(Sorry for another off-topic message, and this thread hogging the AGI list in general. It's winding down already, though, and some of it might have been useful to some folks here).
On Mon, Nov 28, 2005 at 08:22:36AM -0800, [EMAIL PROTECTED] wrote: > Unfortunately, the Cell was designed specifically for gaming. So if game > devs don't think it meets their needs, that's a bad sign. Not really, the Cell is a generic architecture strongly influenced by Blue Gene (which has a better OPS/Joule ratio, incidentally, and a real signalling interconnect while the Cell by itself is pretty autistic). Sony PS3 is a game console allright. If the game developers universally don't like it, I guess it means it won't sell too well. I don't really care about their sales, provided they will ship the console with a Linux dev kit. > > All game dev punks can think of is how to render > > more and prettier polygons > Wrong. Among many other things, we do extensive GOFAI (including minimax > and pathfinding), collision detection, audio processing, and physical > simulation. Several of these are about as ideal candidates for the Cell as > you're likely to find. Collision detection, audio processing (full acoustic model), flocks, particle systems etc. are all instances of physical simulation. The Cell should do very well for such workloads, particularly because of parallelism, on-die memory bandwidth (granted, 2 MBytes is not very much) and good float (and in-register float and integer array SIMD) performance. > > preferrably without throwing away their old codebase. > I'll say it again: if you don't understand why this matters, you simply > don't understand how making real software works. Given a choice between > writing new code and re-writing old code, real programmers strongly prefer > the former. I perfectly understand the economics of software development. The point is that the rules of the game are changing as single-core CPUs will become the exception, not the rule on desktops and consoles of the near future. You will have to rewrite the old code for multiple cores (2, 4, 8, soon more). Multithreading will work well on the systems of next couple generations. It will no longer work very well after that. Hopefully, these future systems will support lightweight message passing in hardware. > The point is, there needs to be a payoff. If programming the Cell was > harder, but gave 3x performance, nobody would be whining. We're whining > because it's harder and gives 1x performance. If this is so, the PS3 is going to be a dead brick. It is is quite possible. I don't know what's going to happen. > > Local-connectivity integer automata networks and large scale MD would > > run just fine on SPEs. > Do you know of any real programmers (ie, people who get paid real money for > solving real problems) who've actually used the Cell and believe that? Nobody has used the Cell yet. All we have are simulators (I haven't even bothered with the simulator because I'd rather not jeopardize this FC4 machine, and a VMWare instance of it is going to be a bear). Actually, I should ask the computational chemistry folks what they think of the architecture. -- Eugen* Leitl <a href="http://leitl.org";>leitl</a> ______________________________________________________________ ICBM: 48.07100, 11.36820 http://www.leitl.org 8B29F6BE: 099D 78BA 2FD3 B014 B08A 7779 75B0 2443 8B29 F6BE ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
signature.asc
Description: Digital signature
