Skip site navigation (1)Skip section navigation (2)
Date:      Sat, 22 Jan 2005 20:21:26 -0700
From:      Scott Long <scottl@freebsd.org>
To:        Sten Spans <sten@blinkenlights.nl>
Cc:        freebsd-amd64@freebsd.org
Subject:   Re: Would this system work correctly with FreeBSD 5.3 ?
Message-ID:  <41F31836.2010403@freebsd.org>
In-Reply-To: <Pine.SOC.4.61.0501230246080.13561@tea.blinkenlights.nl>
References:  <20050122141741.9690.qmail@web26804.mail.ukl.yahoo.com> <Pine.SOC.4.61.0501230246080.13561@tea.blinkenlights.nl>

next in thread | previous in thread | raw e-mail | index | archive | help
Sten Spans wrote:

> On Sat, 22 Jan 2005, Claus Guttesen wrote:
> 
>>> I'm buying a Quad AMD Opteron 64 bits, with 4GB
>>> memory. I'd like to know
>>> if FreeBSD 5.3 will support a quad-cpu system (with
>>> HTT, 8) correctly,
>>> without performance losses and consistent operation,
>>> no crashes etc. Is
>>> the 4GB memory still a problem in amd64 ?
>>
>>
>> I have a tyan quad-opteron @ 2 Ghz with 4 GB RAM.
>> Works fine, using it as a postgresql-server. Running
>> 5.3-stable. Very nice hardware.
> 
> 
> A short hardware blurp fyi:
> 
> - HTT:
>   Amd is indeed working on dual core cpu's,
>   samples are rumoured to be out there.
>   But there are rumours of heat issues and/or
>   socket changes. I guess we'll see later this
>   year.

Hyperthreading is not the same as dual core.  Hyperthreading
takes a single core and bolts on an extra L1, instruction
decoder, and register array.  Each 'logical' core competes
for the same execution units.  Ironically, HTT only works
because the pipeline on P4 is so long.  An expensive stall
gives the other core an opportunity to jump in and salvage
some of the cycles that otherwise would have been lost.

> 
> - Quad motherboards:
>   It's a shame that tyan connects all
>   the pci interfaces on their quad motherboard
>   to one cpu. The hp/compaq quad proliant has
>   seperate buses to two cpu's for extra bandwith.

This is really academic right now since FreeBSD does not take
topology into account when making scheduling decisions or
routing interrupts.  Without this affinity, it's only luck
if you happen to wind up on a CPU that is closer to the data.
Also, I don't believe that PCI buses are connected to CPUs
at all.  They are connected via HT-PCI bridges that act as
normal HT peers.  Only memory is connected directly to the CPU.
But I might be wrong.

Scott



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?41F31836.2010403>