Slashdot Mirror
IBM-Sony-Toshiba Reveal New Cell Processor Details
BBCWatcher writes "The three main partners in the Cell Processor initiative announced technical details of the new architecture. IBM's documents are particularly revealing. There's much more information on how developers, including open source developers, can access the SPUs (Synergistic Processor Units). As reported earlier, Sony will put the Cell into every Playstation 3 game machine, due early next year. And yes, Cell runs Linux."
who is the actual manufacture?
I heard a rumor awhile back that the PS3 will be running a stripped down version of Linux, just like the XBox/XBox 360 run a stripped down version of Win 2k. It does seem to make sence since the Cell processor runs Linux and NVIDIA(the PS3 will use a NVIDIA graphics card) has been known for great Linux and OpenGL support(I also heard all PS3 games will use OpenGL).
Sony offered Linux on the PS2 partly as PR stunt, partly as a way to offer a cheaper dev system than their $100K dedicated HW. But they distributed it with only a proprietary "BIOS"/bootloader, which meant developers couldn't distribute bootable discs even if they blew off the license which said they weren't allowed to, not without Sony's approval. And they distributed their proprietary boot disc only bundled with their $200 ethernet/HD. Plus it only worked with a select few "sync on green" monitors. So the whole thing was mostly a really tiny niche hobby, rather than a new Linux architecture. Let's hope Linux on PS3 has a chance to play with the big dogs.
--
make install -not war
I wasnt thinking of running some sort of desktop OS/languages that we think of today, but starting work on something that would take advantage of the cell architecture.
---- Booth was a patriot ----
When your "technical specs" are filled with marketing buzzwords, you KNOW you are in trouble.
These look like multi core CPUs with modified Altivec instructions to handle some extra elements.
My impression is that this is an optimized chip for situations where you have a known compiler (no branch prediction) plus known hardware and workload (games + gpu).
So they are likely to get swank game performance, but not sure this is a revolution as much as a nice optimization for a specific tasks.
Not to be a buzz kill, but it looks like we'll have to wait for a lot of development and middle ware maturity before we see the real potential in cell processors.
Progress isn't made by early risers. It's made by lazy men trying to find easier ways to do something. -Heinlein
Not just buzzword compliant, but confusing as hell for those of us who have been in the know for a while.
To me, SPU always made me think "Scalar Processing unit", while PPE made me think "Parallel Processing Element".
Of course that's exactly backwards.
That, and I choke on words like "synergistic" because they peg my bullshiat-o-meter way off in the red.
In my opinion one of the coolest features of this architecture are the way the reciprocal estimate and reciprocal square root estimate instructions work.
In a single cycle you get 13 good bits of precision -- with the low order bits filled with information to be used by the floating point interpolate instruction.
You can get a full precision (32 bit ieee float) reciprocal in about 6 cycles, and a 1/sqrt in 7 or so. Oh, and that's 4 results in that time. Averaging 1.5 cycles per FP divide, and slightly more for sqrt. times 7, times 3.2 billion per second, and the bandwidth to feed it.
That's several orders of magnitide faster that you could do with any x86 part out there.
Ian Ameline
If PS3 runs Linux & Firefox & Thunderbird & Emacs & Open Office; and has access to a network and a hard drive, I will buy one and probably use it as my primary computer both at work and at home.
(from a former Apple / NeXT / Amiga fan who doesn't mind spending "too much" on interesting architectures)
First the SPUs have the ability to initiate DMA. That means they can do stuff like calculate memory-mapped addresses and request more data, or select different destinations for a calculation. Or even load in a different program to do specialized execution. All independent of the main processor. BIG improvement.
2nd is the integer instructions. They really have everything... shifts, rotates, all SIMD. One of the big problems with PS2 VUs was that you had to resort to real sorcery to do simple things like shifts. But these seem to be real actual general-purpose CPUs. There's nothing that really strikes me as "OMG, I can't believe they didn't include X! Idiots!" Branch prediction, maybe.
Intel and AMD, with market share to keep, have chosen to tackle the MHz-cap by going symmetrically multicore. The IBM-Sony-Toshiba alliance was free to tackle it differently, and I think they might be on to something. Remember that Cell is a family of CPUs, not just the one chip. Future versions might be even better suited for a PC.
* And by 'shine' I mean 'like a supernova'. The known demos of the Cell in action promise a leap in power (in this class of problem) bigger that getting an 8087 gave us back in the day.
"A language that doesn't affect the way you think about programming, is not worth knowing" - Alan Perlis
Linux developers (who had evidentally already had access to the documentation, but couldn't disclose their work until it was officially made public) sent out a new version of the patches to give programs access to the SPUs. It's looking pretty likely that, as soon as you can actually get Cell processors, there will be support in the toolchain and kernel for using them effectively (provided you actually have a task that benefits from massive vectorization with very little control logic, of course). This should be great for the photorealistic rusty steel Enlightenment theme.