It just makes you want to cry, this whole CreditNew debacle. However, I prefer to see the comical side of it instead.

Sten you are doing good science, the kind we can only hope Dr. A appreciates.

---

If they at least could openly admit that doing AP is so much more important than MB, and explain the reasons for that.

IIRC AP gets some NSF funding and I don't think MB does. If anybody knows differently please correct me.

---

If they at least could openly admit that doing AP is so much more important than MB, and explain the reasons for that.

IIRC AP gets some NSF funding and I don't think MB does. If anybody knows differently please correct me.

See http://www.nsf.gov/awardsearch/showAward?AWD_ID=0307956 for when the NSF grant related to Astropulse started and was expected to expire.

As to AP vs. MB credits, one merely needs to stick to stock CPU processing to get about the same credit per time for either. Perhaps the mistake was allowing GPU processing, since BOINC is clearly not ready yet to balance things properly across different kinds of hardware. And there may well be similar effects from using anonymous platform. OTOH, perhaps the crooked yardstick doesn't really matter as much as analyzing the data.

                                                                   Joe

It just makes you want to cry, this whole CreditNew debacle. However, I prefer to see the comical side of it instead.

Sten you are doing good science, the kind we can only hope Dr. A appreciates.

Be careful what you wish for. The easiest solution to level the playing field might involve reducing the credit values for AP results, not raising the MB credit values.

It just makes you want to cry, this whole CreditNew debacle. However, I prefer to see the comical side of it instead.

Sten you are doing good science, the kind we can only hope Dr. A appreciates.

Be careful what you wish for. The easiest solution to level the playing field might involve reducing the credit values for AP results, not raising the MB credit values.

Yep, as other informed people have stated, all you would have to do is replace the Stock AP CPU App with the Optimized version. AP credit scores would plummet instantly. That might level the playing field on nVidia cards, however, there would still probably be a large difference on AMD cards. Seems the current AMD MB App just isn't up to the task.

In any case, Everyone's RAC would plummet once again. I'm sure that would go over well...

Perhaps the mistake was allowing GPU processing, since BOINC is clearly not ready yet to balance things properly across different kinds of hardware.

                                                                   Joe

Considering where mistake could be I would rather assume that RAC/credit per se was big mistake... And definitely not to use so powerful devices as modern GPU are.
And more realistically: there are 2 things:
1) to measure FLOP performance
2) to stimulate users to participate via competition stimulating.

And BIG mistake is to consider that these 2 different goals can be implemented uniformly.

As any physicist in particular and any scientific-oriented man in general knows one should always consider extreme cases to understand if found solution is right or wrong.
In our case that extreme case will be, for example, non-CPU intensive project.
Should peoples be stimulated to participate in such project if it has scientific value - surely yes.
Can we measure peoples participation in such project by FLOPs they did - surely no!
Obvious discrepance!

And now back to other cases: any calculation depends on data. Each particular scientific task requires own algorithm to solve. Each algorithm (recall my example with y=sin(x) or y=a+x ) requires different amount of floating point operations per memory access.
In part this ratio depends on optimization level of course, as Richard stated after my example, but largely, VERY largely this depends from algorithm per se.
As in my example, it's not possible to optimize a+x to be so compute-intensive as sine x.

But computer (CPU or GPU or any other device) needs to access to memory. It's inevitable. But this access in no way accounted for in FLOPs measurements.
So, not only in extreme case (where it's just obvious) but in any other projects case we see that FLOPS measurement and "payment" for participation to stimulate that participation are DIFFERENT goals and should be solved differently, by different methods.

That's the root of big initial design issue I see in BOINC credit scheme. No matter how good or bad we will be in FLOPs measurement, it's just not that needed for "payment" in participation.

Better or worser via FLOPs measurement can be compared only inside same algorithm (and on the same hardware architecture), One can say that one opt app better/faster/does more FLOPS than another by measurement on the same hardware and solving same task, but between projects and even between AP and MB algorithms differ. All we can say in this case is that one hardware/software more or less suitable for particular task than another one, that is, relative comparison, not some absolute measurement unit as RAC pretends to be.

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

It just makes you want to cry, this whole CreditNew debacle. However, I prefer to see the comical side of it instead.

Sten you are doing good science, the kind we can only hope Dr. A appreciates.

Be careful what you wish for. The easiest solution to level the playing field might involve reducing the credit values for AP results, not raising the MB credit values.

... which could in turn result in more people dropping Seti altogether in favor of higher-crediting projects.

---

David
Sitting on my butt while others boldly go,
Waiting for a message from a small furry creature from Alpha Centauri.

It just makes you want to cry, this whole CreditNew debacle. However, I prefer to see the comical side of it instead.

Sten you are doing good science, the kind we can only hope Dr. A appreciates.

Be careful what you wish for. The easiest solution to level the playing field might involve reducing the credit values for AP results, not raising the MB credit values.

... which could in turn result in more people dropping Seti altogether in favor of higher-crediting projects.

This highlight another fundamental credits flaw. Credits needed for encourage peoples to help in scientific problems, not to discourage them. Hence, any credit payment scheme should always take into account not only "scientific fairness" in FLOPs counting, but should mainly account for human psyhology.
Credit payment can have inflation, that is, at some point it can start to pay a little more for "the same" work. But in no way it can deflate, it should never pay less, it just annoys peoples and always acts against credits goal.
And this is fundamental issue indeed - authors of new credit payment system forgot for what this system is needed at all....

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

After this, surely nobody can accuse me of only crunching to get the highest RAC? :-)

If any does, then they are 5 cans short of a 6 pack :-)

---

Sten, I may have missed it, but what do you consider a point of
leveling off? Something like a score that doesn't vary by more
than + or - 5% for 2 weeks? Or something else?

Sten, Are you running lunatics on your computer?
Im wondering if I went to stock aps on my I7 920 rig if my RAC would go up?

---

[/quote]

Old James

IIRC for GPU MB stocks are the same as Lunnatics, both the same x41zc CUDA code so your RAC will be the same too.
On MB CPU or AP they are not. That´s one of the reason why AP still paids more, AP stock still not use the optimized builds.

---

IIRC for GPU MB stocks are the same as Lunnatics, both the same x41zc CUDA code so your RAC will be the same too.
On MB CPU or AP they are not. That´s one of the reason why AP still paids more, AP stock still not use the optimized builds.

Well, GPU stock AP almost match to opt GPU AP, roughly in the same degree as GPU MB. Only CPU stock and opt AP differs hugely.
Nevertheless issues with RAC...

---

SETI apps news
We're not gonna fight them. We're gonna transcend them.

...
23+ hours later, at 10,907.24 feet (RAC). Lost 24911 feet (RAC) since March 6th. It ain't over yet....

When it firmly crosses below 10,000 I think a post in the milestones thread will be in order. ;-)

                                                                   Joe

Mrs.Trellis of North Wales and her knitting circle say their giant cushion will soon be ready for deployment - but they do need a delivery address - will "Sten the Parachutist, Norway" suffice?

---

Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?

Mrs.Trellis of North Wales and her knitting circle say their giant cushion will soon be ready for deployment - but they do need a delivery address - will "Sten the Parachutist, Norway" suffice?

Make that Sweden though :-)

Damn, it's already been sent,

Claggy, North Wales.

Damn, you Swedish guys are unbelievable!
I love it, keep going on! :)

---

Aloha, Uli

Sten, Are you running lunatics on your computer?
Im wondering if I went to stock aps on my I7 920 rig if my RAC would go up?

Yup, I'm running lunatics "anonymous platform". I believe if running stock, the RAC will indeed go up, but the production rate will fall :-)

Crazy eh?

Sten after you level out, Will you do stock apps and see what happens? If yuo choose not too, I would be willing to try it on my I7 920.

---

[/quote]

Old James

RAC will fall even more with stock CPU apps which are slower for both MB and AP
(you will get the same credit per task but you will do less tasks/day)

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)