Re: MSI K8T Master-2 FAR Mainboard
From: Anton Ertl (anton_at_mips.complang.tuwien.ac.at)
Date: Sun, 31 Oct 2004 09:31:34 GMT
Mike Cox <email@example.com> writes:
>General Schvantzkoph wrote:
>> If I were building a dual Opteron system I'd pick a board that had
>> separate memory for each processor.
Probably for performance reasons.
>>The K8T Master-2 only has DIMMs
>> connected to one processor, the other processor has to access the RAM
>> through the hypertransport bus.
>I thought both chips access ram through the hypertransport bus.
No, one CPU accesses the RAM through its RAM controller (both with
more expensive boards).
>for threaded applications, having two processors is better than one. Next,
>if you have a processor bound to a certain memory dimm, than if you need to
>access lets say more than you have in that dimm, you will have to page out
>the memory to the hard disk.
I don't know what you mean with "bound" here, but in a Dual-Opteron
system where both CPUs have RAM attached, each CPU can access all the
RAM (and a process running on one CPU can allocate and use most of the
RAM without paging).
>If you are running postgresql with 40GB dataset, than that may be a problem.
If you are experiencing a lot of paging, a machine with a lot of RAM
may help a lot (so look for lots of DIMM slots).
OTOH, if the CPUs are always waiting for the disk, one CPU might be
sufficient, certainly in the board you suggested. In a board with
DIMM slots for each CPU you might still want to have a second CPU
because of its RAM controller.
>> Either get an Athlon64 939 system or get a
>> true dual Opteron board that has 8 DIMM sockets.
>I'm not interested in the 939 because it doesnt support dual processors and
>it doesn't take error correction RAM.
I have not researched Socket 939 processors, but Socket 754 Athlon 64s
certainly take and use ECC RAM (I have one, with ECC RAM, with ECC
enabled). I would expect Socket 939 CPUs to support non-registered
ECC RAM, too (but take a look at AMDs docs and the board's BIOS
options before deciding this).
-- M. Anton Ertl Some things have to be seen to be believed firstname.lastname@example.org Most things have to be believed to be seen http://www.complang.tuwien.ac.at/anton/home.html