[HARDWARE] Massive computing power

Joe Zbiciak j-zbiciak1 at ti.com
Sun Sep 27 01:51:11 EDT 1998

'Troy Thoele' said previously:

| Check out Digital Signal processors.  These CPU's are self contained,
| cheap, and easy to run in multiples.  Right now I'm working with Analog
| Devices SHARC processors, and this thing does 120MFLOPS at 40Mhz.  TI
| makes a chip that approaches a GFLOP on a single chip.

Bear in mind that FLOPs isn't as important as integer performance.  An
integer DSP (such as a TI's TMS320C8x or TMS320C62x) would be better suited
to the job.

Awhile back, I did some minor benchmarking of the C62x on RC5 cracking,
and I was moderately impressed.  However, I don't think I can really
post any benchmark results, since I'm a TIer.  :-)

I will say that the C8x's Parallel Processor has almost the exact
hardware you'd need to perform half of an RC5 round in one cycle, if I
recall correctly.  With multiple Parallel Processors (four on a 'C80,
two on a 'C82) and one RISC processor on one die, you should be able to
chow through keys like they were nothing.

| I can't get the RC5 core code ported to this chip yet because the
| creators (in thier infinite wisdom???) chose C++ rather than ANSI C for
| the core routines.  That's where I'm stuck at.  

What you could do is port _just_ the cracking core to the DSP.  Run the
rest of the client on a "host" machine.  That way, the DSP looks like a
function call to the "host" machine.  All of the C++ could live on your
PC, then, and only the assembly language core and some wrapper code which
communicates with the PC would be required.

After all, unless you have TCP/IP and a hard-drive hooked to your DSP,
you'll need some external machine to handle this stuff for you anyway. :-)



 +------ Joseph Zbiciak -----+
 | - - j-zbiciak1 at ti.com - - |  "The meaning of a value is determined 
 |-Texas Instruments, Dallas-|   by how it is used."
 | - #include <disclaim.h> - |                          -- Ousterhout
To unsubscribe, send 'unsubscribe hardware' to majordomo at lists.distributed.net

More information about the Hardware mailing list