🚀 go-pugleaf

Joshy (joshuam@cc.gatech.edu) wrote:
> Pizza (pizza@cc.gatech.edu) wrote:
> : On 29 Apr 1997 20:39:06 GMT, ccocsjo@acmey.gatech.edu (Who? Me?) spewed:
> : Actually you're wrong.  The MMX chips have a larger on-board Cache (32K, I
> : believe) so they are about 5-10% faster than the equivalent P5 w/o MMX.

> : 5-10% faster in EVERYTHING.

> Oh, Wow! I pay a hundred bucks extra for 5 to 10 percent.  
> On the CPU curve that gets me less than two months.
> Oh, Boy!

Hmmm...  Going in cache size from 16K to 32K is *NOT* likely to provide
much of a performance increase (not even 5 to 10 percent) except for code
that fits entirely within a 32 KB cache that won't fit in a 16 KB cache.
There is very little code that does anything significant that will fit
mostly in a 32 KB cache.  I would think a 1 percent increase due to the
cache size increase would be much more likely than a 10 percent increase.

It won't even decrease cache misses by very much.  This is not to say it
is not an improvement, but the performance improvement won't be that much.

--
John Cronin 				phone: (404) 894-7563
Office of Information Technology Customer Support Center 0710
Georgia Institute of Technology, Atlanta Georgia, 30332
Internet: john.cronin@oit.gatech.edu