Jed Rothwell wrote: > Stephen A. Lawrence wrote: > >> > There was a tremendous effusion of computer CPU and ALU architecture in >> > the 1970s and 1980s, as minicomputers and microcomputers competed. Now >> > there is only Intel. >> >> Really? That's odd -- at work we use an awful lot of X86-64 systems. >> Last I heard that wasn't an Intel part; Intel second sources under >> license, it but it's AMD's baby. > > Okay, there's only Intel and AMD. And maybe IBM. > > My point is that back in 1975 there were dozens of companies developing > CPU architectures, and the resulting computers were incompatible and had > very different performance characteristics. Whereas today there are few. > Perhaps MPP architecture will take off in the coming years and we will > be in a new era CPU design effusion. These things tend to come in waves, > or perhaps I should say punctuated equilibria. There were few > fundamental changes to automobile design between 1930 and 1990 and then > suddenly, boom, we have hybrids and electric cars.
Right, point well taken. Sorry, I was picking nits. (Something new and different!) Back when IBM and Dec introduced the first large-market RISC chips it looked like there was a real revolution in CPU instruction set design. And here it is 20 years later (give or take a few) and the Alpha is dead, the Intel 680 (or whatever it was called) never got out of the embedded systems market, the Motorola 88K series is dead, the Sparc with its innovative register windows is long gone, the Power is holding onto a few niche markets by its fingernails, and it's business as usual with most systems running with a stack of bandaids 9 miles high piled on top of the horrible old Intel 8080 architecture. (Can you run an X86-64 in 8 bit mode? I wonder...) The skill of the chip designers betrayed them -- when the Power instruction set was introduced nobody believed you could make a bloated-instruction-set CPU go as fast as a RISC chip, but the designers at Intel eventually proved that wrong, and so we're stuck with the same old same old massively updated but still hyperugly 8008/8080/8086-series architecture. The waves came, they hit the rocks, and they recoiled. Last I heard the X86 series still had exposed pipes in the FPU -- I mean, really... > > A sign of the diminished importance of CPU design is the fact that the > #5 fastest supercomputer in 2005 was the Dell Thunderbird at Sandia > National Laboratories. Dell is not an engineering-oriented company. They > take things out of boxes, assemble them, and ship them. I doubt they > have much influence on CPU design, whereas I'll bet Apple does. I wouldn't bet much on that. Apple had influence way back in the past, back when they were a really big fraction of the 68000 and Power chip markets and they were partnering with Motorola and IBM to design a "generic" MacOS CPU board. Then they got into a big spat with Motorola and IBM over licensing MacOS and since then I think they're just a me-too company when it comes to processor design. These days Apple ships nothing but X86-64 boxes and they account for such a tiny fraction of the market I doubt either Intel or AMD cares a cat's sneeze what Apple thinks of the chip. After all, what's Apple going to do if they don't like it -- switch back to Power? Don't think so. > > - Jed >
