[Gulliver64]

Originally Posted by halimedia 

Question: Is there a custom-generated wisdom-file present on the machine you have linked to? That may explain why the differences are greater on your machine than I observed during my benchmarking runs...

Curiously,

Ron

Ron,

I use the wisdom file that was posted for the G5 in this thread. The latest result was slightly faster than the previous 2.

[halimedia]

Originally Posted by Gulliver64 

I use the wisdom file that was posted for the G5 in this thread. The latest result was slightly faster than the previous 2.

Interesting! May I suggest you let your G5 run the way it is until you have a representative set of v8-results, and that you then temporarily remove the wisdom file to see how the situation changes? That would be helpful!

TIA,

Ron

[Gulliver64]

Originally Posted by halimedia 

Interesting! May I suggest you let your G5 run the way it is until you have a representative set of v8-results, and that you then temporarily remove the wisdom file to see how the situation changes? That would be helpful!

TIA,

Ron

Ron,

interesting development: Just finished the first longer WU (7,400 seconds). This one was faster on a credits/hour basis than the 7.1 results were. I noticed this earlier but did not think it was relevant: When changing the worker-app the first 1 or 2 WUs are much slower than later ones. Could this be due to cache handling/use?

[adream]

weh hey !!

many thanks alex and co for releasing the new clients !

much appreciated,

sorry about whinging the other day in my last post a bit, but i thought i was just reminding you that there where eager crunchers out there chomping at the bit ..

all loaded up on my 8-core and lets see what it can do.....great !

regards

adream

[arkayn]

Installed on my C2D iMac and crunching away.

[arkayn]

Posted in the optimized apps thread on SETI.

[alexkan]

Originally Posted by Gulliver64 

interesting development: Just finished the first longer WU (7,400 seconds). This one was faster on a credits/hour basis than the 7.1 results were. I noticed this earlier but did not think it was relevant: When changing the worker-app the first 1 or 2 WUs are much slower than later ones. Could this be due to cache handling/use?

Actually, I'm more interested in the fact that you're seeing the new apps being slower than the old apps primarily on shorter WUs, which points to some sort of difference in the FFTW libraries between the two builds. This also fits with your observation that v7.1mb is also slower than v7.1, since v7.1mb was built against the same FFTW library as v8.

Have you made modifications to the app_info.xml file I've provided to copy the old wisdom file into your WU directories? If you could find some way to attach the wisdom file you're using and tell me more about your system setup, it could be useful for diagnosing the problem.

[Gulliver64]

Originally Posted by alexkan 

Actually, I'm more interested in the fact that you're seeing the new apps being slower than the old apps primarily on shorter WUs, which points to some sort of difference in the FFTW libraries between the two builds. This also fits with your observation that v7.1mb is also slower than v7.1, since v7.1mb was built against the same FFTW library as v8.

Have you made modifications to the app_info.xml file I've provided to copy the old wisdom file into your WU directories? If you could find some way to attach the wisdom file you're using and tell me more about your system setup, it could be useful for diagnosing the problem.

Hi alex,

I just installed v8 on my second G5. Let's see how it behaves there.

No, I have not modified the app_info.xml. On this second G5 I do not use a wisdom file at all.

[Gulliver64]

O.k., now here is what I see after 5 results on my seconds G5 (Dual 2.7GHz): As with my other G5 (Dual 2.5GHz) the first 2 results with v8 are appr. 5-10% slower than with the original 7.1, all other results are appr. 10% FASTER. Very interesting ...

[fortilan@mac.com]

> Re getting the latest enhanced optimized SETI running on my Mac Pro.

Thanks QSilver

It worked, but I just cant seem to get the install right... every time I change workers, I seem to need to reinstall BOINC and when I do it launches itself immediately and cant find the right worker and I need to muck around to get it to load the right one

Is there a simple stepwise way if upgrading the worker? (I looked back through many previous posts and have not spotted one... yet)

[arkayn]

The 2 files that you have to put in the folder are the application itself and the app_info.xml file. The second one tells the Boinc Manager to use the correct application.

[fortilan@mac.com]

Originally Posted by arkayn 

The 2 files that you have to put in the folder are the application itself and the app_info.xml file. The second one tells the Boinc Manager to use the correct application.

Thanks - I guess its just 'permission' issues I get every time I copy in something new.

Another quick question, I am now using V8 --- is a wisdom file needed? or will it make things even better? I seem to recall a program to generate the right wisdom file. If I do need one, where can I get the generator program? I use an 8 core Mac Pro.

[beadman]

Just installed v8 in my core duo MBP; will keep you advised. Now off to put it on my iBook G4.

beadman

[beadman]

Installed V8 on my iBook G4 without a problem - picked up on the work in progress with troubles. Did have the permissions problem when I tried to start BOINC, but on both the MBP and the iBook, once I re-installed BOINC, it started right up.

beadman

[beadman]

First unit completed, started with 7;1MB and finished with v8 G4. Had an error in the results:
<core_client_version>5.10.20</core_client_version>
<![CDATA[
<stderr_txt>
MacOS Error -5000 occured in /Users/alexkan/seti/boinc/api/mac_icon.C line 107
MacOS Error -5000 occured in /Users/alexkan/seti/boinc/api/mac_icon.C line 107
OS X optimized S@H Enhanced application by Alex Kan
Version info: OS X Altivec (G4/G5, G4-optimized v7.1mb-nographics) V5.13 by Alex Kan

Work Unit Info:
...............
Credit multiplier is : 2.85
WU true angle range is : 0.840107

Flopcounter: 9403489698861.500000

Spike count: 1
Pulse count: 0
Triplet count: 0
Gaussian count: 0
</stderr_txt>
]]>

WU was valid, and I received 31+ credits.

[Gecko_r7]

Install went fine on iMactel T7200. Had to do the BOINC-reinstall as well. No big deal.
VERY nice speedup

AR=.407423 Credit= 54.24

V7.2mb = 5734.13
V8 = 5104.82 11% faster!

Pretty kickin' for only 2.0Ghz saddled w/ rather pokey DDR2 667 CL 5-5-5-15 memory

Nice work Alex!

[alexkan]

Originally Posted by fortilan@mac.com 

Another quick question, I am now using V8 --- is a wisdom file needed? or will it make things even better? I seem to recall a program to generate the right wisdom file. If I do need one, where can I get the generator program? I use an 8 core Mac Pro.

No wisdom files are needed if you're on an Intel Mac. Even if you're on PowerPC, wisdom files from the past should probably not be used with v8 builds because of version differences between the FFTW libraries in those builds and the fft_test applications.

Originally Posted by beadman 

First unit completed, started with 7;1MB and finished with v8 G4. Had an error in the results:

<snip>

WU was valid, and I received 31+ credits.

From the looks of things, that WU was never touched by the v8 app, since it would have announced its presence in the stderr output. The -5000 error you saw should be familiar to a lot of people by now, as it's what happens when the SETI app fails to update its icon because of the file permissions used to keep BOINC and its apps sandboxed. You shouldn't see this anymore, since I made sure that every app already had its icon before I packaged the releases.

On a related note...Thanar, if you're out there, have you tried running this latest crop of apps? I'm curious if they solve the issue you were seeing with running apps at less than 100% CPU utilization.

[Elphidieus]

Performance on a Mac Pro 8-core with only 3GB RAM (what an embarrassment)....

Result based on crunching of 4 simultaneous cores

WU Credit= 54.26

V7.2 = 5800-6000s
V8 = 3800-4000s

Improvement = 33% faster

Now that's a v8 on a V8....

[fortilan@mac.com]

Hi,

is anyone else getting lots of 'Computation errors' running the new v7 and v8 workers?

I was getting them before on v7 then (I think) they went away, now I'm getting them on v8 as well. The units processed are faster than before but many are computation errors after a few mins/secs. Most with a 'missing output file message' (sorry dont have one to show) but some with this message...

"process exited with code 250 (0xfa, -6))"

*** Update ***

The results for most of these errors contains the following:

<core_client_version>5.10.7</core_client_version>
<![CDATA[
<message>
Can't create shared memory: system shmget
</message>
]]

any one else seeing this? or know what it is?

[halimedia]

Originally Posted by fortilan@mac.com 

Hi,

is anyone else getting lots of 'Computation errors' running the new v7 and v8 workers?

I was getting them before on v7 then (I think) they went away, now I'm getting them on v8 as well. The units processed are faster than before but many are computation errors after a few mins/secs. Most with a 'missing output file message' (sorry dont have one to show) but some with this message...

"process exited with code 250 (0xfa, -6))"

I've received the following error quite frequently as of late: 'SETI@home error -6 Bad workunit header' - could that be the same? (using BOINC 5.10.7 and .20 CLI here). There appear to be quite a few corrupt WUs like that out there at the moment...

A good way to check whether errors are caused by some problem with your particular configuration or by the WU itself is to look at the quorum results of the WU in question. If the WU fails for all quorum members, it's most likely that the WU's at fault.

HTH,

Ron