Greetings all,

Just throwing this out there for discussion. I now crunch other boinc 64 bit projects out there, ABC and Riesel Sieve in particular that offer up around 6k per day of credits on a quad Intel Q6600 processor. I, like others get less that 4k per grid WU on this project. It used to be that this project returned alot more credit before in comparison to other projects before these new 64 bit apps showed up and closed, then passed this project.

If this cell type of ps3 processing is that much more performant that pc type processing, shouldn\'t the WU credit be reevaluated to adjust for the changing times?

Open for discussion...

Well... the two projects you took for comparison, are maybe not the best ones... ;)

Riesel Sieve is pretty much over granting credits, and ABC with the 1000 credits \"present\" for longer WUs, may also not be the best choice to compare...

But I understand what you mean... Even if I compare Cosmology@home (64bit Linux) and Einstein@home (32 bit app), I can easily get 3500 - 4500 or even more credits per day on my Q6600 hosts...

Actually I was also thinking about the credits here at PS3GRID since I started crunching, because if the Cell CPU really has the power of 16 (I think) normal CPUs the credits should be higher than what I can crunch with a normal Quadcore...

ButI don\'t think that the credits here should be reevaluated for the changing times... How should that work? Every time a new CPU comes out we should get more credits here? Why the Cell CPU hasn\'t changed, but the other CPUs got faster... ;)

Just my two cents...

---

pixelicious.at - my little photoblog

The PS3 is approximatley 16 times faster than a 32bit single core CPU of a certain speed, the reference unit. I forget what that is.

You could say a quadcore is 4 times faster than a single core, but that really is not true, it all depends on instructions per second that each specific processor can compute. Each needs to be referenced back to a common element, the benchmark computer.

All benchmarks are suppose to be based on the same factor and equal across projects, but in reality this does not happen. The developers are trying to change this and get those project that over grant back into a common range and even get windows ewaul to linux for the same speed processors. ITs an ongoing process not yet solved.

The credit should not change over time for a particualr host, unless it wears out and slows down.

As newer CPU\'s get faster and use reduced cycles for instructions, have higher bus speeds for data mvoing and so on, they may be granted more credit than any PS3 which remains the same.

I can however tell you that this x16 is about true. My single PS3 earned more credits in less than a year than all my other computers put together took a year to earn on other projects, thats about 27 CPU\'s. None are quadcore but some are HT CPU\'s. I posted this somewhere before, at this time I forget the exact details or where. None of my hosts got faster during that time, they all still run at the same speed, so they should be getting the same credit today they got 1 year ago. They only way I could get more credit, is replace the CPU\'s or computers with faster ones.

Credit has always been based on number of calculations per second * time.

So for that to be correct for the ps3 then it should be at least 16000/day if it is 16x faster. Even an old P4 can earn 1000+ credits a day so I feel 16000/day is a reasonable number if not a little on the low side.

*edit*
looking at the front page it says the cell is 10.3x faster than a single core cpu so even 10,000 a day is closer to what it should be than what it is now...

Credit has always been based on number of calculations per second * time.

So for that to be correct for the ps3 then it should be at least 16000/day if it is 16x faster. Even an old P4 can earn 1000+ credits a day so I feel 16000/day is a reasonable number if not a little on the low side.

*edit*
looking at the front page it says the cell is 10.3x faster than a single core cpu so even 10,000 a day is closer to what it should be than what it is now...

Well, you can go to this page to see that it is not that simple:
http://www.ps3grid.net/forum_thread.php?id=219

In practice, there are two ways of counting credits in BOINC. One is the standard one which runs a simple benchmark and computes the speed of your computer, then it multiplies this speed (number of average integer and floating point operations per second) for the time duration of the computation. The second one is to explicitely say how may average ops are performed by the application.

The discrepancy is due to the fact that the first case largely overestimate the Gflop/s of your machine on the real application because it measures it on a simple idealized benchmark.

The second method is more reliable, and it is the one we use. If you apply the benchmark used by BOINC to estimate credits on a GPU, you probably get a factor 50, but our measure of credits although conservative is more realistic. We hope that in the future the credits system normalize to the second method.

GDF