Re: AMD64-based laptops from HP/Compaq are crippled

From: General Schvantzkoph (schvantzkoph_at_yahoo.com)
Date: 07/06/05 

Date: Tue, 05 Jul 2005 20:49:24 -0400

On Mon, 04 Jul 2005 20:46:58 -0700, alex goldman wrote:

> I've been researching my next laptop purchase, and whether it will run
> Linux, and completely accidentally stumbled upon an alarming issue that
> people need to be more aware about.
>
> Apparently (and if my understanding is correct), HP/Compaq disables
> dual-channel memory access in Socket 939 motherboards, making CPU <-> RAM
> communication slower, and overall performance less than what people would
> expect when buying a system with the outward characteristics of what they
> think they are getting.
>
> http://zv6000forums.com/viewtopic.php?t=148

OK, read some of the posts. Here is the likely reason that they turned off
dual channel memory, they fucked up the design of machine and it won't run
reliably in dual channel mode. Assuming that they aren't complete idiots
what happened is that they intended to use both channels. If they were
planning for it to be a single channel machine they would have used a 754
pin part, which is what they did in the nx6125 or the ze2000z. When they
got to the lab it didn't work when both memory channels were running at
the same time. Rather then re-layout the pc board they disabled dual
channels in the BIOS and shipped it. They probably have a pc board spin in
the works, when it's ready they'll change the model number and ship it as
a new machine. The current ones are never going to run in dual channel
mode, it's not a bug in the BIOS it's a bug in the printed circuit board
layout. If you are going to buy an HP laptop today buy the nx6125 or the
ze2000z, those are available with Turions which are lower power then the
939 A64. Also most Turions have 1M of cache, the only exception
is the ML32 so stay away from that one. As I said in an earlier post cache
matters, dual channel doesn't. I know you think that dual channel is
important but it's not. I challenge you to name one real world program
where it makes even a 10% difference. I did some extensive benchmarking
doing heavy duty simulations, FPGA place and routes, GCC makes. The 3400+
with 1M of cache and a single data channel was never slower then the 3800+
with two channels, a faster clock, but only 1/2M of cache. On Verilog
simulations the 3400+ was twice as fast as the 3800+.

Quantcast