From: "Geo" 

"Tony Ingenoso"  wrote in message
news:42788f9b$1{at}w3.nls.net...
> If you've only got 16K of L1 and the chip runs 10X slower on an
> L1 miss, what good is the pipeline when its sucked instantly dry by
> the execution unit waiting on an L1 line load?
>
> Modern OOP bloatware will blast the crap out of that tiny L1.
> These machines aren't running nearly as fast as they're advertised.

These machines are designed to run bloatware, there is nothing else on the
market today. That's why they do this prefetch stuff and execute things in
advance, so they can see the 10x slower jump coming and start loading that
now before it's needed.

> There's more embedded CPU's sold than desktops anyway.  Intel
> still sells embedded 386SX/DX and 486SX/DX's.  This is a big
> market.

and I believe your tweaks are both needed and justified in the embedded
market, a change like the one you mentioned could make a huge difference in
a digital camera as it converts an image to jpg compressed format.

Geo.

--- BBBS/NT v4.01 Flag-5