problem installing fedora 2 x86-64 on a intel xeon 64 bits

James Wilkinson james at westexe.demon.co.uk
Tue Jul 6 23:19:21 UTC 2004


ne wrote:
> Seems I stand corrected. I did some Googlle research myself and saw
> http://www.pcpro.co.uk/?http://www.pcpro.co.uk/news/news_story.php?id=59918.
> I quote:
> 
> 'With its newly announced 'Nocona' Xeon processors - which clock 
> speeds ranging from 2.8 to 3.6GHz - Intel brings 64-bit memory 
> addressing to its 32-bit server processor. The aim is to provide 
> greater support for the high-level number crunching tasks of the 
> workstation market.'
> 
> So I guess the question becomes do FC kernels have support for
> this chip yet?

Jeff Vian objected:
> This is 64bit memory addressing on a 32bit processor. This is PLAINLY
> stated above.  It is not likely the 64bit kernel will work. 
> 
> I also would guess the chipsets to support that are very new.

Umm ... no. And yes.

Decoding Intel statements is often an artform.

In this case, Intel has literally spent billions of dollars on its
Itanium range, and billions more on support for it. Sales have not
been impressive.

Intel has been carefully attempting to preserve sales on its 32 bit Xeon
workhorse range (which is extremely lucrative: until last year Intel had
basically no competition on the PC-based server market except at the
very low end), while spinning its 64 bit Itanium processor as being what
you really need if you want real computing power [1] while remaining
compatible with "the industry-standard 32 bit Intel architecture"[2].

AMD's Opteron has basically spoilt this. It is very competitive with
Xeon on single-CPU systems, scales much better than Xeon (largely due to
HyperTransport and the NUMA memory architecture), really has superb 32
bit performance, and has 64 bit options (with the promise of even better
performance). So lots of people have been interested.

Intel has responded by adding a (mostly) Opteron-compatible [3] 64 bit
mode to recent Pentium 4 based processors, and has started turning it
on in some of the most recent Xeons. It now has a problem: how does it
advertise its new 64 bit chips without cannibalising sales for Itanium?

Its answer is statements like the one above: it's trying to portray the
64 bit mode as solely about letting Xeons efficiently address lots of
memory, whereas Itanium is still about real power and the future of
computing, and what you want if you want real 64 bit support. Honest,
guv.

In fact, although there are areas where Itanium makes sense, the market
is limited. At the moment, it's largely limited to more than eight-way
SMP boxes, in roles where they need fast access to shared memory (so
clustering won't work).

Itanium was supposed to support this market, but it was supposed to do
so with chips that could be sold in PC quantities, and the forecasts for
the project assume these economies of scale.

In the real world, the Intel AMD64 compatible processors are just as
much 64 bit processors as their AMD counterparts.

Unfortunately, there are some hardware level differences between the
two platforms, which need to be reflected in the way the kernel handles
them [4]. I know there are patches in the current upstream kernel,
but I don't know whether they made it to the default x64-64 FC2 kernel,
nor whether the surrounding software knew about the possibility of
Intel x86-64 processors. So you're probably right, Jeff: these CPUs
may not be reliably handled in 64 bit mode until FC3.

HTH,

James.

[1] Until recently this was a joke. These days, Itanium is becoming
competitive at certain tasks, especially those requiring lots of FPU.

[2] But not if you want any speed out of your 32 bit programs...

[3] It's supposed to be as compatible in 64 bits as it is in 32 bits:
at the moment, AMD doesn't have SSE3, while Intel doesn't have 3D-Now.
This is reputedly at the insistence of Microsoft, who didn't fancy
having to port to *four* processor families...

[4] For example, Intel chipsets also don't have the IOMMU that 64 bit
AMD CPUs have.  This isn't strictly an instruction set compatibility
issue, but the IOMMU is supposed to be very handy in supporting existing
hardware on systems with more than 4 GB of memory.

-- 
E-mail address: james@ | "Come on, son, give us your best shot."
westexe.demon.co.uk    |     -- Goliath





More information about the users mailing list