SI
SI
discoversearch

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor.  We ask that you disable ad blocking while on Silicon Investor in the best interests of our community.  If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.
Technology Stocks : Advanced Micro Devices - Moderated (AMD)
AMD 249.66+7.6%3:59 PM EST

 Public ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext  
To: Daniel Schuh who wrote (4175)8/9/2000 6:00:53 PM
From: pgerassiRead Replies (1) of 275872
 
Dear Daniel:

RE: Tradition Pooled vs NUMA SMP

There is no additional programming required in a NUMA architecture than with a BUS SMP. All of the differences are handled with hardware. The only differences are in configuration, management, and optimization (performance). Beowulf was built using Linux as a base. Beowulf uses a switched ethernet instead of LDT to connect the CPUs. From many sources, Beowulf clusters are very good supercomputers. The biggest advantages are no new hardware is required and the PC architecture has extremely good price performance ratios.

Using LDTs instead of switched ethernet just means that bandwidth just went to 100x and thus very large Beowulf clusters become possible (say 8K to 16K CPUs instead of current 36 to 144 CPUs). Each Sledgehammer CMP CPU could have 8 gigaflops peak and thus, the huge cluster could have as much as 128 teraflops peak (now that could do nuclear blast simulations in hours instead of weeks).

AMD was going to do PTP, point to point, SMP anyway. All this does is move 2 CPU cores, a Northbridge, and possibly L3 onto one chip instead of three (although FSB speeds will be nearer to CPU speed (remember what happened when L2 was put on die?)).

Pete
Report TOU ViolationShare This Post
 Public ReplyPrvt ReplyMark as Last ReadFilePrevious 10Next 10PreviousNext