08 April 2018

The Asymptote of Progress - part the seventh

Another note of asymptote existence. Anandtech just reviewed the latest SSD from Western Digital. For the first time in a while, they implemented their own controller. One of the comments complained that the review didn't reveal the entire design/implementation of said controller. Billy Tallis replied (in part):
We asked repeatedly, and all we could get was that it isn't RISC-V. But every other NVMe controller used in consumer SSDs uses Cortex-R, and there's no reason to suspect WD is doing anything different. There aren't many alternatives.

Which comment points to, at least, two asymptotic factors:
1) CPU availability (not the whole controllers, of which there are legion) boils down to something ARM or RISC-V
2) The designers assert that new principles are becoming rare in instruction set design, as the most successful designs of the last forty years have become increasingly similar. Of those that failed, most did so because their sponsoring companies failed commercially, not because the instruction sets were poor technically. So, a well-designed open instruction set designed using well-established principles should attract long-term support by many vendors.

That second is from the wiki article on RISC-V. This is the quote's link. That old Zeno problem, stepping halfway to the wall. Or as stated in recent missives, CPU is a manifestation of maths, and in the end there will be one "best" resolution to a maths problem. You really should read the whole thing. One, therefore, might wonder how much microarchitecture "evolution" is pointless wheel spinning?
If you are one of the few hardware or software developers out there who still think that instruction set architectures, reduced (RISC) or complex (CISC), have any significant effect on the power, energy or performance of your processor-based designs, forget it.
Not to say that engineering of chip production has hit the wall (it is close, though). (And an amusing stackoverflow discussion.)

The commoditization of even CPU and OS might lead one to ask the next question. What becomes the high-value/high-wage occupations? Do we really need to repeat the fiasco of FIRE burning down the global economy? But if real science/engineering becomes unnecessary (in the creation and production of consumer goods; cosmology need not apply), how should we educate all those little grey cells (yes, that's not original, and you should look it up) for what sorts of work? The soft questions are always the most difficult to answer.

And, it turns out, that RISC-V machines are being made. Not all is peaches and cream, naturally. A short tour of the innterTubes will dig up criticism. But CPU was, is, and always will be a manifestation of maths, and as every other type of maths, there's one best way to do it. Guido says so.

No comments: