I've noticed a lot of this truism-spouting nonsense going about on the forums. Why is it that somebody always goes "BINGO!!! WINNAR" and then essentially restates a narrowly focused obvious truth? Yeah, "being of a narrow focus" makes a DSP chip better suited for sound processing than an Intel chip, after cost analysis. A CD-ROM used for read-only applications will be cheaper and more effective if you cut out write-related functions. Jam is better on toast than caviar, and the reverse is true on crackers. SO WHAT?Originally Posted by neuropolitique
Bingo, we have a winner! Microcomputer CPUs are general purpose processors. Being of a narrow focus gives custom chips many advantages over multi-purpose chips.
Finally, I've already shown that - in today's world only, of course (things WERE different in the past, and will possibly be different again soon), late 90s-era Intel design proved good enough for Microsoft's purposes for a number of reasons:
1.) Complexity of modern code. Think of the applications being run on a single processor in PC games: sound displacement and filter types, fractal geometry and generated textures, AI. Many of these applications can be offloaded, but instead of being a "must-do" point, that allows me to segue into my next point -
2.) Availability/cost analysis. First off, Microsoft has to give somebody a good reason to create something new, and that good reason might be a cash prize of some sort. In this case Microsoft decided that the old Celeron was a Good Enough processor for their purposes.
3.) Creating a coherent design. The Saturn scared the industry away from having lots of small processors doing things, I think, but for good reason. When you have one strong general purpose processor, that allows people to come up with a wide range of solutions for their problems. Nobody wants to limit a console's functioning out of the gate these days.
Before I continue, it seems that there's some misunderstanding about RISC versus CISC. RISC was a specific type of processor, but the general idea was that the instruction set was reduced to the bare minimum, not expanded. CISC, of course, is the complex version. There's nothing RISC at all about a modern video card's processor and techniques - the complexity of any single architecture is about as baroque as you can find.
RISC has had its heyday. As applications of all sorts are relying more heavily on brute force these days, CISC processors take the guesswork out of a hardware developer's solution.
In fact, the classic example of a tightly focused application running on narrowly focused hardware - polygonal rendered graphics on 3D accelerators - is starting to OPEN up to a CISC approach, NOT go the other way around. The amount of RAM on a new graphics card dedicated to providing program space for creating custom shaders is still far below the 1MB mark (and will likely be there for a while), but the fact is these features aren't being added for laughs - this increase in processing power reflects the need for specialized applications to borrow techniques from traditional CISC computing.
Reply With Quote