Subject: Re: Prob with 2.0 build
To: None <current-users@NetBSD.org>
From: Matthias Scheler <firstname.lastname@example.org>
Date: 07/19/2004 12:27:10
In article <40FC1ED1.email@example.com>,
> Well looking at the mid and high end servers on hp's website shows up
> some serious hardware which is well capable of beating any existing
> Opteron system.
Yes, but I'm not sure if NetBSD works one those.
> I think they have to do more than just increase the number of registers,
> they need to abandon x86, ...
Why? I don't think that x86 is *that* bad. Many problems with existing
x86 system are related to PC architecture problems and legacy
> ... but I've read that RISC systems perform better than CISC (x86), ...
Which ones? Beating a 3.2GHz Pentium IV in single chip integer performance
is a hard job. I'm not sure if any RISC chip can do that, maybe IBM's
latest Power5 CPUs.
> .. i.e. designing RISC chips ...
Modern x86 CPUs are RISC chips which interpret x86 instructions.
> ... and optimizing code for RISC is more straight forward.
That's not true. Optimizing code for RISC CPUs is very difficult.
Just look at the lacking performance of IA64 or UltraSPARC binaries
compiled with GCC.
> I think that's the reason why Intel want to depricate x86 in favour of EPIC
Intel wants to depricate x86 because the Itanium architecture (IA64) is
covered by a lot of patents and they finally get rid off competitors like
AMD or VIA. And I believe that IA64 is going to die anyway. The performance
advantage over Xeon or Opteron system doesn't justify the price. And
thanks to AMD64 (or EMT64) the 4GB limit is no longer an issue.
Matthias Scheler http://scheler.de/~matthias/