[sdiy] Function Generator ICs

Magnus Danielson magnus at rubidium.dyndns.org
Fri Sep 21 22:11:36 CEST 2007


From: ASSI <Stromeko at compuserve.de>
Subject: Re: [sdiy] Function Generator ICs
Date: Fri, 21 Sep 2007 21:46:02 +0200
Message-ID: <200709212146.02895.Stromeko at compuserve.de>

> On Freitag 21 September 2007, Scott Gravenhorst wrote:
> > BTW, I don't totally
> > understand the "gate count" thing, I have the "500K gate" Spartan-3E
> > (XC3S500E), oddly though, in my polysynth design, the ISE says the
> > design represents 1,290,000 "equivalent" gates...
> 
> The gate counts don't really tell you anything useful unless you are 
> comparing them across the same set of tools.  In theory they should 
> give you an idea how costly it would be to implement your design as an 
> ASIC.  These numbers have been going through various iterations of 
> marketing fudge factors with each generation of the FPGA architecture 
> (not just at Xilinx).  A system gate is supposed to be equivalent to 
> the sea-of-gates available in some mask programmed gate arrays.

It used to be a almost meaningfull measuring method. But with the LUTs, CLBs
and other larger blocks, the lines have become fuzzier. I just don't use it
anymore for FPGA work. It is a meaningless measure now. Ths fit-size into a
certain architecture gives better feeling. 495 CLBs in Virtex 2VPX says more
(if you used and encountered one... :).

Also, for some things you do want to optimize the design for a better fit for
the architecture you are actually using. Engineering smart fits costs more, but
the gain can be significant. It may come in speed or in size of fit.

Another aspect is the convergence of a synthesis. For larger designs, you
really don't want to use much more than 70% of the CLBs anyway. If you want to
do more, the need for engineering smart fits grows, and the tricks to pull out
of the hat may include steering the placement.

I agree fully with Achim that there have been a large amount of marketing
fudge around those numbers over the years. I think it has become less so
nowdays. Still, the numbers are more or less bogus regardless if they are
overly optimistic or conservative.

Cheers,
Magnus



More information about the Synth-diy mailing list