Pimpwerx said:
But let's make one thing clear. For all its FLOPS power, that's only single-precision math. For double-precision math (supercomputing), Cell is much less formidable. It's powerful, but not nearly the beast it is in single-precision math. Guess what the most common application of single-precision math is. Exactly.
Lets make something else clear, you have no fucking clue what you're talking about.
When talking of preformance per area per cost, the Cell|BPA pretty much destroys most other ICs. For example, the NEC
Earth Simulator, which untill Blue-Gene/L held the preformance crown with 35 TFLOPs/sec (Double-Precision), is made up of clustered nodes based on the SX-6 Series vector processor.
-Each SX-6 processor peaks at 8 GFLOPs (DP) and was a more limited-run IC (read costly).
-Each ISSCC-Cell, runs at 26 GFLOPs (DP) and will sell in a $300 game console.
3 times the preformance in a commodity processor over a set-piece processor designed for supercomputing is nothing less than a beast.
Pimpwerx said:
*sigh* Read what I said again. Cell is a FLOPS machine. Tell me, what consumer electronic applications need anywhere near 8 SPEs. AFAIK, Toshiba is looking for a low-powered version of the chip for their devices. What's IBM planning this thing for?
Anything dealing with HD multmedia, be it the movement, manipulation or viewing of it. Toshiba's demo of ISSCC-Cell decoding and scaling 48 SDTV screens to a 1920*1080p display is a good indication of the net movement in CE.
IBM has already committed to a line of BPA|Cell based servers.
"
CELL is not limited to game systems. IBM has announced a CELL-based "blade" leveraging the investment into the high-performance CELL architecture. Other future uses include HDTV sets, home servers, game servers, and supercomputers. " -
IBM Research
TVs? You don't need this kind of power for a DSP. PC processors? For who? Apple? Certainly not Intel or AMD. I wouldn't bet on it being a S/390 app either. You can't file "etc" under the category, "far reaching technical implications". That's a copout. There is a definite, immediate application, and it's not suprising that it's Sony that's doing it. That would be the PS3. And as I mentioned, and has been pointed out before, Sony doesn't plan on putting Cell into its tvs and other devices until further down the road, b/c PS3 production will be more than enough of a drain. For all the talk about the other applications, no one can come up with one. Someone should find me a Toshiba or IBM application for this chip so I can stfu already. I just don't see it. Again, no TV needs this kind of power. The most interesting thing was the comment that the PPE core could be replaced with some other core, even an x86 core. So that gives IBM some flexibility in farming this product out to some customers.
I'd state that you're just ignorant and should "STFU already" as you stated. Toshiba demoed such an application (decoding 48 concurrent MPEG streams and resizing for output on a 1080p display with 7 SPUs). Philips used to cite STI-Cell as an example of the architecture it wants to put into the future livingroom (I wish the presentation was still available) to power the HD streams it wants to put everywhere, from your portable devices to your OLED covered walls. Was quite a neat presentation for looking 10 years into the future.
And the reason isn't because there isn't an application, that's asinine. The problem is Cell is fabbed on 10S PD-SOI, Sony has limited fab capacity for this as they are only now finishing installation of the lithography equiptment at Nagasaki. It's not the same as, say, producing on the bulk CMOS5 90nm process that the EE+GS@90 uses, or ther PSP's IC uses.
Which, if you'd take note, currently supplies EE+GS@90's for all PS2s as well as for Sony's HDTVs. The applications are already there.
