[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Tinycc-devel] (no subject)

From: Dave Dodge
Subject: Re: [Tinycc-devel] (no subject)
Date: Wed, 13 Sep 2006 18:07:43 -0400
User-agent: Mutt/1.4.2i

On Wed, Sep 13, 2006 at 01:10:13PM +0200, Guillaume POIRIER wrote:
> Can't Opteron natively scale up to 8 sockets (so 16 core with
> dual-code) natively?
> Can't Opteon address already enough RAM for all realistic workloads?

SGI for example has hardware today that goes up to 512 IA-64 sockets
and I think 128 terabytes of RAM in a single machine.  That's one
computer booting one copy of Linux.  If you need more, they can build
clusters of these with multiple 10 gigabit/sec interconnects between
them.  The costs of course are insane -- high-end IA-64 is something
like $10K per cpu just for the chip, and you don't even want to think
about RAM prices since you have to use the highest-capacity sticks to
reach thoses sizes.  But people do build them.

> The only shortcoming I see with Operon (right now) is the size of its
> cache, not the size of the addressable memory nor the number of
> core...
> ... or am I missing smth?

I guess the real problem is that nobody's selling the boards to hold
them right now (please prove me wrong!).  One really nice feature of
SGI's stuff is that it's all modular.  If you find that you need more
cores, or more RAM, or more PCI slots: you buy another 2U box, cable
it into the rack, and reboot.  The system doesn't need to share a
single physical case/backplane/board for it to behave as a single

I think you can recently finally get an Opteron system at the same
size as the "small" Itanium machine we were able to buy two years ago,
and I expect to be at least testing on such hardware within a year.
It's only a matter of time before someone does build Opteron machines
much bigger and for a lot less money than an IA-64 equivalent.  I'd
love for SGI to start producing Altix-style bricks with Opterons in
them but I don't really hold out hope for that.

                                                  -Dave Dodge

reply via email to

[Prev in Thread] Current Thread [Next in Thread]