Pascal Schmidt wrote in a message to Roy J. Tellason:

 PS> Hi Roy! :-)

 RJT> This is going back a few years,  but there was a local computer 
 RJT> store around here that at one time advertised a machine they called 
 RJT> their "Pentium-killer". It was a 486,  but of a special
design,  so 
 RJT> that _all_ of the ram in it was as fast as most cache in those days 
 RJT> was.

 PS> Well, the best one could do on a 486 would be to make RAM twice as 
 PS> slow as the internal cache. I've been reading old reports about the
 PS> 486 design lately, and operations with data from the cache take 1
 PS> cycle, while going to main memory has a minimum additional 1 cycle
 PS> penalty for going to the on-chip memory interface. Note that it
 PS> still would require glue logic between CPU and memory that could do
 PS> its job "instanteneously", which is impossible. ;)

I don't know,  as I said I never did get much in the way of details back then.

 PS> Now, maybe the comparision was against the 386s of that day, which 
 PS> had external cache taking a few cycles to do its work. Still much
 PS> faster than going to the real memory.

The comparison was specifically against the Pentium machines that were
being pushed everywhere at that time.

 RJT> I never did get much in the way of details as to just what brand 
 RJT> of MB or what kind of ram they were using in it back then,  but 
 RJT> it sounded pretty nifty...

 PS> Yeah, but that was still a CPU running at a rather low frequency 
 PS> (the fastest true 486 designs run at 120 MHz for AMD and 100 MHz
 PS> for Intel) compared to what we have now.

There's something different about the 133?  That's what I'm running in this
box I'm typing on now.  AMD part,  they call it a "5x86/133", 
although I believe it has some other markings on it as well,  P75 or
somesuch.

 PS> Doing something like that would be extremely expensive now. Intel 
 PS> charges excessive prices for Xeon CPUs with 2M of cache. That may 
 PS> be for marketing reasons, but in fact it is also not easy to 
 PS> manufacture a 2M cache that can run at GHz speeds.

I can't imagine it'd be easy to make *anything* that runs at those speeds! 
This is RF,  and fairly high frequency RF at that!  It really blows me away
that speeds are getting up into the regions they're into these days, 
although I have my doubts about how much the rest of the system can keep
up...

---