[gulliver]

Originally Posted by Bad to the bone

Look at this WU!

http://setiathome.berkeley.edu/worku...?wuid=80752856

My 2.5 GHz G5 is the one with 1,952 seconds. I wish, ALL would be like this.

How did you do that? My G5 Dual 2.7 GHz (http://setiweb.ssl.berkeley.edu/resu...?hostid=633575) needs at least 5,000 seconds for that credit. My G5 Dual 2.5 GHz (http://setiweb.ssl.berkeley.edu/resu...?hostid=638749) takes even longer. Could you post the wisdom file please?

[rhettmaxwell]

Originally Posted by gulliver

How did you do that? My G5 Dual 2.7 GHz (http://setiweb.ssl.berkeley.edu/resu...?hostid=633575) needs at least 5,000 seconds for that credit. My G5 Dual 2.5 GHz (http://setiweb.ssl.berkeley.edu/resu...?hostid=638749) takes even longer. Could you post the wisdom file please?

Dunno what gave the kick. Nothing special. Went back to BOINC 4.4.4 superbench recently from 5.2.13 and another attempt to run 5.4.9 again. 5.4.9 (stock and superbench) always pukes on me with segmentation violation. Good old 4.4.4 is yielding credit very inconsistently but appears to be slightly more productive than 5.2.13. I wonder if it'll outperform 5.2.13 over time.

Check other results here. There is one Wu with 5,747 sec that harvested 31.86 credits. Not bad compared to 7.2 to 7.4 k sec worth about 30-31 credits.
http://setiathome.berkeley.edu/show_...hostid=1441575

I'm using Todd Madson's wisdom like I also did with the old worker. Gotta find some time to shut down all apps and create my own wisdom file one of these days. Will be interesting how the various combinations will perform.

[Gecko_r7]

Originally Posted by Bad to the bone

Dunno what gave the kick. Nothing special. Went back to BOINC 4.4.4 superbench recently from 5.2.13 and another attempt to run 5.4.9 again. 5.4.9 (stock and superbench) always pukes on me with segmentation violation. Good old 4.4.4 is yielding credit very inconsistently but appears to be slightly more productive than 5.2.13. I wonder if it'll outperform 5.2.13 over time.

Check other results here. There is one Wu with 5,747 sec that harvested 31.86 credits. Not bad compared to 7.2 to 7.4 k sec worth about 30-31 credits.
http://setiathome.berkeley.edu/show_...hostid=1441575

I'm using Todd Madson's wisdom like I also did with the old worker. Gotta find some time to shut down all apps and create my own wisdom file one of these days. Will be interesting how the various combinations will perform.

I might be able to shed some light. Seti Enhanced uses FLOPs counting as part of the new credit scheme. Benchmarking is no longer relevant w/ the way Enhanced claims credit and does not affect credit when using the current Boinc 5.x clients. Until just a couple of weeks ago under the previous Seti project (pre-Enhanced), optimized applications that completed work quicker, did so w/ lower claimed credit vs. the standard Berkeley distributed aps. In essence, quicker computers would complete more WUs and therefore perform more work but may have claimed only similar credit to a much slower machine thaat only completed 1 WU in the same time. Optimized Boinc applications (Superbench) equalized this to a large degree so that faster rigs and those running opt. science applications (there are other reasons to run opt clients as well, esp if running mult. projects, hosts, etc.) would claim the appropriate credit due.

There has been some observation within the Seti forums that running certain Boinc 4.x clients w/ Enhanced can cause erratic operating behavior, and in the case of Opt. 4.x clients, also substantial overclaiming of credit (because of higher "multiplier" used vs. Boinc 5.x distributed clients....Gent's, please correct me if this is not correct). This WU from your computer appears to show exactly this:

http://setiathome.berkeley.edu/worku...?wuid=80785923

You're claiming 40.69 w/ client 4.44 Super Bench and computer# 2436605 is claiming 12.41 using the distributed Boinc 5.4.9 and computer #239002 is claiming the same 12.41 using 5.2.13.

The subject of "claimed credit" is an extremely passionate one right now given the new credit structure w/ Enhanced and the resultant longer WUs and disproportionate lower credit awarded for time spent. To sum it up, people used to receiving a lot of credit w/ Seti Std from running opt. clients and aps have seen their RACs fallen more sharply than those using the standard clients and aps from Berkeley. The new credit system also makes overclaiming stand-out like a sore thumb. There have been recent inquiries/suggestions/demands that Berkeley should require clients of at least 5.x to process Enhanced WUs and would do this by denying work to older clients. There have also been demands for overclaimed work to be "adjusted" or removed from the totals. There are number of issues the Berkely has to deal w/ right now w/ various priorities but given the massive consternation by the most vocal (and substantially productive group) members you will see on the Seti and team forums, the "credit issue" looms large and is perceived to be perhaps the #1 issue and source of frustration in the active Seti communities.

Depending upon your views of Seti, your participation, motivations, role in community etc. you may want to consider going back to 5.2.13....or not. There is a very good liklihood you will soon need to use a min. of 5.x anyway. It's your call, I am not passing judgement, only offering an explanation to your observations and educating on recent developments w/ the new migration to Enhanced. Regardless, this does seem to explain your observations.
Good luck!

[arkayn]

Interesting read there, I have always run pretty much the newest thing that was out there. I never did get into the super bench distributions.

[alexkan]

Alright, I have some good news and some bad news.

First, the good news. I've finished v5, complete with Crunch3r-style stderr output. The current binary is optimized for G5 but will still run faster on G4s than any previous versions. I'll still come out with a G4-specific compile, but for now, I'll make sure that the quads take their rightful places on the Top Computers list. Give it a try, post reference WU times, etc.

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

The source will be up shortly, although I actually played with compile flags this time, so hopefully recompilation won't be needed for G5. G4s need more than recompilation, so don't bother.

And now, the bad news. It's not going to get much faster. I estimate that at most, I can get maybe 5% out of this if I go back and schedule every single Altivec loop I've written optimally, and frankly, I don't have the time or patience to do this. Real Life calls, after all.

I'm thinking about whether or not to post this in the main SETI boards. I'd hate to rub salt in the wounds of Windows crunchers reeling from Crunch3r's departure, but Mac users need their optimized clients, too. Also, this comment by Crunch3r in this thread needs to be set straight:

Regarding the G4/G5 macs... well i know that alex kan is looking at it but i wouldn't expect much speed improvements.

Of course, I wouldn't complain if one of you decided to post the link there before I finish deciding.

[BTBlomberg]

Originally Posted by Bad to the bone

Look at this WU!

http://setiathome.berkeley.edu/worku...?wuid=80752856

My 2.5 GHz G5 is the one with 1,952 seconds. I wish, ALL would be like this.

I am amazed you got credit for it. The error it shows basically says the Client (SETI) stoped responding or communicating with BOINC for over 30 Seconds and the WU processing died, so your submittion should have been deamed invalid unless it really was finished and stalled at the very end. If that is the case it's a great time, but likely you were lucky to get points.

[rhettmaxwell]

Originally Posted by BTBlomberg

I am amazed you got credit for it. The error it shows basically says the Client (SETI) stoped responding or communicating with BOINC for over 30 Seconds and the WU processing died, so your submittion should have been deamed invalid unless it really was finished and stalled at the very end. If that is the case it's a great time, but likely you were lucky to get points.

Yeah, after I had a little more time to investigate into this behavior I reached the same conclusion like you and others. It's really interesting to see how different the 4.4.4 client behaves on various machines. I actually stuck to this old workhorse to compare performance before upgrading but on this box I accidentally launched it instead of 5.2.13 after another failed attempt to run 5.4.9 and was surprised what happened.

Thanks also to Gecko for elaborating on the various aspects involved! I've read many of the posts regarding the credit system and it's getting a little bit ridiculous at times.

But, part of the motivation to participate in BOINC projects is accumulating credits for various reasons. Otherwise, the whole line of distributed computing projects would've never got off the ground, I'm afraid...

[applefanss]

Thanks, Alex, v5 goes like the proverbial clappers. I don't have the means to provide you with precise stats, but just 2 units from the quad (202253)

http://setiathome.berkeley.edu/worku...?wuid=80647824

and

http://setiathome.berkeley.edu/worku...?wuid=80774400

show a gain of some 20% over v4. That's a great contribution to the project, like all your previous efforts without which the platform, the team and all of us would look very ordinary.

Let's hope your new employers appreciate their catch.

[arkayn]

Just switched over to v5 and will see what it does on my eMac.

I am running Crunch3r's SSE2 app on both of my Dells, so far they are outstripping my eMac.

[liebsmaschine]

Originally Posted by alexkan

I've finished v5, complete with Crunch3r-style stderr output. The current binary is optimized for G5 but will still run faster on G4s than any previous versions. I'll still come out with a G4-specific compile, but for now, I'll make sure that the quads take their rightful places on the Top Computers list. Give it a try, post reference WU times, etc.

Ah, Alex, we love you. How can we ever repay you? :-)

I'll hold off until you post the G4-specific compile (to minimize the hassle of timing it right not to interfere with currently-processing WUs and then restarting BOINC, etc.). But thanks in advance for everything!

One other note (not addressed to Alex in particular): My <a href="http://setiathome.berkeley.edu/results.php?userid=8344130">results</a> show a bunch of Client Error WUs. However, I don't think it's Alex's client, because if you look at the Work Unit ID for the ones that have errored out, every computer that has attempted to process the WU has returned a Client Error. Anyone know what's up? Why is Berkeley sending out so many faulty WUs?

[Gecko_r7]

Originally Posted by alexkan

Of course, I wouldn't complain if one of you decided to post the link there before I finish deciding.

Thanks Alex for V5, every bit helps. You've been so good to us and this group. I know we all wish you well in your new job At the same time, I can tell you personally that reality sometimes creeps in way to far and it's nice to have a diversion that provides a creative outlet. We hope you will continue to bless us w/ the benefit of your creativity when you feel the urge. Your efforts very much encourage and fuel the passion that many of us have for the project. You help keep it real and fun! This forum is a great gathering place of positive attitudes and "can do" spirit.

The main Seti forum is dreadful at the moment and not a very encouraging environment to spend much time.
Any newbe's checking it out could only be disuaded from joining the project.
I'll post the link as a "beacon of light" for the lost PPC crunchers who may be wandering in the current darkness of the main forum.

BTW, I PM'd you w/ an interesting read. File is dowloadable at Megaupload (you may have to click the upright link on the interim page to get to the actual download page.) If there is a problem, e-mail me at my hotmail in the PM or let me know a different way to send it to you.
Sorry, I don't have a personal host site.

[alexkan]

Originally Posted by jackal

I'll hold off until you post the G4-specific compile (to minimize the hassle of timing it right not to interfere with currently-processing WUs and then restarting BOINC, etc.). But thanks in advance for everything!

I wouldn't hold off on the transition if I were you. I think the performance gain over v4 is already enough to negate the inconvenience (I was of the impression that the transition is less troublesome than in the past?), and I don't have a rigid timeframe for the G4 version, anyway.

So Alex, does this new build need wisdom files? Is it 10.3 compatible? Do we just replace app and xml file?

[alexkan]

Originally Posted by Andrew F

So Alex, does this new build need wisdom files?

No wisdom files are needed, although you can provide pregenerated ones for a tiny performance gain.

Is it 10.3 compatible?

The build is compatible with 10.3.9, as are all its predecessors.

Do we just replace app and xml file?

Yes.

[BTBlomberg]

Originally Posted by alexkan

I wouldn't hold off on the transition if I were you. I think the performance gain over v4 is already enough to negate the inconvenience (I was of the impression that the transition is less troublesome than in the past?), and I don't have a rigid timeframe for the G4 version, anyway.

BOINC and the WU see it as the exact same app so there will be no issue. I just switched with one in the last 2 hours of Est. 9:30 hour crunch. Hopefully I will see this mess up the estimate and crank through the rest.

[BTBlomberg]

Originally Posted by alexkan

And now, the bad news. It's not going to get much faster. I estimate that at most, I can get maybe 5% out of this if I go back and schedule every single Altivec loop I've written optimally, and frankly, I don't have the time or patience to do this. Real Life calls, after all.

I wish I saw your post last night after I posted. 14 hours of crunching optimized missed. Oh well.

Thanks again Alex. As for the quote above, I think we all know there is a wall out there to how optimized you can get this. Any extra edge you can give us we appreciate.

Does anyone have a general estimate of how much faster v5 may be over the stock client? We have had a little here and a little there. It would be nice to know start to finish.

[Gecko_r7]

Originally Posted by Bad to the bone

But, part of the motivation to participate in BOINC projects is accumulating credits for various reasons. Otherwise, the whole line of distributed computing projects would've never got off the ground, I'm afraid...

Yes, there is a truth to your comment. By using dist. compt. to go beyond the scope of paid professionals and the dedicated groups of core supporters, Seti had to come-up w/ a reward system to expand it's appeal to casual volunteers. Like anything, if one can create a contest and foster friendly competiton in the mix, it generally can accelerate appeal and production. With more people, come more motivations and expectations. "Credit awards" may carry very little significance vs. the fundemental intent and focus of the project as far as Seti project professionals are concerned, but it's THE prime motivator for a sizable portion of the contributors that Seti now so heavily relies. It may be considered an evil by some, but it's a necessary evil to appeal to the masses. Hopefully, they will iron-out the bugs and this whole controversy will calm down so everyone can focus on the science again.
Good crunching!

[Todd Madson]

Some things I've noticed about V4 before upgrading to V5 - it was speeding up.
How cool is that?

It was producing work units in the 3 hours and 30 minute range and towards
the end of its short lifespan we had 2 hrs 50 mins to 3 hours 10 minutes.

There's a 3:08 block at 100% still on my screen ready to submit.

So what I'm seeing so far on my machines' screen seems VERY encouraging
with regards to work unit production. Stay tuned - I've got an errand to run
and will be back with work unit URLs to see the times,