Re: DECstation question.

Eric Young (eay@cryptsoft.com)
Fri, 1 May 1998 16:04:10 +1000 (EST)

• Next message: staym@accessdata.com: "fast modular multiplication inverse"

On Thu, 30 Apr 1998, Eric Young wrote:
> So the
> question is, does anyone know what the clock speed for a DECstation 5000/125
> was, or know how to find this out under ultrix?

Well, the answer was to use the 'pixie' profiling tool. As part of its
output, it lists the clock speed.

So for those interested in RSA in software and which CPUs are best, I've
compiled a table. The numbers are the number of RSA private key operation per
second assuming the CPU is running at 200mhz.

CPU type mhz 512 1024 2048 4096
Pentium 100 32 102.6 20.0 3.1 0.4 linux
PPro 200 32 142.9 29.4 4.8 0.7 linux
R3000 25 32 93.0 16.6 2.5 0.3 ultrix 4.3
R4400 200 32 73.0 13.9 2.1 0.3 irix 5.3 o32
R10000 180 32 182.1 35.7 5.7 0.8 irix 6.4 n32
R10000 180 64 326.8 74.6 12.6 1.9 irix 6.4 n32
DEC 21164 400 64 227.3 47.6 7.8 1.1 OSF

I only list these machines, because I've done assember for them and it is
apples to apples comparison (4 way loop unrolling, same algorithms) and the
scheduling is reasonable, +-15% perhaps. Under this table, the old R3000
looks quite good, but I suspect this is because that particular box has 64k
I+D caches. The R4400 only has 16k caches and the R10000, 32k. Still, the
R10000 is quite a fast multiply engine. I have only briefly looked at the
PPC, but it appears to have 32*32 -> 64 and/or 64*64 -> 128, so it will be
interesting so see what it can do. Still, considering the fastest R10000 is
250mhz, or 93.2 1024 bit RSA/s, and the 21164 can do 600mhz, or 142.8 1024 bit
RSA/s, the choice for a SET accelerator will still be a PC motherboard running
linux/NT and an Alpha 21164PC :-).

eric (perhaps sparc v8/v9 next?)

• Next message: staym@accessdata.com: "fast modular multiplication inverse"