greg at franklin.burnham-inst.org (Dr. Greg Quinn) writes:
> Does anyone have any recommendations or experience about the current
> dedicated sequence analysis computing systems (eg. Paracel, etc) in
> terms of effectiveness/maintenance/expandability in relation to cost.
I cannot say anything about dedicated sequence analysis systems, but
I did recently look at some timings for conventional unix workstations.
Here are the numbers on Work/Mhz for the different architectures for
the ssearch3 (Smith-Waterman) program.
Smith-Waterman cells/sec/Mhz: cells*10^-6/sec/proc OS cost
PowerPC 604e (300 Mhz): 58,365 17.5 (MkLinux) $1800 (300 Mhz G3*)
Alpha EV4 (275 Mhz): 54,777 15.1 (DUnix)
Alpha EV5 (533 Mhz): 65,448 34.9 (DUnix)
Alpha EV5 (533 Mhz): 57,305 30.5 (Linux) $3500 (linux only)
SGI R10K (180 Mhz): 72,277 13.0 (Irix)
Intel PPro (200 Mhz): 86,336 17.3 (Linux) $2000 (450 Mhz PII**)
*A 300 Mhz G3 machine is probably 1.5 - 2.0X as fast. **The 450 PII is
probably 2X as fast as the 200 Mhz PPro.
At the moment, a dual processor 450 MHz PII Linux machine for $4,000
should provide about 70*10^6 cells/sec - impressive. 30-40*10^6
cells/sec is available on PowerPC G3's and 450 MHz PII's for $2,000.
Bill Pearson
