Re: Diminishing bandwidth performance with multiple quad core X5355s - Mailing list pgsql-performance

From Arjen van der Meijden
Subject Re: Diminishing bandwidth performance with multiple quad core X5355s
Date
Msg-id 465081D6.1080306@tweakers.net
Whole thread Raw
In response to Re: Diminishing bandwidth performance with multiple quad core X5355s  (jlmarin <jlmarin@gmail.com>)
List pgsql-performance
On 14-5-2007 0:00 jlmarin wrote:
> I wanted to post this even if it's a bit late on the thread because
> right now I have exactly this kind of problem.
> We're trying to figure out if a dual-Quadcore (Xeon) will be better
> (cost/benefit wise) than a 4-way Opteron dualcore, for *our* program.

We've benchmarked the Sun Fire x4600 (with the older socket 939 cpu's)
and compared it to a much cheaper dual quad core xeon X5355.

As you can see on the end of this page:
http://tweakers.net/reviews/674/8

The 4-way dual core opteron performs less (in our benchmark) than the
2-way quad core xeon. Our benchmark does not consume a lot of memory,
but I don't know which of the two profits most of that. Obviously it may
well be that the Socket F opterons with support for DDR2 memory perform
better, but we haven't seen much proof of that.
Given the cost of a 4-way dual core opteron vs a 2-way quad core xeon,
I'd go for the latter for now. The savings can be used to build a system
with heavier I/O and/or more memory, which normally yield bigger gains
in database land.
For example a Dell 2900 with 2x X5355 + 16GB of memory costs about 7000
euros less than a Dell 6950 with 4x 8220 + 16GB. You can buy an
additional MD1000 with 15x 15k rpm disks for that... And I doubt you'll
find any real-world database benchmark that will favour the
opteron-system if you look at the price/performance-picture.

Of course this picture might very well change as soon as the new
'Barcelona' quad core opterons are finally available.

> As you say, Opterons do definitely have a much better memory system.
> But then a 4-way mobo is WAY more expensive that a dual-socket one...

And it might be limited by NUMA and the relatively simple broadcast
architecture for cache coherency.

Best regards,

Arjen van der Meijden

pgsql-performance by date:

Previous
From: PFC
Date:
Subject: Re: Postgres Benchmark Results
Next
From: Arjen van der Meijden
Date:
Subject: Re: Postgres Benchmark Results