From owner-freebsd-hardware  Mon Feb 24 11:32:13 1997
Return-Path: <owner-hardware>
Received: (from root@localhost)
          by freefall.freebsd.org (8.8.5/8.8.5) id LAA10393
          for hardware-outgoing; Mon, 24 Feb 1997 11:32:13 -0800 (PST)
Received: from george.lbl.gov (george.lbl.gov [128.3.196.93])
          by freefall.freebsd.org (8.8.5/8.8.5) with SMTP id LAA10388
          for <freebsd-hardware@freebsd.org>; Mon, 24 Feb 1997 11:32:11 -0800 (PST)
Received: (jin@localhost) by george.lbl.gov (8.6.10/8.6.5) id LAA24844; Mon, 24 Feb 1997 11:27:26 -0800
Date: Mon, 24 Feb 1997 11:27:26 -0800
From: "Jin Guojun[ITG]" <jin@george.lbl.gov>
Message-Id: <199702241927.LAA24844@george.lbl.gov>
To: freebsd-hardware@freebsd.org, robsch@robkaos.ruhr.de
Subject: Re: Memory speed of P6-200 (256k)
Cc: skynyrd@opus.cts.cwu.edu
Sender: owner-hardware@freebsd.org
X-Loop: FreeBSD.org
Precedence: bulk

Message-1:
} >> Not good :-).  I have Triton-1 (ASUS P55TP4XE) with non-EDO RAM and the
} >> benchmark runs at about 119.2MB/sec (1MB = 1048576).
} 
} >Which BIOS are you using, and which revision of the motherboard?
} 
} Award BIOS from about 1/1/96 and 2.4 motherboard.
} 
} Bruce

Message-2:
} I found:
} dd if=/dev/zero of=/dev/null bs=1m count=2000 
} 2097152000 bytes transferred in 27.298230 secs (76823735 bytes/sec)
} The motherboard is a P6NP5 (Natoma chipset) with 64 MB EDO-RAM.
} The kernel is 3.0-current.
} 
} Is this value normal for a P6-200?
} If not, how can I speed it up?
} 
} TIA
} Robert

I do not know what is the orginal messurement, but it looks very interesting.
By invesgating such idea, I found that 2.1.x and 2.2 or higher have different
memory sub-system or dd implementation.

So, you guys should not worry any memory speed on your system, and there is
no thing can change the speed beside change the memory sub-system, such as
inteleaved memory system.

The command "dd if=/dev/zero of=/dev/null bs=1m count=XXX"

really does is to allocate 1MB memory and try to put zero into the piece
of memory and repeat XXX times.

Under 2.1.x-RELEASE, it looks like dd/memory-sub-system (more like dd)
uses 4-bytes aligned register to memory operation, and 2.2/higher use
8-bytes aligned register to memory operation.
I tested this operation under 6-different mortherboard and 4-different
CPU speed.  I have not compare the dd or memory sub-system between 2.1.x and
2.2/higher yet; but I am sure I got the right assumption. 

So, for these people want to know what is your best memory system performance,
refer to ftp://george.lbl.gov/pub/ccs/performance.ps, page 6-8, and look for
register to memory copy and memory copy.

Do be comfused by this dd command :-)

-Jin