Distributed Playstation

withinavoid writes "News.com has a story up about the next generation Playstation 3. Apparently the game developers are asking for a 1000 times performance increase and that's just not possible, so they are looking at distributed computing as a possibility. "

Comments

Sony Hype Machine

[Spankophile]

Hmm, the last time Sony announced their "Next Generation Console" (aka PS2), it was still over a year away, while the PS1 was still selling well. The real kicker (aka purpose) though was that it was enough to keep a large number of gamers away from the DreamCast, which was a great system.

It would appear to me that hyping a PS3 while the PS2 is still selling strong would be an attempt at keeping people from getting an Xbox or GameCube.

Missing the point

[Anonymous Coward]

Distributed != Over the internet.

Distributed in that the chips involved are linked to each other without any knowledge of the number or place.

Think "NUMA memory and cluster of processors in a box".

Rather than needing a large room for a Beowulf, they want to do it in a much smaller box.

They aren't stupid guys, and while they may be tight lipped about the OS in the article, look at the K42 project for an idea of one of the ones in the running.

"The K42 group is developing a new high performance, general-purpose operating system kernel for cache-coherent multiprocessors. We are targeting next generation servers ranging from small-scale multiprocessors that we expect will become ubiquitous, to very large-scale non-symmetric multiprocessors that are becoming increasingly important in both commercial and technical environments. "

Comes from current PS2 architecture

[iamr00t]

Basicaly, we all know that it was hard for PS2 developers to make games for it initialy.
The reason was the Emotion engine in PS2, that it explicitly multithreaded, i.e. you have to make your program use all threads (unlike PPRO for example, where CPU does it for you).

It's really a whole new way to program.

Now it seems that Sony convinced some developers to lean it there's nothing stopping them from making more threads (there are 16 in Emotion if I am not mistaken).

Oh, and it has nothing to do with distiributed computing over the Internet. The application architecture is similar, but that's it. And yeah, no batches here :)

As for IBM involvement, here is the article in Wired Magasine about their cell computer

Oh, and ahother one about PS2 and PS3, that one is quite old, but explains where Sony is going.

Re:Comes from current PS2 architecture

[voronoi]

Err, the reasons the PS2 is hard to work with is:

1) Poor tools
2) Arcane DMA alighnent issues
3) Misguided selection of VU integer instructions
(no imul, you have to jump through hoops to do xor, only 16bit, yet flags are in the upper 16bits)
4) Hard to singlestep the VUs
5) Very limited blending modes in the GS
6) Very limited iterator precision in the GS
7) No hardware clipping
8) Wierd GS rules where rendering horizontal triangles is much slower than large vertical ones
9) Non perspective correct iterated vertex colour
10) Limited vram
11) 1.0 * 1.0 == 0.999999999 in the VUs

I.e. it's cheap and flawed (but hey that's a challenge and some people seem to like it)

Then you go

"It's really a whole new way to program."

You are not a programmer are you? Didn't think so.

"Now it seems that Sony convinced some developers to lean it there's nothing stopping them from making more threads (there are 16 in Emotion if I am not mistaken)."

What are you talking about? I can't possibly imagine why you would want 16 threads in PS2 game.
Genrally CPU multi threadding is quite costly. Most games are written to run at a solid 60fps*, so you can often get away with out multithreadding
stuff like the AI, the Renderer or some trickle loader.

BTW The only console that really had seriously multi-thredded games was the N64.

* Due to field rendering you have 2x the fillrate (and more vram) at 60fps than at 30fps. Dropping to 30fps is bad!

Re:Hohumm

[hcdejong]

Use pictograms. They are much more intuitive.

No, they're not. Icons (with VERY few exceptions) are only obvious after they've been explained. If the icons are good, and there aren't too many of them, you need to explain them only once. Try to replace everything with icons (which you're going to have to do if you want no localization problems), you end up with way too many icons for anyone to remember. Do you know (without looking at the tooltip) what every single icon in say, Word is for? Didn't think so.

Also, icons don't completely solve the localization problem. Images (especially the abstract drawings used in icons) can have different meanings in different cultures.