The Power of the NORC 3770

4x in the supercomputing world isn't significant. My old office mate wanted increases on the order of 50x. 8-16x only starts to get interesting. O(n^3) and O(n^4) are HEC applications. This is part of why 360-91s and 370-195s were marginal improvements. You have to play with exponents if you expect application improvement. If a maker goes for the 90% market as CDC did, then they will ultimately lose.

Well the real construction problem is to decide what to modularize. Be it hardware or software, we don't buttemble individual transistors by hand. Same goes critically for software. MP's failing was one of software. Parallel algorithms aren't.

People did do a little of that. And that would have been 1-3 of all that model made. The software wasn't there. Machines back then didn't even have names.

Yep true the von Neumann architecture bottleneck.

GHz. I'm a software type and at this level I leave to the hardware guys.

No idea on niobium.

Only one the shorter time horizons based on the infrastrucure of existing CMOS fab lines.

Depends on the algorithm and application. 1-2 for 1000 doesn't cut it with Congress.

Running Windows? On the slow laptop disks? The extra big people don't have the influence that had in the 50s and 60s.

--