Silicon Investor (SI) -- The First Internet Community

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 : Intel Corporation (INTC) -- Ignore unavailable to you. Want to Upgrade?

To: Tenchusatsu who wrote (88167)	9/12/1999 8:35:00 PM
From: Rob Young	Read Replies (1) \| Respond to of 186894

Tench, <needs a chipset> Yes of course. Guess that wasn't my point. How about: "needs a dumbed down chipset as the memory controller and network (CPU<->CPU) controller are on-chip... thereby a relatively cheap server". <Plus, with the way you're describing the configuration, it sounds like this system will be true NUMA> Actually better than NUMA. In fact their is a presentation entitled "Alpha Microprocessor Futures" by senior consulting engineer Bill Herrick available at The Register. The Register is by far the best source of Alpha info... After all, the vast majority of the others are Wintel lapdog rags (dig, dig). Anyhow, if you click on slide 37 you find the following (typing this in for your reading pleasure by the way): Problem: - Shared memory bus limits bandwidth, increases latency Solution: - Distribute memory across processors o High performance local access - Maintain shared address space - Interconnect processors in a mesh o High speed point-to-point connections o Directory-based coherence o Fast, high bandwidth global access When Pete Bannon presented at last year's MPR Forum and was fielding questions a Intellite posed: "Could you describe how you went about your directory coherence?" Smart question... directory based coherence WAS a very hard problem (from what I understand). Without missing a beat, Bannon replied: "Next question". I've been sandbagging a bit.. just how good is the directory based coherence and 21364 bandwidth? Well if you click on slide 40 you see a 4 processor box (which is why I brought up 4 processors to begin with): - RAMBUS memory - Memory size = 16 GByte - Memory bandwidth 40/20 GByte/sec (that's right! local versus remote hence the "high performance local access" found above. These guys sandbag ... I mean the low-performance 20 GByte/sec is pretty good, don't you think?) - Memory latency 60-100 ns Of course the latency is the big thing. That IS an outstanding number for a 4 processor config. I went and looked at others and this is from my old and foggy memory but I believe the best I found for current day is 270 ns with 4 processors. So what do you think, do you think the 21364 has a chance to sell up against the all-powerful all-mighty Merced? Silly question, eh? Rob