bit and will return to command prompt when done), exit Terminal and Click on Applications directory.
Wisdom.sah should be there. Move to same folder as your opt. v5-g4 application.
All set.

I got "bus error" at the bottom of the test... I was booted in to single user mode. I don't know if that makes a difference or not... No wisdom file seen after I did another ls.

[alexkan]

Originally Posted by Tyler McAdams

I got "bus error" at the bottom of the test... I was booted in to single user mode. I don't know if that makes a difference or not... No wisdom file seen after I did another ls.

fft_test3 doesn't need to be run in single user mode. I haven't seen bus errors with fft_test3 before--what version of OS X are you running, and how far into the test did it go before bus erroring (was it right after the first line mentioning FFTW)?

Also, as a preemptive question, has anyone run fft_test3 on 10.3.9 without any trouble? I just want to make sure, since I put more effort than last time into making sure my clients were at least partly backwards-compatible with older releases.

[Tyre MacAdmin]

Originally Posted by alexkan

fft_test3 doesn't need to be run in single user mode. I haven't seen bus errors with fft_test3 before--what version of OS X are you running, and how far into the test did it go before bus erroring (was it right after the first line mentioning FFTW)?

Also, as a preemptive question, has anyone run fft_test3 on 10.3.9 without any trouble? I just want to make sure, since I put more effort than last time into making sure my clients were at least partly backwards-compatible with older releases.

I'm on the latest version of tiger 10.4.6. The bus error happens at the very end of the run... printed at the bottom line of the output.

[alexkan]

Originally Posted by Tyler McAdams

I'm on the latest version of tiger 10.4.6. The bus error happens at the very end of the run... printed at the bottom line of the output.

Looks like write permissions in whatever directory you're running fft_test3 in are causing the act of writing wisdom to bomb. I'd move fft_test3 to a directory you have write permissions for, and then try running it again.

Also, when/if your fft_test3 run finishes, please post the console output to this thread. I can't remember if I've seen numbers from a processor like yours. Don't worry about formatting or alignment, since I can read it fine either way.

Incidentally, what's the speed of your FSB and the size of your L2 cache?

[Tyre MacAdmin]

I re-ran the file in regular mode... it made the file just fine. I started the boinc and I'm getting 100% cpu activity but the graphics are not displaying the activity. Is that normal?

edit... nm it's showing the graphics now. Strange! Well whatever... I'll see what kind of difference it makes and post the results.

What do I have to do to join Team MacNN?

@alexkan
Here is the output of the fft file:
Apple vDSP op / ip / ipm
8 1043.142964 1094.166261 703.939133
16 2097.152000 2314.098759 1525.201455
32 4194.304000 4301.850256 2867.900171
64 6100.805818 6100.805818 4473.924267
128 7962.068610 8099.345655 6435.096548
256 8134.407758 8801.162492 6710.886400
512 9151.208727 9901.307803 7843.893195
1024 8715.436883 9192.995068 7803.356279
2048 8294.353978 9113.549432 7308.886178
4096 6710.886400 7973.330376 6340.995024
8192 3104.680541 3061.106077 2692.639605
16384 2485.513481 3834.792229 2366.559436
32768 576.206617 1548.666092 671.984619
65536 459.453070 702.710618 543.666746
131072 313.851634 387.912509 341.879139
weighted time 0.042334 0.032891 0.038418
FFTW3 interleaved op / ip / ipm (this may take a minute)
8 805.306368 808.540530 671.088640
16 983.280059 976.128931 877.240052
32 3226.387692 1465.259039 1369.568653
64 4575.604364 2532.409962 2262.096539
128 5659.783711 3819.203642 3428.920058
256 6100.805818 4709.393965 4227.330016
512 5298.068211 4408.611504 4136.847781
1024 5284.162520 4827.975827 4168.252422
2048 4732.035282 3482.063698 3195.660190
4096 4353.007395 3133.487813 2876.094171
8192 3744.271382 2097.152000 1727.554915
16384 1651.184703 1842.204110 1192.289462
32768 852.356444 1347.567550 813.769572
65536 545.046611 548.107108 418.123763
131072 452.001065 386.860186 358.307377
weighted time 0.030273 0.034131 0.038252
Thaidogs-G4:/Applications tylerm$

The cache is 512kb full speed and the fsb is 133mhz

[arkayn]

Originally Posted by Tyler McAdams

What do I have to do to join Team MacNN?

Go to this link and click on join team.
http://setiathome.berkeley.edu/team_...p?teamid=30246

[Tyre MacAdmin]

Originally Posted by arkayn

Go to this link and click on join team.
http://setiathome.berkeley.edu/team_...p?teamid=30246

I'm in! thanks for all the help! It looks like my G4 is now crunchig just as fast as my 3Ghz P4!

[bobpalmer]

Alex,
Here's the output from fft_test3 on a Quad
Apple vDSP op / ip / ipm
8 2548.437873 2757.898521 1781.651257
16 3834.792229 3947.580235 2767.375835
32 6710.886400 6710.886400 5500.726557
64 9151.208727 8947.848533 7743.330462
128 11457.610927 11457.610927 9786.709333
256 10737.418240 10737.418240 9586.980571
512 12582.912000 12582.912000 10981.450473
1024 7540.321798 11374.383729 10485.760000
2048 11534.336000 9464.070564 8112.060484
4096 10324.440615 10882.518486 9702.486361
8192 5975.446795 6231.537371 5894.697514
16384 6348.135784 6391.320381 6022.590359
32768 4357.718442 6100.805818 5470.831304
65536 2434.788717 4948.119005 3976.821570
131072 1289.096823 2295.474221 1960.224550
weighted time 0.009883 0.005508 0.006504
FFTW3 interleaved op / ip / ipm (this may take a minute)
8 2757.898521 2684.354560 1864.135111
16 2485.513481 2485.513481 2334.221357
32 5687.191864 2449.228613 2236.962133
64 7743.330462 3947.580235 3660.483491
128 9586.980571 5872.025600 5399.563770
256 10129.639849 7354.396055 6795.834329
512 8628.282514 7190.235429 6226.595629
1024 9192.995068 8085.405301 7294.441739
2048 9002.408585 6001.605724 5635.095450
4096 8659.208258 5965.232356 5592.405333
8192 6815.744000 3894.710857 3776.689316
16384 5559.314178 3997.974877 3819.203642
32768 4617.582385 2952.002815 2851.651445
65536 3103.300069 2803.503457 2462.710606
131072 2895.560122 2210.950946 1994.494210
weighted time 0.004854 0.006221 0.006904

[ChillyWilly5280]

So I FINALLY managed to get into the forums and I would like to take this opportunity to thank Alex & Rick for their optimized clients - they have been great in both the previous and current versions of SETI. And also to those that have given them the feedback needed to make the apps what they are. Thanks guys, you've done a fantastic job!

Also, a thanks to Gecko_r7 for walking me through the process of running fft_test3 in single user mode over at the SETI forums, and to boog for getting the ball rolling!

Alex, I am running 10.3.9 and as far as I know fft_test3 ran w/o any problems. WU's crunch seemingly ok and validate, is there something in particular I should look at/for? I ran it on three different 7450's, all on 10.3.9.

[alexkan]

Looks like I was wrong when I said that there wouldn't be much in the way of speed gains in versions after v5...

v6:
G4 version
G5 version

(Even though I've been posting on the main SETI boards, I haven't forgotten my roots--you guys get the jump on the non-MacNN people by...oh, 5 minutes or so. )

I won't try to guess what kind of speed gains you guys will get out of this one, since my optimizations probably affect certain angle ranges much more than others, but I'm pretty sure that you guys'll be able to see the difference. Oh, and this is linked with the (slightly) faster FFTW compile I rolled for fft_test3, so that doesn't hurt, either.

[Gecko_r7]

Originally Posted by alexkan

Looks like I was wrong when I said that there wouldn't be much in the way of speed gains in versions after v5...

(Even though I've been posting on the main SETI boards, I haven't forgotten my roots--you guys get the jump on the non-MacNN people by...oh, 5 minutes or so. )

Thanks Alex!
Feels like X-mas whenever you release a new version to us.
I'll run and post the results of the Ref WU against G4 later this evening.
BTW, are we still running quicker than the new penguin build CofA has been working on?

[brysonda]

Definitely a big thanks Alex.

I have transitioned all of my 11 systems over to v6. Also regenerated the widsom file on the quad using the new fft_test3.

Hopefully the group of quads can continue to overtake those PCs in the top spots.

[arkayn]

I will download the new build once I get back home to the Grand Canyon, for now I only have 2 systems processing as I have the laptop with me.

[BTBlomberg]

Shoot Alex. Boinc just gets calibrated to give a somewhat accurate guess at WU completion times and then you do this. I do not mind at all though. Looks like the G4 client had a bigger gain that I would have expected.

I installed V6 on my PBG4 500 with a WU that was 26.667% complete. It had gone 3:54:13 and said it had 14:09:15 to go with V5. After 34 Minutes I snapped a pick of the data and tryed to make some estimates (given that some operations later may run faster or slower than the current one). First of all the Estimated time lest was counting down faster than the clock for CPU time (always good) at a rate of 40% faster. Now that is not a good basis, but with other calculations it looks like this WU that was estimated to complete in 18:03:28 Hours would complete in 14:14:26 Hours if at the pace it is at now. This would be roughly a 31-27% increase. Very nice in my book, especially for something that could not be speeded up any more

Time will tell how fast it really is but it looks good right out of the gate. I bet those high speed G4s will see hard numbers way before me. Remember guys not all WUs are create equal so you can't just go by the reported time to say it's fast, it needs to be in the same Point/Estimated time category. The estimated time indicates the weight on it often.

[mac1896]

Originally Posted by Tyler McAdams

I'm in! thanks for all the help! It looks like my G4 is now crunchig just as fast as my 3Ghz P4!

Nice to see another 1.8GHz duallie on the board !

What model G4 are you running it in ?

[Gecko_r7]

The latest Ref WU times for Alex's v6-G4 compile on 03' MDD w/ Giga DP 1.33
BTW, all runs are w/ latest wisdom.sah

AlexKan v6-G4 (Test5 6.21.06)
real 225m56.512s (= 13556.512s)
user 225m23.616s
sys 0m30.942s
wu_cpu_time = 13497.792725

AlexKan v5-G4 (Test5 6.9.06)
real 258m57.563s (= 15537.563s)
user 258m24.204s
sys 0m31.797s
wu_cpu_time = 15298.139857

v6-G4 vs.v5-G4

real: +15% faster
cpu: +13% faster

Stock 5.13 Worker (Test1 5.24.06)
real 499m12.054s ( = 29952.054s)
user 498m11.342s
sys 0m57.878s
wu_cpu_time = 29915.352618

v6-G4 vs. Standard 5.13 worker

real: +121% faster!
cpu: +122% faster!!!

I think w/ v6 the G4 1.33 will now crunch a WU faster than my M-1.6 Banias w/ Crunch3r ap! Got to crunch a few more WUs to find matching ARs to compare the two. This should be fun! Nice work Alex.

[BTBlomberg]

Update on note above. Alex this is going even better than my first calculations. After running 4:27:00 Hours with V6 on this WU it appears it will be done in about 13:30:47 with 26.667 done in V5. This is compared to 18:03:28 with all V5. Estimate based on current progress would suggest this WU in V6 only would complete in 11:10:40 Hours. This points to an increase I discarded in calculations earlier of a 38.5% increase on this G4 over V5. Now it maybe the WU or something, but if not this is great. I look forward to see if others confirm this type of increase on G4.

My PBG4 500 was already beating my IBM P3 550 with Cruch3r's last client. This seams to take us back to the days before enhanced with this improvement.