Sun's UltraSparc-T1 (http://www.sun.com/processors/UltraSPARC-T1/)
makes some interesting trade-offs...  They ditched out of order
execution and branch prediction (making for a much simplified (and
thus smaller) processor core) and designed each core with 4 sets of
registers and enough logic to keep track of up to 4 threads, executing
instructions from one at a time, in order.  Fetcheds from cache take 4
cycles, so by the time a thread rolls around again, it's associated
data is there, ready and waiting.  A design like that, especially with
multiple cores, trades some memory (in the form of extra registers)
and some latency for nearly 100% CPU utilization so long as you have
enough threads.

--tim
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to