PS3

[IronBits]

Originally Posted by garo

Sorry to nitpick but you are talking about KWh and not KW/h. Power per unit time does not make sense.

How many solar panels would it take to run these ?

[Kevin]

My guess is the same thing keeping GIMPS from being ported to the PS3 is the same thing keeping all those other programs from being ported. They all probably need need double precision operations (for large integer multiplication, or something along those lines), the PS3 hardware is only capable of single precision operations, and even with complete optimization the performance hit you get from emulating double-precision operations with single precision operations makes the whole venture pretty much worthless.

[jasonp]

Originally Posted by Kevin

My guess is the same thing keeping GIMPS from being ported to the PS3 is the same thing keeping all those other programs from being ported. They all probably need need double precision operations (for large integer multiplication, or something along those lines), the PS3 hardware is only capable of single precision operations, and even with complete optimization the performance hit you get from emulating double-precision operations with single precision operations makes the whole venture pretty much worthless.

Yes. See this thread for the last time this discussion came up. Bottom line: no hardware double precision floating point? No fast 32x32 -> 64-bit integer multiply? No service.

Jason, you need to be much more comfortable with the software development process in general before gearing up to do a hard job like porting a distributed project to a boutique processor. I understand that not being able to help is really frustrating, but as Bob has said before, you don't learn to climb mountains by starting with K2. A complete environment for developing software on the PC you have now costs $0.

[jasong]

Okay, forget the PS3.

Based on a hunch, I'm considering buying a graphics card for LLR. Assuming there IS an app for LLR made for graphics cards(/me winks at someone in the shadows, not George) which would be the best card to buy?

[IronBits]

Greenbank is currently accessing one of my PS3s to see what he can get something ported over to it. :)
http://www.rieselsieve.com/forum/viewtopic.php?t=1136

[jasong]

Originally Posted by IronBits

Greenbank is currently accessing one of my PS3s to see what he can get something ported over to it. :)
http://www.rieselsieve.com/forum/viewtopic.php?t=1136

I know this is an old thread, but since anything made for the PS3, or Xbox 360 for that matter, would be proprietary, I think we should focus on graphics cards. Besides, if you're a supers-serious cruncher, building a multi-graphics card setup would be better bang for buck than a PS3 or an Xbox 360.

[Uncwilly]

Originally Posted by jasong

I think we should focus on graphics cards. Besides, if you're a supers-serious cruncher, building a multi-graphics card setup would be better bang for buck than a PS3 or an Xbox 360.

Only if you want to do the far LMH stuff. Don't waste your breath on GPU's until they can do 64-bit ops.

[jasong]

Originally Posted by Uncwilly

Only if you want to do the far LMH stuff. Don't waste your breath on GPU's until they can do 64-bit ops.

Since I haven't been here awhile, I'm guessing people are still chanting the same stuff about single-precision being too slow.

Ask yourself how much data graphics cards have to shuffle around in order to simulate things like a first-person view of someone blowing apart a demon. I know that doesn't absolutely prove that graphics cards would be better at LLR, but I guarantee the people that keep saying this haven't ever actually TRIED to make an LLR program that uses graphics cards.

[joking]
Oh, and for those of you who believe so-called common knowledge, I guess you also believe that bees can't fly and photosynthesis makes plants too hot to touch.
[/joking]

Maybe someone should actually take the time to port some code over to graphics cards. I think even unoptimized code would prove that it was a worthwhile endeavor.

[cheesehead]

Originally Posted by jasong

Since I haven't been here awhile, I'm guessing people are still chanting the same stuff about single-precision being too slow.

... because the arguments are still valid. You aren't showing us any disproof.

Ask yourself how much data graphics cards have to shuffle around in order to simulate things like a first-person view of someone blowing apart a demon.

Find out how much data Prime95 pumps through an FPU each second!

I know that doesn't absolutely prove that graphics cards would be better at LLR, but I guarantee the people that keep saying this haven't ever actually TRIED to make an LLR program that uses graphics cards.

... and you aren't trying to show us why any of the slow-single-precision argument is invalid!

Oh, and for those of you who believe so-called common knowledge

"So-called common knowledge" is a phrase more applicable to your casual and unsupported claims about graphic card capability than it is to the detailed explanations about floating-point arithmetic considerations that have been posted in previous discussion.

Maybe someone should actually take the time to port some code over to graphics cards.

Show us a theoretical basis (with details about the floating-point arithmetic) for thinking that such an effort would be worthwhile!

Maybe I should try entering the snow-plowing business around here by attaching a plow blade to the front of my Saturn sedan. How could I possibly know that wouldn't work unless I actually try it? (Answer: doing some research beforehand. How many models of plow blade are available that can be fitted to Saturn sedans? How much would it cost to construct the necessary fixtures on the front of my car to mount a plow blade? Survey the snow-plowing business to see which vehicles others use -- hmmm, lots of pickup trucks, no sedans -- wonder why? How likely is it that I'd get through a season of actual, productive snow-movement without ruining my car?)

Can you explain why we need guard bits, and how many, in single-precision versus double-precision floating point arithmetic, and how this affects (a) the number of bytes needed to represent each Mersenne number, (b) the number of floating-point operations needed in each FFT transform and (c) the resulting ratio of throughput using single-precision versus double-precision? (Hint: you can glean this from the earlier discussions. Once you do, you'll be on the way to understanding the "stuff about single-precision being too slow".)

[jasong]

The proof would have come by now, but the person who made the code is very much under the weather. When they get better, or infinitely worse, the proof will show up. As a comparison, though the illness isn't the same, if you were in the last stages of AIDS with very little hope for recovery, would you be interested in getting a lot of attention?

To be brutally honest, the probable reason the proof hasn't shown up at all is probably because he knows that my crappy reputation here, though he probably doesn't agree with it, is his best protection possible.

I know that will get me a lot of jeers, but unlike a lot of people here, I don't enjoy making myself look greater than I am. Better a wise pauper, than a foolish billionaire.

I am egging you people on because the greater your fall, the more likely it is that will change your ways, in my opinion.

[crash893]

Why not do it anyway as a proof of concept

we didnt start on p4s we had to work are way up to them but its a good thing that we started before hand for lessiosn learned.

if a graphic card method came out now it would probably suck but in a year or two when graphic cards come around i think we will all be glad they put in the effort ahead of time