imgod2u

Distinguished
Jul 1, 2002
890
0
18,980
The L1 data cache is 8 KB with a 2 cycle access latency for a 64-byte cacheline with accesses every cycle.
The L2 cache is 512KB for the Northwood models with a 7 cycle transfer latency for a 64-byte cacheline and can be accessed every 2 cycles.
There is no traditional L1 instruction cache. Instructions which are decoded are stored in the execution trace cache and all further calls to those instructions are taken from the trace cache in the form of already decoded micro-ops. The trace cache stores 12k micro-ops. Each micro-op is similar to 1 or more x86 instructions depending on the complexity, which themselves vary in size from 1 bit to 86 bytes.
There's no real way to compare the two caching structures (based on size and speed), they just work differently.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
 

juin

Distinguished
May 19, 2001
3,323
0
20,780
I read on chip architect that Presscott have 16 KB of trace cache.

They have increase the ligne size from 3 uops to 4upos so it mean that also a increase in bandwith to 4 instruction have occur per cycle.


Simple reminder or new tecnical information

A cpu fetch cache ligne and bit bus is the bus to the core so the bottle neck is the ability to fetch as many as posible ligne of cache.<P ID="edit"><FONT SIZE=-1><EM>Edited by juin on 03/07/03 11:45 PM.</EM></FONT></P>