On TechRepublic: 12 tech terms that make you sound old
BNET Business Network:
BNET
TechRepublic
ZDNet
TalkBack 3 of 3:
« Previous
You missed one
The Intel solution has another problem: due to the physics of high-speed memory busses, you can only hang 2 or 3 DIMMs on each channel. Thus, a dual-channel Intel system can handle at most six DIMMs. Assuming 1 GB/DIMM, that's 6 GB max memory.
In contrast, each AMD processor has its own 6 GB memory, so a 4-way Opteron can handle four times as much memory as the Intel. With Intel, as you add processors the memory doesn't increase, but with AMD the memory available increases with the processor count.
We've benchmarked these servers for our workloads, and they scale almost linearly: a job that takes 9.5 hours with a single CPU takes 5.2 with two and 2.7 with four (That's compared to 46.4 and 26.2 on our previous HP 64-bit servers with one and two CPUs respectively.) In other words, going from 1 to 2 processors with PA-RISC/HP-UX cost 13% over linear, but with Opteron/Linux going from 1 to 4 CPUs incurred the same 13% penalty. The Opteron/Linux combination is, therefore, about twice as scalable as the PA-RISC/HP-UX combo. This helps explain why HP is doing an EOL on PA-RISC and HP-UX.
Posted by: Yagotta B. Kidding   Posted on: 04/20/04 You are currently: a Guest | Members login | Terms of Use

Alert moderator to an offensive message

Subscribe to this discussion via Email or RSS

One problem  Enterprise Analyst. | 04/19/04
What Problem?  L A D | 04/20/04
You missed one  Yagotta B. Kidding | 04/20/04

What do you think?

SponsoredWhite Papers, Webcasts, and Downloads

advertisement

SmartPlanet

Click Here