[Thanar]

Finally, the WU exited again, twice, so I reset the project. I got the exact WU back from the server. If it does it again, I will switch back to the stock client which works fine under 50% CPU (I just can't listen to the fans revving at maximum in my work place) and wait for Alex to compile a new optimized one from a more recent CVS.

[Odysseus]

Originally Posted by kennet 

Index of /~alexkan/seti

Here are the new Alex Kan Seti Enhanced nographics versions 7.2. I take it that the mb in the version name means multibeam?

I do not know if they process multibeam or not but these were made available the 21 and 22 of August 2007, as indicated at Alex's download site.

Any S@h Enhanced version can process Multibeam data; note that there’s been no jump in version numbers since last summer. AFAICT most of the development effort for Multibeam was server-side, especially in retooling the splitters to accommodate the new data format, but I guess also in handling & interpreting the results. The only obvious differences in the recent science apps, aside from the continuing, non-CPU-specific performance tweaks, are a new reporting item or two, in the text output and the screensaver display, not affecting crunching per se, and an adjusted credit claim. (In the bizarre case of v5.23, one might better say “maladjusted”. )

[Gecko_r7]

Alex,
Seeing a strange error in stderr report.

CPU time 4501.188005
stderr out
<core_client_version>5.10.7</core_client_version>
<![CDATA[
<stderr_txt>
MacOS Error -5000 occured in /Users/alexkan/seti/boinc/mac_build/../api/mac_icon.C line 107
MacOS Error -5000 occured in /Users/alexkan/seti/boinc/mac_build/../api/mac_icon.C line 107
OS X optimized S@H Enhanced application by Alex Kan
Version info: OS X SSSE3 (Intel, Core 2-optimized v7.2mb-nographics) V5.13 by Alex Kan

Work Unit Info:
...............
Credit multiplier is : 2.85
WU true angle range is : 0.601678

Flopcounter: 12938162691841.638672

Spike count: 0
Pulse count: 1
Triplet count: 0
Gaussian count: 0
</stderr_txt>
]]>
Validate state Initial
Claimed credit 42.6944023978573
Granted credit 42.6942978617936
application version 5.27

Doesn't appear to affect processing or validation.

lib issue?

[Knightrider]

Seeing a strange error in stderr report.

I am getting this as well, but I also had it with the V7.1 as well. Alex is aware of it, but I don't know what causes it. It seems to have no adverse effect.

K.

[Odysseus]

Originally Posted by Gecko_r7 

Seeing a strange error in stderr report.

See upthread. I just installed the new “mb” app—thanks, Alex!—on my G5 running BOINC v5.4.9, but it hasn’t got any work yet; I’m expecting not to see the message because the icon problem doesn’t seem to affect clients of this vintage. (It may even have been fixed in an alpha version by now.)

[Gecko_r7]

Originally Posted by Odysseus 

See upthread. I just installed the new “mb” app—thanks, Alex!—on my G5 running BOINC v5.4.9, but it hasn’t got any work yet; I’m expecting not to see the message because the icon problem doesn’t seem to affect clients of this vintage. (It may even have been fixed in an alpha version by now.)

Thanks. One of the disadvantages to this long thread. Little gems like this get lost.
I ran 5.4.9 until a few weeks ago. Never happened w/ it.
Interesting also that the V8 pre-releases didn't show this w/ 5.10.7, but the revised 7.2 displays the error.
Thanks for the heads-up Odysseus.

[Todd Madson]

With Boinc Manager 5.50 and the new Alex Kan optimized G5 bits I'm getting
computation errors with every downloaded work unit. Ideas?

Interestingly, it's working fine on my G4.

[Thanar]

An update:

I think I am giving up on Alex' builds, since on my Mac Mini I had another failed task:
Sat Aug 25 23:50:37 2007|SETI@home|[file_xfer] Started download of file 11fe07ad.17065.3748.10.5.142
Sat Aug 25 23:52:16 2007|SETI@home|[file_xfer] Finished download of file 11fe07ad.17065.3748.10.5.142
Sat Aug 25 23:52:16 2007|SETI@home|[file_xfer] Throughput 3828 bytes/sec
Sat Aug 25 23:52:17 2007|SETI@home|Starting 11fe07ad.17065.3748.10.5.142_1
Sat Aug 25 23:52:17 2007|SETI@home|Starting task 11fe07ad.17065.3748.10.5.142_1 using setiathome_enhanced version 527
Sun Aug 26 02:34:39 2007|SETI@home|Restarting task 11fe07ad.17065.3748.10.5.142_1 using setiathome_enhanced version 527
Sun Aug 26 03:21:52 2007|SETI@home|Restarting task 11fe07ad.17065.3748.10.5.142_1 using setiathome_enhanced version 527
Sun Aug 26 04:30:30 2007|SETI@home|Restarting task 11fe07ad.17065.3748.10.5.142_1 using setiathome_enhanced version 527
Sun Aug 26 06:20:18 2007|SETI@home|Restarting task 11fe07ad.17065.3748.10.5.142_1 using setiathome_enhanced version 527
Sun Aug 26 09:19:35 2007|SETI@home|Task 11fe07ad.17065.3748.10.5.142_1 exited with zero status but no 'finished' file
Sun Aug 26 09:19:35 2007|SETI@home|If this happens repeatedly you may need to reset the project.
Sun Aug 26 12:25:54 2007|SETI@home|Task 11fe07ad.17065.3748.10.5.142_1 exited with zero status but no 'finished' file
Sun Aug 26 12:25:54 2007|SETI@home|If this happens repeatedly you may need to reset the project.
Sun Aug 26 15:14:45 2007|SETI@home|Restarting task 11fe07ad.17065.3748.10.5.142_1 using setiathome_enhanced version 527
Sun Aug 26 16:19:12 2007|SETI@home|Restarting task 11fe07ad.17065.3748.10.5.142_1 using setiathome_enhanced version 527
As you can see, lots of restarts and exits.

However, on my iBook G4, a WU completed successfully, although the one currently under process is 14 hours in computation which -according to BOINC manager- means 0.01% complete! Huh?

Switching mack to the stock worker for moth G4s, waiting for a new optimized client based on earlier CVS source version. Mind you, the above issues occur ONLY when CPU usage percentage is set to any lower than 100% value, through BOINC manager's preferences.

[Gulliver64]

Originally Posted by Odysseus 

See upthread. I just installed the new “mb” app—thanks, Alex!—on my G5 running BOINC v5.4.9, but it hasn’t got any work yet; I’m expecting not to see the message because the icon problem doesn’t seem to affect clients of this vintage. (It may even have been fixed in an alpha version by now.)

Odysseus,

as far as I understood alex this new version for G4/G5 is NOT multi-beam. It only corrects the multiplier. Version 8 will be MB.

[Todd Madson]

I don't get it. It didn't work at all yesterday (see errors) but it appears to be working today.

Sun Aug 26 09:48:07 2007||Rescheduling CPU: files downloaded
Sun Aug 26 09:48:07 2007||Rescheduling CPU: files downloaded
Sun Aug 26 09:48:07 2007|SETI@home|Starting task 11fe07ae.27849.20522.14.5.185_0 using setiathome_enhanced version 527
Sun Aug 26 09:48:07 2007|SETI@home|Starting task 11fe07ae.27639.21749.12.5.161_0 using setiathome_enhanced version 527
Sun Aug 26 09:48:08 2007|SETI@home|Unrecoverable error for result 11fe07ae.27639.21749.12.5.161_0 (process exited with code 2 (0x2))
Sun Aug 26 09:48:08 2007|SETI@home|Deferring scheduler requests for 24 minutes and 54 seconds
Sun Aug 26 09:48:08 2007||Rescheduling CPU: application exited
Sun Aug 26 09:48:08 2007|SETI@home|Computation for task 11fe07ae.27639.21749.12.5.161_0 finished
Sun Aug 26 09:48:09 2007|SETI@home|Unrecoverable error for result 11fe07ae.27849.20522.14.5.185_0 (process exited with code 2 (0x2))
Sun Aug 26 09:48:09 2007|SETI@home|Deferring scheduler requests for 1 hours, 9 minutes and 5 seconds

Any ideas on what's happening here and how to fix?

[kennet]

When I switched from 523 Boinc Client to Alex's optimizer and 527 the client forced the resending of the previously downloaded workunits (had to reload the Boinc Client so that Alex's optimizer would work). Those previously downloaded workunits in 523 were multibeam and they were restarted with the optimizer.

[Odysseus]

Originally Posted by Gulliver64 

as far as I understood alex this new version for G4/G5 is NOT multi-beam. It only corrects the multiplier. Version 8 will be MB.

That’s my understanding as well; AFAICT v7.2 is still based on the stock v5.13 codebase. My point was that it can still process tasks from Multibeam WUs, and its results should still validate against those of the stock app.

[Todd Madson]

Machines with the MB version:

Macs:
G5 2.5 dual
G4 400

PCs:
Linux Box with Athlon 2400 with Chicken Good 2.2 (checking, it's 2.4 actually)
Pentium M 1.7 Windows XP Box

Observations:
So far it looks like my RAC is dropping like a rock. Not sure if this
is a seti thing or just the effect of the corrected credit system with
the new clients or not.

I've got at least one Mac that needs to be updated to the MB
version that I haven't had the time to get to and theres' another
x86 system that needs to get updated but so far the results
have been curious.

Seems like the time for work units to crunch is longer but
looking at the records for the machines doesn't bear that out - it
seems slightly sooner but I have a lot of pending credit.

[arkayn]

Alex has just released a new version of his app...

i386-v7.2mb1-coreduo-nographics
i386-v7.2mb1-core2-nographics


* pulse-finding fixes to address weakly similar WUs


source code

[kennet]

Thanks muchly

Kenn

[BTBlomberg]

It appears that SETI stats are just dropping with the new WUs. You can have a good WU that takes several hours get 60+ Credits and then have one that runs 5 times as long to error out and get 0.XX credits. Looks like there are a lot of noisy WUs out there and the effect is you can work a long time on junk and get no credit for it. This is not just Mac and not just optimized Clients. I have seen it with 2.4 and 2.2 clients as well as stock all failing with excessive time on a WU.

I also am seeing the Pending credits rising. I went from just a couple hundred a couple weeks ago to nearly 4000. So my machines are cranking them out, they are just sitting there waiting to be confirmed.

On the other hand I am also seeing 100+ Credit WUs from time to time.

My feeling is that the MB WUs are a little junky right now and I think the tech bulitin at Seti confirmed this at one point. SO I am hoping they work throught them adn get things teeked so that we can get back up there.

Going from nearly 4000 credits a day to 2000 credits a day sucks.

[halimedia]

Originally Posted by arkayn 

Alex has just released a new version of his app...
i386-v7.2mb1-coreduo-nographics
i386-v7.2mb1-core2-nographics
* pulse-finding fixes to address weakly similar WUs
source code

Kudos to Alex for releasing this fix so quickly!

Since installing these new builds, my pending credit is decreasig instead of increasing. I'd highly recommend anyone install these new builds. While it's rare that WUs fail or receive no credit with Alex' first Intel MB-builds, the weak similarty of many results provoke SETI to send out a third WU per quorum. This results in the increasing pending credit situation many have experienced, puts unnecessary load on the SETI servers, and reduces project efficiency over-all.

Thanks again, Alex - and thanks to arkayn for posting this here!

Cheers, Ron

[gorbag]

I certainly hope pending credit starts to decrease (I just updated all my x86 workers). I've got over 31K pending!

[halimedia]

Originally Posted by gorbag 

I certainly hope pending credit starts to decrease (I just updated all my x86 workers).

No worries, it will

I've got over 31K pending!

My max was 60k, and I'm down to 49k as of this writing. I installed the updated workers a little less than two days ago.

Cheers, Ron

[arkayn]

I am down to 4300 from over 6000 a few days ago. Still only using 2 computers at the moment.

[Todd Madson]

Okay this is weird. Running Alex's MB on the G5.

One work unit is done at 100%.

CPU time: 23 hours, 23 minutes, 5 seconds.

Another is in process:

CPU time: 30 hours, 35 minutes, 39 seconds, remaining 87 hours, 18 minutes, 09 seconds (progress 0.042%).

What the living heck is going on?

The first work unit is 04mr07ab.10282.4980.3.4.228_7
The second is 04mr07aa.8827.17659.12.4.192_4

87 HOURS? 23 hours?

This is a G5 dual 2.5 ghz model with 3.5 gig of ram.

Holy moly batman! I have two other work units done, one was
complete in 2 hours 51 minutes and 6 seconds,
the other 2 hours 49 minutes and 17 seconds.

Yipes.

Either I have a real alien transmission or something is really weird in Setiland.

[BTBlomberg]

Todd,

These are likely noisy WUs that do not fail, but just keep running and running. I have seen these on G4 and Intel PCs. It's not just Alex's client for Mac. THough my G4 1500 PowerBook had the worst one with 23 hours and stuck at 15%. Most are 3 to 9 hours.

A lot of WUs seam to get -9 too many results type errors on return in all clients. I think they just have a bunch of junk on there tapes and we have to wait through it for some good stuff.

[arkayn]

I had one that was at 14 hours and only 0.045%. I finally aborted that one.

[kennet]

Todd:

It is the noisy WU. I had one a week ago. I aborted the WU and have had no repeat of the situation.

Kenn

[BTBlomberg]

Here is another one I aborted this morning. 18+ Hours for .038%

Workunit

The page offers the note "Too many error results" but has sent it out 8 times.

There was an avarage of 46.8 hours wasted on this WU on the other clients that returned this. I think the clients are not failing right. For some reason they keep trying to run bad WUs or something else is up with the WUs that send things off. My thoughts are bad WUs from new system initial setup. I am hoping they clear up with time or corrections by the SETI team.

[QSilver]

Originally Posted by kennet 

Todd:

It is the noisy WU. I had one a week ago. I aborted the WU and have had no repeat of the situation.

Kenn

@Todd
There's been plenty of dicussion of these noisy / neverending WUs over at the S@H Number Crunching forum (there are many threads on this topic; for example, see Little Credit for a lot of Work!!!). General consensus is that aborting them is best.

QS

[Todd Madson]

Yeah, I'll abort.

Why? I have one that says it will take 103 hours remaining!

[QSilver]

Originally Posted by Todd Madson 

Why? I have one that says it will take 103 hours remaining!

Sounds like a good enough reason!


QS

[[email protected]]

Hi,

I hope this is the right place to post this. I'm having real problems running the latest enhanced optimized SETI on my Mac Pro.

I download Alex's latest files, put them in the seti directory, start boinc and nothing but error messages -- maybe I've forgotten how to copy new enhanced stuff into the right place...

/Library/Application Support/BOINC Data/projects/setiathome.berkeley.edu

is this correct?

Maybe someone can point me to the 'right' instructions for installing and running the new enhanced optimized SETI ?

p.s. I'm also running rosetta - could this be interfering with SETI?

atm, I'm running the standard SETI worker - its a lot slower than 7.2mb

Any help appreciated

[QSilver]

It's not Rosetta...it's probably a permissions or ownership problem.

To fix a permission problem, after installing Alex Kan's optimized apps:

1. In Terminal, cd ~Library/Application Support/BOINC Data/projects/setiathome.berkeley.edu
2. Type chmod 755 seti_enhanced*
3. Quit Terminal

Try running the app again.

If it doesn't work, try fixing the ownership:
sudo chown boinc_master:boinc_project /Library/Application Support/BOINC Data/projects/setiathome.berkeley.edu/*

QS

[reader50]

fortilan, were the errors coming from Rosetta? The Rosetta project had a meltdown earlier this week, defective firmware wiped the partition the project was hosted on. Took them days to recover and restore from backups.

[Gulliver64]

Originally Posted by alexkan, Apr 12, 2007, 06:23 AM 

Give me a couple days to get the PowerPC builds up and running. As you may already know, there are things that I did for v7.2 for x86 that weren't in v7.1 for PPC, so these alone should provide some performance gains. It remains to be seen whether the cache tuning I did for v8 will help PPC, so that means I'll need PPC testers once the code is almost ready. For that matter, I need testers with Core Duos, since I haven't assessed the performance impact on that architecture, either.

I realize my performance tuning has been targeted more towards the leaderboards lately, but hopefully what I've learned there will soon bear fruit for everyone else as well!

Hi alex,

any news on the PPC-version 8? It is now 5 months since the above announcement. Any chance we will ever see an optimized multi-beam PPC-app?

[halimedia]

Originally Posted by E.T from tellus 

...Anyway i don't realize any problem with heat. i think i can try 8 way X-serve next.

E.T, did you ever try that? I'd be curious to know...

Thanx for any feedback!

Ron

[halimedia]

Originally Posted by Gulliver64 

any news on the PPC-version 8? It is now 5 months since the above announcement. Any chance we will ever see an optimized multi-beam PPC-app?

See my post in your thread on this topic....

[adream]

im going back to using the v8 pre-release by the way, gotta try and knock those pcs from the top 5 .. it does the science properly and overclaims slightly but its alot faster than 7.1mb1

doesnt look like v8 is gonna appear anytime soon either ho hum

regards

adream

[Elphidieus]

Originally Posted by adream 

im going back to using the v8 pre-release by the way, gotta try and knock those pcs from the top 5 .. it does the science properly and overclaims slightly but its alot faster than 7.2mb1

doesnt look like v8 is gonna appear anytime soon either ho hum

regards

adream

Don't mind if I ask, where exactly can I find the pre-released v8 client...? Only from Alex..?

[arkayn]

Originally Posted by Elphidieus 

Don't mind if I ask, where exactly can I find the pre-released v8 client...? Only from Alex..?

Pretty much I do believe as it was never released out in the wild.

[Billy]

Originally Posted by arkayn 

Pretty much I do believe as it was never released out in the wild.

Is there a version for Core Duo or only Core 2 Duo?

[halimedia]

Originally Posted by Billy 

Is there a version for Core Duo or only Core 2 Duo?

There isn't at this point. The optimizations in v8 for Intel focus primarily on features exclusive to the Core2Duo-platform. However, there are some more general improvements in the v8-code that will eventually warrant a v8-build for CoreDuos, too. Last I talked to Alex, he was planning such a release. Be aware, though, that the performance improvement of such a release over the current v7.2-build would likely be marginal (max. 5 %).

HTH,

Ron

[Gulliver64]

Originally Posted by halimedia 

Be aware, though, that the performance improvement of such a release over the current v7.2-build would likely be marginal (max. 5 %).

HTH,

Ron

Are you talking about the modified 7.2 multi-beam or the original 7.2? The MB apps (at least for Mac PPC) are at least 20% slower than the original version.

[halimedia]

Originally Posted by Gulliver64 

Are you talking about the modified 7.2 multi-beam or the original 7.2?

Makes no difference, really. The only major difference between the two is an adjusted multiplier, allowing for correct credit claims when computing MB-data. Their performance characteristics are identical otherwise.

The MB apps (at least for Mac PPC) are at least 20% slower than the original version.

Are you speaking of the stock MB-apps or of Alex' optimized workers? And what do you mean by original version?

Curiously,

Ron

[Gulliver64]

Originally Posted by halimedia 

Makes no difference, really. The only major difference between the two is an adjusted multiplier, allowing for correct credit claims when computing MB-data. Their performance characteristics are identical otherwise.Are you speaking of the stock MB-apps or of Alex' optimized workers? And what do you mean by original version?

Curiously,

Ron

Ron,

I tried the alexkan 7.1 PPC no graphics version versus the alexkan 7.1-MB PPC no graphics. The multi-beam version also has the corrected multiplier. The MB-version is appr. 15% slower and also asks appr. 10% less credit due to the modified multiplier. When comparing the granted credit the MB-version produces 15% less credit/hour than the original alexkan 7.1 version.

P.S. There is no 7.2 for PPC, only 7.1.

[halimedia]

I can vouch for the fact that Alex' current Intel MB-apps are identical to their LF-predecessors with regard to performance. This means that they crunch identical WUs in the same amount of time. Of course, this does not take into account the change in the credit multiplier that was applied when MB was rolled-out publicly.

I've done all sorts of baseline testing with different v7 and v8 builds, both on PPC and Intel. The only thing I have not yet done is comparing the G4 and G5 v7.1 MB apps to their v7.1 LF predecessors. Considering your claim, I guess it's about time to do that then...

Cheers,

Ron

[Gecko_r7]

Originally Posted by adream 

im going back to using the v8 pre-release by the way, gotta try and knock those pcs from the top 5 .. it does the science properly and overclaims slightly but its alot faster than 7.2mb1

doesnt look like v8 is gonna appear anytime soon either ho hum

regards

adream

Hi adream:

I'd ask you reconsider doing this as Alex requested we not run his previous LF aps using the old multi. Doing so risks causing contention and consternation in the community. In the end, he's ultimately held accountable since the aps are of his making and release. In appreciation of his efforts and long-time support of our community, I hope you'll respect his request.

V8 is coming soon. Have faith.

[adream]

ok

the last v8 pre release was issued to testers in april

this is in no way a criticism of alex or any other developer, because i know they do actually have proper lives outside seti/boinc he heh just like i do

there appears to me no evidence whatsoever that development is happening on the v8 client apart from the pre release 6 months ago, a fact that appears to suggest that development has stopped. there isnt even any beta testing etc :-)

the pre release does proper science more efficiently than any other client available, the fact is that one in 40 WU's gets granted about 10 extra points... this being on a machine that is capable of 8000 points credit a day, hardly a big deal ?

my 8-core mac pro consumes 500-600 watts when fully loaded with seti (amongst other things). the price of electricity in the UK is very high, so i am faced with a very real cost per science unit completed equation... and to be honest i would rather use all the electricity as efficiently as possible with the risk of upsetting a few geeks (including my self as a geek!) than to plow on with low efficiency for many more months knowing that i have a magic little v8 that could do better !

for average units the v8 crunches are aprox 500-1800 seconds faster than 7.1mb1 (4600 secs instead of 6400 secs on many units over the last two days!) so you can see why its an issue in my tiny little mind (and wallet) :-)

as far as i can tell only my 8-core, UL1's 8-core and Bad to the bone's 2 4-core's are running the pre release, there may be more but i cant be arsed to look through everyone's stats lol

i totally agree about respecting alexs wishes and to be honest have only used the v8 for two days just to bring this issue up again so people don't forget what an advance v8 could be for xeons etc. my dream of an all mac top 10 is frustratingly close

regards

adream

[halimedia]

Hang in there, all of you! v8 will be coming to a Mac near you real soon now...

[alexkan]

O ye of little faith!

No, I didn't forget about all of you, but as time went on, it became easier to just do small-scale (rather than real-world) testing. Also, it eventually became clear to me that if I was going to be a perfectionist about this, I would never wind up releasing anything. So, without further ado...here's v8!

G4
G5
Core Duo
Core 2
Xeon

Note that there are two different Core 2 apps now, one for C2D machines, and one for Xeons. It was thanks to my pretesters that I recognized that the performance characteristics of Mac Pros are sufficiently different from iMacs and MacBooks to merit their own separate builds. Hopefully this app will also address whatever other smaller issues people were seeing, since all of them are now built against more recently BOINC libraries, and should all have their app icons properly baked in.

Source code to this release is also available. Note that there are two separate Xcode projects (since I build with different compilers and libraries depending on whether I'm building for PowerPC or Intel), but all the source code should be the same.

Let me know if anything goes wrong. It was kind of a mad scramble today to unify the source bases and get everything built, so it's entirely possible that I may have overlooked something in all the craziness.

[Elphidieus]

Thanks Alex....

I'm running it on my CDs and C2Ds....

Will have the v8 tested on my V8 soon...

EDIT: Just had a noisy WU on v8, notice that the output file still indicate a built of seti 5.13.

Are these V8 clients optimised for MB workunits, or are they just speed bumps for non-MB WUs plus a re-calibration of the credit multiplier...?

Task details

[Gulliver64]

Originally Posted by halimedia 

I've done all sorts of baseline testing with different v7 and v8 builds, both on PPC and Intel. The only thing I have not yet done is comparing the G4 and G5 v7.1 MB apps to their v7.1 LF predecessors. Considering your claim, I guess it's about time to do that then...

Cheers,

Ron

Hi Ron,

I just finished the first 2 WUs with the new v8. As you can see from my Host stats (results from 8 Oct 2007 9:56:21 UTC) the v8 client now claims the correct credit but is 10% slower than 7.1 was.

[halimedia]

Originally Posted by Gulliver64 

Hi Ron,
I just finished the first 2 WUs with the new v8. As you can see from my Host stats (results from 8 Oct 2007 9:56:21 UTC) the v8 client now claims the correct credit but is 10% slower than 7.1 was.

Interesting! Yesterday, I've benchmarked v7.1mb against v7.1 on G4 and G5. As opposed to the Intel MB-builds, there was a slight difference between the two: the MB-builds were 2 to 5 % slower. I ran this past Alex, and he thinks this is due to the fact that all v7.xmb-workers (including those for PPC) were compiled on an Intel Mac, whereas v7.1 was compiled on a PPC Mac. We'll look into this further...

The same issue will is likely to crop up on short WUs with v8 PPC-workers, as your example shows. However, the new workers will be significantly faster on long WUs (by ~15 %).

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