Silicon Graphics, Inc. (SGI) 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 : Silicon Graphics, Inc. (SGI)

SGI 91.87

+1.8%

Dec 5 9:30 AM EST

Public Reply Prvt Reply Mark as Last Read File

Previous 10 Next 10 Previous Next

To: John M. Zulauf who wrote (3640)	11/21/1997 9:54:00 PM
From: Jerry Whlan	Read Replies (2) of 14451

Q1. Are you saying that modern parallel program tools make the sort of hand work I've described unnecessary? Recently, there was an interesting discussion along these lines in one of the usenet groups. One well supported point was that the same kind of work required to port an application to a cc-NUMA machine and tune it was the same as that to port and tune it for a distributed memory machine like IBM's. The bonus of doing it for a system like IBM's was that using MPI (a message passing interface for distributed memory systems) was extremely portable, all high-performance vendors support MPI so it is not a big deal to move your code from that Cray to the IBM and a lot of the tuning you did for MPI on the Cray is just as useful on the IBM or the Origin. I think this reasoning is partially behind SGI's support for the OpenMP standard, as it would theoretically bring the same portability to NUMA tuned code. Q3. Are you saying you think the HP messaging, and SCI-ring Sun clusters are even nearly as easy to program well as a single image flat memory system? For those who don't know, SCI stands for "Scalable Coherent Interface" which, as the name implies, was designed to support cc-NUMA. Sun's SCI based products will initially not be cc-NUMA. However they have purchased the COMA (cache-only memory arch) patents of the now defunct KSR. I would expect to see Sun pop up with a COMA scheme on top of their SCI implementation sometime soon. This will provide a flat memory model. You also seem to be confused about the HP product. HP's technical boxes implemented cc-NUMA long before (years?) SGI. SGI has an edge on the latency to remote memory, but their nodes are only 4 (or 8?) cpus whereas HP's nodes are 16 cpus. The effect is that a lot more memory is local on an HP machine. As for HP's scalability, they announced an operational 256 cpu, single system image, machine at this week's SuperComputing 97 conference. And they too had a 128 cpu machine there, but I don't think they said how long it took them to set it up on the floor.

Report TOU Violation

Share This Post

Public Reply Prvt Reply Mark as Last Read File

Previous 10 Next 10 Previous Next