How the Pentium Fell Short of a 360195 4035

That is part of it.

If you go back to CACM about 1988 or so plus or minus a few years, Alan Karp then of IBM wrote a letter and started his $100 Karp Prize for 100x speed up on an MIMD code. Maybe it was 1986. I have it somewhere. It was won by Gustafson and Montry who introduced the concept of scaled speed up. Now I am some what skeptical of the argument, but who am I to argue how Alan spends his money? So Gordon Bell then started his annual prize at $2000 and various friends are judges, but the issue of scale still looms.

Part of the problem is that computer builders and systems designers think "things" are static. Far from it. The reality is that earlier simulations were yesterday's toys. Many where 1-D. Barely, then they went 2-D, then 3-D and then 4-D. Do you think the storage increments linearly like I just incremented dimensionality? That says nothing of the code. And if you are doing this on 1 CPU (could be small n cpus), then you have Ivan Sutherland pointing out to you that this is the von Neumann bottleneck: adding memory just made the problem worse. Thing started out simple, elegant, symmetric, spherical (point or circular), linear, and they evolved to the asymmetric, exceptional, ugly, confused, political, etc.

Another thing is real time robotics. It's not a problem of geometry.

--