My first seti machine was a pentium 133 on a 430fx board with 32 MB EDO ram.
I think you may be content with your PIII. Hours is good, with the version 2
screen saver my (first) machine too 90-100 hours to process a WU. (It was
then upgraded to P166MMX and 48MB, then to a TX board, 128MB EDO and a
200MHz Cyrix 6x86 PR266), not to mention an added Duron 750@824 machine.
Trying to envision the technical and economical realities of such solution,
I guess what could be done is devising a unified cache as large as the
entire dataset (about 16 MB), that should reduce penalties from reading main
memory and/or accessing disk, and a dedicated FPU that would be optimized
for the specific FFT algorithm, in order to be as small as possible (the
cache would take the most area of the chip about 95%, and so would be its
contribution to cost). It would operate in a mode similar to 3D
accelerators - telling the OS that FFT routines should be commited to the
seti chip, like direct3D or O/GL routines are directed to 3D accelerators.
The rest of the program could run on main memory, but it's most time
consuming routine would be performed (hopefully) faster.
Regarding the "obsolete" PIII architecture, it could be argued that PIII
chipsets are getting obsolete, but comparing with P4, the processor
architecture is much more efficient in terms of instruction per cycle. maybe
the HT variants are closer to PIII in this regard. a PIII tualatin can crush
a PIV, when both are running on the same freqency and similar memory
subsystem technonlogy (eliminating diversity in areas such as bandwidth).
The new mobile processors generation (centrino / banias) draws more on the
efficient design of PIII (and enhaces it) than the (too) long pipeline
PIV's.
Now, as per cost, i think implementing such a large cache would make it
quite expensive, but it could be made to run with just any processor, as
weak as a pentium...


"Filipe Santos" <filipesan@sapo.pt> wrote in message
news:3f536a98$0$22650$a729d347@news.telepac.pt...

Hello,

I own a Pentium III machine, so my units take hours to process. I blame

the

obsolete Pentium III architecture and I thought that a dedicated hardware
expansion card could be a solution. The card could have its Seti math
processor and all the cache necessary. The most expensive ones could also
support dual Seti math processors and other technology wonders.
I guess that the driver of this hardware device could be the well know
screen saver.

Why such a card hasn't been made?

What could the card have in its hardware to make it more powerful?  Any
ideas?


Filipe