The ND-100 was a bottleneck for the 500 series, and initially also for the 5000 series. But as time went by the 5000 series acquired intelligent I/O interfaces

The ND-100 was a bottleneck for the 500 series, and initially also for the 5000 series. But as time went by the 5000 series acquired intelligent I/O interfaces sitting directly on the Multi-Function bus on the 5000, and there was less and less of a bottleneck. The systems I worked on had to get the absolute last possible drop of I/O performance out of it, and by having the ND-100 do its own I/O in parallel with I/O on the 5000 the whole thing worked very efficiently in the early nineties. The bottleneck, near the end, moved to the 5000 (or 5700, as it were) CPU instead. A Sun 4 could run circles around the 5700 in processing time - at one point I ran all our Fortran code through f2c and executed it on the Sun. Wow. That's when I started to talk with the customer about a possible rehost (which we did, to an SGI).