Hi Folks:

In a recent benchmarking discussion it was suggested that one way to
"equalize" (in terms of testing times) older machines and increasingly fast
machines-- all the way up to supercomputers-- was to simply make the image
larger for the hotties.  Then use a correction factor.

The thinking goes:  Use a standard image, let a 486/33 render it at 320x240,
and let the monster render it at 320,000x240,000.    If they both take
exactly 5 minutes, say, the Hottie Monster is 1,000,000 times as efficient.
(One million times the number of pixels.)

My concern with this approach is that while we are scaling the image in two
dimensions, we are calculating lighting effects in THREE dimensions.

Mark Wagner, who mentioned the scaling approach, feels that antialiasing
would skew the results.

My question is:  What OTHER functions WILL or MIGHT make the compute times
go longer than the scaling factor for 2D (area) and more toward 3D (volume).

Armed with suggestions, I'd like to conduct tests.

Matt

Post a reply to this message

A suggestion: radiosity?

	Graham Redway

Matt Swarm wrote:
> 
> Hi Folks:
> 
> In a recent benchmarking discussion it was suggested that one way to
> "equalize" (in terms of testing times) older machines and increasingly fast
> machines-- all the way up to supercomputers-- was to simply make the image
> larger for the hotties.  Then use a correction factor.
> 
> The thinking goes:  Use a standard image, let a 486/33 render it at 320x240,
> and let the monster render it at 320,000x240,000.    If they both take
> exactly 5 minutes, say, the Hottie Monster is 1,000,000 times as efficient.
> (One million times the number of pixels.)
> 
> My concern with this approach is that while we are scaling the image in two
> dimensions, we are calculating lighting effects in THREE dimensions.
> 
> Mark Wagner, who mentioned the scaling approach, feels that antialiasing
> would skew the results.
> 
> My question is:  What OTHER functions WILL or MIGHT make the compute times
> go longer than the scaling factor for 2D (area) and more toward 3D (volume).
> 
> Armed with suggestions, I'd like to conduct tests.
> 
> Matt

Post a reply to this message