Tuesday, April 18, 2006
Mitosis and K10 speculative multithreading
Mitosis - Cell division principle in biology is applied for Intel's new multicore microarchitecture in 2009.
Replicating small cores with its own register file and speculative L1 cache, and using Global Register file+ Register Versioning Table, and version control Logic to glue multicore, boost a lot of gains in performance that might be obtained inside single die. Intel is pretty good in large L2 so, it is promising new architecture. Drawback is however, the lack of memory controller on this picture, that means still no CSI I/O for such a fast multicore !?.
Will AMD be competable by 2009?
AMD will do almost the same with its future K10 processor , that should be at disposal about the same time or somewhat earlier. However, K10 will use large integrated L3 (due to recent Z-RAM deal) that naturally fits into its SOI technology, but not into Intel's bulk. By the year 2008 Intel will use 45nm technology, that will leak a lot at higher temperatures and clock, but that will not significantly hit speculative multithreading directed, naturally lower clocked processors. And finally, Mitosis processors will be produced at Intel's 45nm fab in Qiryat Gat in Israel, currently under construction, less than 50 kilometers from troubled Gaza strip in Palestine.
If you can't wait so much, you have an option to join opened Sun T1(Niagara) multicore project that with Rock implementation, a year or two earlier, will offer very similar core collaboration concept.
This post link
Replicating small cores with its own register file and speculative L1 cache, and using Global Register file+ Register Versioning Table, and version control Logic to glue multicore, boost a lot of gains in performance that might be obtained inside single die. Intel is pretty good in large L2 so, it is promising new architecture. Drawback is however, the lack of memory controller on this picture, that means still no CSI I/O for such a fast multicore !?.
Will AMD be competable by 2009?
AMD will do almost the same with its future K10 processor , that should be at disposal about the same time or somewhat earlier. However, K10 will use large integrated L3 (due to recent Z-RAM deal) that naturally fits into its SOI technology, but not into Intel's bulk. By the year 2008 Intel will use 45nm technology, that will leak a lot at higher temperatures and clock, but that will not significantly hit speculative multithreading directed, naturally lower clocked processors. And finally, Mitosis processors will be produced at Intel's 45nm fab in Qiryat Gat in Israel, currently under construction, less than 50 kilometers from troubled Gaza strip in Palestine.
If you can't wait so much, you have an option to join opened Sun T1(Niagara) multicore project that with Rock implementation, a year or two earlier, will offer very similar core collaboration concept.
This post link
Comments:
<< Home
indeed csi is there:
what you think are the line which connects cores whith GRF VCL and L2 cache?
about vista: if needed you'll see a MS Service Pack enabling CSI in 2008
p.s a 3 to 5 yl is not bad for a short lived solution!
Post a Comment
what you think are the line which connects cores whith GRF VCL and L2 cache?
about vista: if needed you'll see a MS Service Pack enabling CSI in 2008
p.s a 3 to 5 yl is not bad for a short lived solution!
<< Home