wuid=310019506

122,440.10 CPU Seconds for 719.06 Credit on my Q6600

I figure if only crunching AP WUs, my RAC would drop from 6100 to 2040.... This ain't going to be pretty!

BOINC on...on...
JDWhale

So your Q6600 took about 34 hours to complete the unit.

That probably means I can expect about 40 on my iMac and maybe 36 on the Dell.

---

So your Q6600 took about 34 hours to complete the unit.

That probably means I can expect about 40 on my iMac and maybe 36 on the Dell.

That Q6600 runs at 3.20Ghz....

My E4500 at 2.42Ghz is taking ~52 hours and should finish the first one ~1700 UTC today. There are 10 more AP WUs in that queue (~3 weeks running one core AP). This rig might compare more favorable to your Dell (similar clock)

wuid=310019506

122,440.10 CPU Seconds for 719.06 Credit on my Q6600

I figure if only crunching AP WUs, my RAC would drop from 6100 to 2040.... This ain't going to be pretty!

BOINC on...on...
JDWhale

Think yourself lucky on Beta AP unit on Q6600 @3.0 done in 125,220 (34.75hr) claimed 675.69, will probably get that granted.

wuid=310019506

122,440.10 CPU Seconds for 719.06 Credit on my Q6600

I figure if only crunching AP WUs, my RAC would drop from 6100 to 2040.... This ain't going to be pretty!

BOINC on...on...
JDWhale

Think yourself lucky on Beta AP unit on Q6600 @3.0 done in 125,220 (34.75hr) claimed 675.69, will probably get that granted.

Had 1 AP WU, 22hours on my X9650, now running Ubuntu 8.04.
Haven't trackt it down.
Is it between the 'normal' SETI WU's in pending tasks?
Sorry, have an eye infection, can't see to good :(.
It's still processing 11Hours an 31% completed, BOINC tell's 16 hours, to complete. Gonna take longer then expected!

---

.... This ain't going to be pretty!

...
JDWhale

Totaly agree with you! The credit should be three times the current!

---

.... This ain't going to be pretty!

...
JDWhale

Totaly agree with you! The credit should be three times the current!

Thats being greedy. My estimate from results on Beta site, looking at a small sample of different cpu's and comparing to default MB app says credits should be about 1000cr/AP unit. Obviously Eric et al disagree.

wuid=310019506

122,440.10 CPU Seconds for 719.06 Credit on my Q6600

I figure if only crunching AP WUs, my RAC would drop from 6100 to 2040.... This ain't going to be pretty!

BOINC on...on...
JDWhale

You surprise me. My first AP WU took about 26.5 hours (CPU time) (95239.55 secs) on my Q9450 @3560MHz. That makes the credit claimed / 24 hours about 25% less than I get for my CPDN model so I would reckon that the 722 creds / WU should be upped to 954 for parity (if CPDN is used as a benchmark).

F.
[edit]
Brain cleared a little after first cup of coffee; estimate for parity revised.
[/edit]

---

Totaly agree with you! The credit should be three times the current!

Thats being greedy...

Well, that can be a long discussion which I not intended to :)

My estimate from results on Beta site, looking at a small sample of different cpu's and comparing to default MB app says credits should be about 1000cr/AP unit. Obviously Eric et al disagree.

And yes, Eric is admin and has the last word.

---

Some more CPU times for AP Wu's

OS: XP Pro SP3 & Boinc ver. 5.10.45

Quad CPU Q6600 @3.00GHz CPU Time 110,122.70 (~30h) Credit granted 684.33

Pentium D CPU @3.00GHz CPU Time 285,520.70 (~79h) Credit granted 684.33

[thinking]
hmm.... I'll better buy a new battery for this 8 year old UPS.
[/thinking]

Completed this AP workunit. It took 40 hours to complete (machine is a Q6600 @ 2.4Ghz). The claimed credit is 715, so that works out at 17.875 cr/hour, assuming it gets granted.

Is anyone looking at optimizing the AP app, or is it too early for that?

---

BOINC blog

...Is anyone looking at optimizing the AP app, or is it too early for that?

Yes and Yes.

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

...Is anyone looking at optimizing the AP app, or is it too early for that?

Yes and Yes.

Yes it too early and yes its being looked at, sounds a bit of a contradiction doesn't it?

So we can expect SSE3, SSSE3 and SSE4.1 variants real soon now with 40 minute run times? Just kidding, although 40 min run times would be good :-) Even JD would be happy with that.

There is this message thread where Dotsch mentions porting the AP app to other platforms, including the PS3 and using the SPE's.

---

BOINC blog

LoL, Sorry, I'll give more detail,
- Yes a few people have taken initial looks at some of the code, but it is far too early to tell what the optimisation opportunities exactly are, given that the project has wrinkles to iron out etc. (so not quite a contradiction). Though the stock AstroPulse app looks to be operating (ignoring the backend growing pains), I personally wouldn't want to go grabbing a codebase, building on that, then the project makes some small changes & tweaks to fine tune processing & results etc.
- Also I feel that clouding the rollout with a different build may pollute any quality control measures in place to stabilise the system that Berkeley may be monitoring to get things running just right (though others may disagree)
- There are also tools and techniques to be built that must be adapted/extended to validate build output results, namely special test WUs timing and validation tools that compare stock versus new builds. Based on validator code, they need to really be built first.
- There are ports to other platforms, as you indicate, but I should point out that the stock one is already Windows, so no (or very little) 'porting' should be required as such. Optimisation, on the other hand, will probably commence with simple rebuilds with faster libraries & better compiler etc, but 'real' optimisation requires studying the algorithms in detail, along with detailed profiling and the leveraging of microarchitectural improvements... In other words, not overnight, it will happen, but carefully and incrementally. That typically involves disappearing into a cave somewhere for an extended period, with no human contact or sunlight and large amounts of caffeine.

Jason

---

"Living by the wisdom of computer science doesn't sound so bad after all. And unlike most advice, it's backed up by proofs." -- Algorithms to live by: The computer science of human decisions.

Completed this AP workunit. It took 40 hours to complete (machine is a Q6600 @ 2.4Ghz). The claimed credit is 715, so that works out at 17.875 cr/hour, assuming it gets granted.

Is anyone looking at optimizing the AP app, or is it too early for that?

Your numbers are consistent with mine. 40 hours for an AP unit with only 715 credit.

No second million by Christmas. :(

I hope the data is worth it.

---

Join the PACK!

If the new server-side credit adjustment mechanism works as advertised, the credit rate for AstroPulse should increase to near parity over the next 30 days. Because Eric was the author of that change and the basic value for an AP WU flop count is hard coded in the AstroPulse application, I think we'll just have to watch and wait.

                                                                Joe

If the new server-side credit adjustment mechanism works as advertised, the credit rate for AstroPulse should increase to near parity over the next 30 days. Because Eric was the author of that change and the basic value for an AP WU flop count is hard coded in the AstroPulse application, I think we'll just have to watch and wait.

                                                                Joe

The kitties be watching.....and waiting. Thanx Joe.

---

"Freedom is just Chaos, with better lighting." Alan Dean Foster

wuid=310159610

Since my first wingman "errored out" after 40+ hours, I'm a bit leary about giving this WU the green light. I'll wait 'til second wingman reports before deciding to run or cancel.

It's sad enough that AP WUs are underclaiming credit when completed, but a total waste of substancial CPU time is not going to be acceptable, IMO.

BOINC On...on...
JDWhale

[edit] Oops... too late. This WU is already 50 CPU hours invested and 90% complete on the mighty E4500. About ~5.5 more CPU hours to completion. [/edit]