On 4/30/06, <b class="gmail_sendername">Robert Bradbury</b> <<a href="mailto:robert.bradbury@gmail.com">robert.bradbury@gmail.com</a>> wrote:<br>
<div><blockquote class="gmail_quote" style="border-left: 1px solid rgb(204, 204, 204); margin: 0pt 0pt 0pt 0.8ex; padding-left: 1ex;"><div style="direction: ltr;"><div><div>Not really! Hypertransport & multi-gigabit ethernet = good good good.
<br>Cell microprocessors are kinda good good good.<br>Highly parallel graphics cards are somewhat less good good good.</div></div></div></blockquote><div><br>
There are two different issues here, and you're making correct statements about one but overlooking the other.<br>
<br>
Architecture matters when you're choosing a system for your particular
job, and yes, you're right about which architectures are good choices
for things like molecular modeling.<br>
<br>
But in the long run architecture matters less than process technology.
The Athlon-64 I'm typing this on is far from optimized for molecular
modeling (which is what it's spending most of its time on right now) -
but it's much more effective at it anyway than an optimal design of the
past. Process technology accounts for most of the hardware R&D cost
and most of the performance improvement, and process technology is
driven by the aggregate demand for denser chips - and most of that's in
things like PC and console CPUs and graphics chips, PDAs and cell
phones, and RAM (and increasingly nowadays Flash - that market is
starting to pay for top of the line fabs, not just take the leavings).<br>
<br>
So yes, people using lots of transistors in "bad" architectures are
still paying their share. By all means encourage better architectures -
but let's also make the scarcity to abundance transition in computing
resources generally, and stop behaving like an animal that's spent its
life in a ten foot cage, and on release keeps pacing in a ten foot
circle.<br>
</div></div>