> Thank-you very much for all the vendor info. I've spent tons of time
> looking at vendors for big-memory alpha systems and found nothing--now I
> know that spending a ton more time probably won't help!
>
> I am confused about the 8GB Xeon machines--are these SMP? How is the
> memory addressed?
I may be just plain confused, but I remember reading somewhere
that some of the new Intel processors (maybe some of these Xeon 450)
where actually changed to get a 36 bit adessing limit instead of 32.
I'm still wondering if I was dreaming, but having machine which are sold
with 8Gig of Shared memory (unless they don't follow Intel SMP specs
and use proprietary mechanisms) tend to prove that Intel actually
manufactures such beasts. If yes there is probably some initialization
magic to switch to this extended adressing mode ... Wasn't Intel supposed
to provide the Linux community with these kind of informations ?
In the opposite case they can probably clarify this point. A 36bits
adressing model would make perfect sense (except from a chip generation
point of view, rebuilding the masks, starting a new process :-( ).
I bet that even the 4Gig machines are also build with those chips
memory is getting inexpensive, but loosing all this space in the IO
range... Hum did Intel reinvent the 640+KBytes upper memory madness ?
Am I just dreaming ? Yeah sure, we will never be able to get such
a beast anyway !
Daniel
P.S. No I'm not dreaming, searching on altavista for +Xeon +36 +bits
+adressing popped out the following announcement ! Quote:
http://www.idg.net/idg_frames/english/content.cgi?vc=docid_9-65177.html
"The chip implements Intel's Extended Server Memory Architecture,
a 36-bit addressing mode that enables memory support beyond 4GB.
Using 36 bits to address memory provides head room for as much
as 64GB, Gupta said. "
P.P.S. Any of the "Gods" knows anything about supporting this extended
adressing range ? Contrary to what other posts seems to have suggested
this wouldn't requires to change completly the way user space is
mapped into the kernel memory, it certainly requires playing with
some MMU registers and some kernel structure, but this shouldn't
cost a large rewrite. Getting linux on those fast would certainly
be a good thing. Maybe the initial poster could try to clarify this
with the vendor and invites a good kernel hacker for some time to
play with the machine. People like VAResearch may be interested in
funding/helping such a development.
--
[EMAIL PROTECTED] | W3C MIT/LCS NE43-344 | Today's Bookmarks :
Tel : +1 617 253 5884 | 545 Technology Square | Linux, WWW, rpm2html,
Fax : +1 617 258 5999 | Cambridge, MA 02139 USA | badminton, Kaffe,
http://www.w3.org/People/W3Cpeople.html#Veillard | HTTP-NG and Amaya.
