On 3/3/06, James Richard Tyrer <[EMAIL PROTECTED]> wrote:
> James Richard Tyrer wrote:
>
> > The GX 533 costs $80 so it would probably make the price point but might
> > now be fast enough.  Note that there is a rule of thumb that the selling
> > price of a production product is less than the parts to make one would
> > cost.
>
> Correction TYPO: that is 'might *not* be fast enough'.

I think it would definitely not be fast enough.  You'd be better off
with no graphics acceleration and using the host CPU to do all the
computation.  You might point out that the peripheral bus (PCIe or
whatever) is a major bottleneck, but even in 2D graphics, there's
often enough computation overhead that the raw data movement over the
bus would be faster than an embedded processor could compute it. 
You'll gain some efficiency by parallel processing, but that's about
it.  And it's pretty well established that you'll get far greater
performance out of a specialized design in an $80 FPGA than you will
out of an $80 CPU.

This reminds me of this one 2D GPU design I was working on once.  The
idea was to create a middle ground between pure software rendering and
full hardware acceleration.  Even in simple graphics processor
designs, you could never saturate the GPU's memory bandwidth by doing
software rendering (unless the memory were really slow).  So I thought
about creating a hardware renderer that had very simple span-oriented
primitives.  By encoding the software rendering in this way, you could
typically get fast GPU performance.  The drawback, of course, is that
you're saturating your host CPU and bus utilization, but you would
have been doing that anyway with a software renderer, only worse.
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)

Reply via email to