Lpc2000

Do you really need the extra memory of the 2138?  Why not spend 
about the same money for an LPC with an external memory bus?  That 
would provide (I believe) the fastest possible parallel interface to 
the ethernet chip.  Or am I missing something?  It seems really 
weird to go through a third chip (FPGA) with a serial interface when 
what you are after is speed.
James 

--- In lpc2000@yahoogroups.com, "Anton Erasmus" <antone@s...> wrote:
> On 30 Jan 2005 at 6:30, Rick Collins wrote:
> 
> > 
> > 
> > --- In lpc2000@yahoogroups.com, "Anton Erasmus" <antone@s...> 
wrote:
> > 
> > > In your previous posts you said you are considering ADDING a 
FPGA to
> > enable
> > > you to easier / faster access a ethernet controller using a MCU
> > without external
> > > bus. 
> > 
> > Boy, leave it to engineers to nit pic.  But if you reread my 
message,
> > that is not what I said.  I said I would "use" an FPGA to do the
> > conversion.  
> 
> :-)  Yes,  I have been involved in many HUGE arguments between 
engineers,
> where someone made a statement, and because of some sort of slip 
misused 
> a word without realising it. Everybody else picks up on this, and 
even though they
> fully agree with what the other guy meant to say, they do not 
agree with what
> he actually said.  Non-Technical people who listen to this, find 
this totally beyond 
> comprehension.
> 
> Anyway, I accept your argument.
> 
> > > If you already have an FPGA, and it has enough spare capacity, 
then
> > it
> > > makes sense to use it. Even if you can use SPI interface which 
is in
> > the order
> > > of 5MHz, it will also be a lot slower than the ethernet's 
10MB/s. 
> > 
> > I didn't say I would use SPI.  The Atmel SAM7 chips have an SSC 
port
> > which is similar to the serial ports on DSP chips and will 
interface
> > directly to many codecs at very high speeds, >10 Mbps.  
> > 
> > > The overhead of
> > > accesing a normal ethernet chip together with all the data you 
have
> > to handle
> > > as part of the TCP/IP stack means that you will not get that 
high a
> > speed overall.
> > > With the Wiznet chip, even though the I2C is fairly low speed, 
you
> > ONLY need to
> > > transfer data you actually are going to use in your app. The 
TCP/IP
> > stack overhead
> > > is handled within the Wiznet chip. Hence the ethernet 
interface has
> > got no overhead
> > > asociated with it, until there is data for the specific socket 
you
> > have opened. 
> > > In a previous message someone pointed to an Olimex LPC board,
> > together with
> > > one of these Wiznet chips, where they could serve web pages 
using
> > the I2C
> > > interface at 350kb/s if I recall correctly.  using 14 port 
pins to
> > emulate a parallel 
> > > interface, should be even faster.
> > 
> > Certainly this is interesting.  But like I said, it is a far cry 
from
> > 10 Mbps.  The overhead is not that great and regardless of how 
much
> > overhead you have, the time required to transfer the data across 
the
> > CPU/LAN chip interface will still add to that.  So having a 20x 
higher
> > interface speed is still a significant boost.  
> > 
> 
> The higher speed serial interface would definitaly help. What is 
the cost of the ethernet 
> controller you have in mind compared to the Wiznet W3100A chip ? 
If the pricing are 
> similar, it might still be worthwhile using the Wiznet chip in 
parallel mode via the FPGA
> and high speed serial combination. To get bootstrap code going to 
re-program or boot 
> the board via ethernet needs very little code because of the 
hardware TCP/IP stack. If 
> one then uses an RTOS or something else with a full TCP/IP stack, 
then one can use 
> the W3100A chip as a normal type ethernet chip. Opening a TCP/IP 
socket using the
> W3100A takes something like 20 lines of C code.
> 
> Regards
>    Anton Erasmus
>  
> 
> -- 
> A J Erasmus