Advanced Micro Devices - Moderated (AMD) Message Board

STOCKTALK

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 207.67

+2.2%

Jan 12 3:59 PM EST

Public Reply Prvt Reply Mark as Last Read File

Previous 10 Next 10 Previous Next

To: Joe NYC who wrote (11125)	10/2/2000 3:50:14 PM
From: combjelly	Read Replies (2) of 275872

Joe, if all you want is bunches of memory, then a NUMA style system would not be a lot of help for the latency reason you noted. I haven't seen anything as to what the latency would be for a request from two directly connected north bridges, but I have to assume it is high. It is my understanding that most applications that require very large data sets often have a fair amount of parallelism also, so a box with pots of memory and a single processor might have only a limited market. Looking at the specs for LDT, I would guess that you would have motherboards with 2 or 4 processors and additional motherboards would plug together with cables or a backplane that is on the edge. This way an 8 way would be fairly cheap and could be cooled without going to extreme measures like water cooling. The nice thing about NUMA is that it looks like a SMP system from the application (or OS) point of view. Now if you want to optimize to reduce communication overhead or latencies, then you need to know something about the layout. Not really sure if it would be worth the coding effort though.

Report TOU Violation

Share This Post

Public Reply Prvt Reply Mark as Last Read File

Previous 10 Next 10 Previous Next