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.

Politics : Formerly About Advanced Micro Devices -- Ignore unavailable to you. Want to Upgrade?

To: kash johal who wrote (40608)	11/1/1998 9:35:00 PM
From: Ali Chen	Respond to of 1573816

kash <Perhaps you could enlighten us on how Alpha's do it.> That was great, thank you! Regards, - Ali

To: kash johal who wrote (40608)	11/2/1998 12:55:00 AM
From: Tenchusatsu	Read Replies (1) \| Respond to of 1573816

<I guess a guide might be how Alpha does 2-way,4-way, 8-way etc systems as the K7 follows that architecture and likely will follow the ALpha philosophy. Perhaps you could enlighten us on how Alpha's do it.> I don't know that much about Alpha servers, but I do know how expensive they are. As for how the Alpha does 4-way, 8-way, and higher, I explained it before in the Intel thread. For the 4-way server, I guess that four processors are connected to one crossbar. The crossbar chains four P2P connections into one SMP cluster, then connects up to the north bridge in yet another P2P fashion. The advantage is that multiple P2P connections have more well-behaved electricals than a bus, and the processor's own bus interface is less complex. The disadvantage is that the crossbar cluster kind of defeats the purpose of multiple P2P connections in the first place, since it effectively turns those P2P connections into a bus. Not only that, but the crossbar adds an extra component to the server chipset, not to mention an added layer of latency. Perhaps the Alpha has a very high appetite for data, meaning that each processor would have to be connected to its own north bridge. This is the ultimate in performance, but it's also crazy expensive and complicated. (I have no reason to believe that the K7, even if it beats the P6 clock-for-clock, will approach the Alpha's appetite for data.) Like I said before, P2P is not necessarily better or worse than an SMP bus, just different. P2P is awesome for uniprocessor systems, though. I wonder if Intel is going to create a uniprocessor Pentium II or Celeron which has a low pin-count P2P connection instead of the 64-bit GTL+ bus. Tenchusatsu