Simon Lemieux wrote:
> 
> > You can still do a lot of things with such machines, but I don't think
> > rendering is one of them, unless you get several dozens of such PCs and
> > have the space at home to set-up such a rendering farm (and watch for
> > the electricity bill!).
> 
> Yes, that's what I thought...  I'll be looking for old used Pentiums instead!
> 
> Thanks,
>         Simon

  Just want to add my 2 cents in this discussion.

  I also considered salvaging older machines to do the grunt work for
me, and came to the conclusion that for me it wasn't worth it. If an
individual doesn't have the finances for buying any new hardware, then
it might be a good idea, as old hardware is certainly better than none.

  What I did realize however, was that todays machines are *so* much
faster than even the P100's or so of yesteryear, that it will take a
bunch of them to compete.

  A P133 scores around 50 of the infamous Linux BogoMips, while a
PII-350 comes in at 700. This is not the whole story obviously, but it
shows what you are up against. In addition to the raw CPU speed the
motherboard busses, chipset and RAM speeds are also a very influential
factor. My experience with distributed.net, Povray and similar number
crunchers shows that the raw speed scales roughly as the BogoMips
rating. Not precisely, but in the same ballpark.

  So if you took one of the new MB's with a pair of FC-PGA P-III sockets
and threw a pair of 700MHz P-III's, a stick of SDRAM, a network- and a
graphics card on it making it boot diskless, then you would in essence
have something that let a room full of P1XX's in the dust. It would be
worth at least 2800 BogoMips. Give the P133 the benefit of the doubt
worth a factor of two, and it would still take 28 of them to compete.
That is not considering the noise, heat output, electricity bill,
reliability (or lack of same), networking components, setup and
administration time, additionally needed expansion cards and RAM modules
etc, etc.

  I consider building two of these crunchers in a pair of ATX
miditowers, which are already standing next to the server that will feed
them. But right now my time doesn't even allow me to participate in this
discussion in a timely fashion, so time will tell if my interest in
raytracing gets strong enough for me to implement this idea.

  Regards

    Frank.

Post a reply to this message