If the AGI community could benefit from Cell Broadband maybe they could come up with their own design...
What attributes would the AGI community want in their Cell Broadband Engine? Dan Goe ---------------------------------------------------- >From : Tony Lofthouse <[EMAIL PROTECTED]> To : <agi@v2.listbox.com> Subject : RE: [agi] /. [Unleashing the Power of the Cell Broadband Engine] Date : Sun, 27 Nov 2005 19:42:18 -0000 > Are any of you guys familiar with the SAI Architecture? > > http://iris.usc.edu/~afrancoi/sai/ > > I guess you could classify it as a async message passing model (sort of). > > It has some nice features and may be a good fit to the CBE platform. > > Is there anyone familiar with SAI and CBE who has a view on this? > > > -----Original Message----- > > From: Eugen Leitl [mailto:[EMAIL PROTECTED] > > Sent: 27 November 2005 7:31 PM > > To: agi@v2.listbox.com > > Subject: Re: [agi] /. [Unleashing the Power of the Cell Broadband Engine] > > > > On Sun, Nov 27, 2005 at 10:21:07AM -0800, [EMAIL PROTECTED] wrote: > > > > > In general, I think it's fair to say that there's a lot of concern in > > the > > > game dev community about how useful the Cell will really be. It will > > > certainly have far less effective, usable power than the raw specs > > suggest. > > > > Repeat after me: "it depends on your problem". Some (many) codes will > > bite. > > Some will run like rabid foxes on meth. > > > > > There are a couple of specific issues: > > > > > > - Multithreading is really hard to work with. Even for a programmer > > with > > > extensive experience, multithreaded code is simply slower to write and > > > harder to debug than single-threaded code. This is a fundamental > > problem > > > that is largely unfixable by tools and libraries. > > > > Yes, people are lousy at parallelism. But it doesn't matter: the easy > > payoffs on ramping up clock of single cores are past. It's time to dive > > into the parallel programming model. If you can't state your problem > > in terms of asynchronous message passing (not just threads), you've got > > a problem on your hands that will only get worse with time. > > > > > - The Cell programming model is asymmetrical (ie, not all processors are > > > identical), which obviously makes things harder. > > > > 8 SPEs are homogenous allright. The main CPU is just the caretaker, and > > runs > > your vanilla OS, and feeds the SPEs (or sets them up to do the work > > by themselves, and just twiddles thumbs). > > > > > - The individual processors in the Cell are relatively underpowered by > > > today's standards. > > > > Not if you consider in-register SIMD performance, and codes running > > directly out of their individual on-die memory areas. > > > > > - Perhaps most critically, the SPEs just don't have enough memory to > > > actually do much real work. There are very few interesting problems > > that > > > fit in 256K, and the performance penalty for accessing "external" memory > > is > > > > You can assume 2 MByte on-die memory. > > > > > very substantial. > > > > How is that different from today's machines? > > > > > The first three points above apply to a somewhat lesser extent to the > > Xbox > > > 360 processor also. > > > > The Xbox is a closed shop. It doesn't matter which hardware it runs, > > it will never become available to generic programming projects. > > > > > Although it's clearly true that parallelism is the future, don't > > > underestimate the difficulties it will bring. You can make a pretty > > strong > > > > It doesn't matter: you will have to deal with the problem sooner or later, > > anyway. The longer you wait, the worse it's going to get. > > > > > case that one effect of the move to parallelism will be to substantially > > > raise the skill bar for professional programmers. A mediocre programmer > > who > > > might be able to produce useful code in the old world will likely be a > > net > > > liability in the modern world, simply because parallel architectures > > make it > > > so easy to make catastrophic mistakes. > > > > AI is not a domain well-suited for mediocre programmers. > > > > > Finally, remember that consoles are approximately closed platforms, > > meaning > > > that you can't just buy one and start writing code for it. > > > > This is only true as long as Sony doesn't ship a Linux kit (it took them > > a while to ship one for the PS2, but I presume they learned something > > since). > > IBMs Cell-based blades (assuming, they will ever ship) will be probably > > not priced competitively, agreed. > > > > The current sweet spot is AMD64, especially dual-core. By the time the > > Cell > > ships it will be cheap DDR2 dual cores, or even DDR4 quad cores. > > > > -- > > 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] > > > > -- > > No virus found in this incoming message. > > Checked by AVG Free Edition. > > Version: 7.1.362 / Virus Database: 267.13.8/184 - Release Date: 11/27/2005 > > > > > > -- > No virus found in this outgoing message. > Checked by AVG Free Edition. > Version: 7.1.362 / Virus Database: 267.13.8/184 - Release Date: 11/27/2005 > > > ------- > To unsubscribe, change your address, or temporarily deactivate your subscription, > please go to http://v2.listbox.com/member/[EMAIL PROTECTED] ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
