[149] in cryptography@c2.net mail archive
re: Weiner machine
daemon@ATHENA.MIT.EDU (Dave Koontz)
Tue Feb 4 16:06:14 1997
Date: Tue, 4 Feb 97 06:33:50 PST
From: koontz@netapp.com (Dave Koontz)
To: PADGETT@hobbes.orl.mmc.com, cryptography@c2.net
An MP-2 with 4K processors has been clocked doing 300,000 crypt(3)'s/sec,
(14 MHz clock rate, narrow data paths), for password checking against
a library. That should scale well for brute force key search, call it
3/4 million DES ops/sec. If we had ever finished the MP-3, the clock
speed was 2.5 times higher, and execution efficiency was another 3 times
better, and scale that to 65,536 processors (although we were going to
do clusters - ARPA paid us for it), call it 900 million DES ops/sec.
You can do better with less - .35 micron or smaller, standard cell(or
full custom) should easily hit in the neighbor hood of 300 MHz. Thats
three chips. (Of course people are giving away MP-1s and MP-2s today.)
There are something like 22 NOR/NANDs in the worst case path for DES,
but they are all small nets. I could easily believe 50 MHz in the
Chip Express libraries. The pin count would be low, which is good,
because they don't have a lot of thermally efficient packages (at least
historically). The one-mask versions of their parts are supposed to
be delay matched, so they don't get any faster.
