On Sat, May 18, 2002 at 01:33:55AM +0100, Ian Molton wrote:
| ...
| Users dont care if 'feature X' has caveats or conditions as to its
| acceleration (generally speaking). They simply expect it will work in
| the 'common case'.
I think this is just as much a problem for users as for developers. For
example, you might say that a driver accelerates multitexturing, but if
the user runs a game with the texture resolution set high enough (or if
something else on the system is competing for texture memory), there
won't be sufficient memory to keep multiple textures resident. The
accelerated multitexturing feature can't be used, and what happens next
depends on the application and the driver -- the app might drop back to
a lower resolution (in which case the user wonders why he didn't get the
high-res texture he asked for), or texturing might be disabled (in which
case the user wonders why the image is completely wrong), or the driver
might fall back to software rendering (in which case the user wonders
why everything slowed to a crawl), or the driver might start swapping
things into and out of texture memory (ditto on slowed-to-a-crawl), or
the app might crash, or something else. In the end, the user is
frustrated because it appears that the system doesn't work as
advertised.
There are dozens of pitfalls of this sort, even in the most recent
generation of graphics cards.
Applications have a huge variety of different behaviors, and lots of
them let users tweak rendering parameters. So how do you describe "the
common case" for a nontrivial class of users?
| well, now that covers things from a 'users' perspective - ie. from a
| 'what card should I buy? is it supported? what features should I tell my
| mates my new baby has?', point of view.
For the most part, I think we'll want to do what the reviewers of most
complicated products do -- adopt a set of tests and publish both the
conditions and results for those tests. Like this:
Car: The Porsche goes from a standing start to 100km/hr on a
dry asphalt track in 5.0 seconds at sea level on a 20 degree
Celsius day.
Card: The GeForce Ti4600 does 11.7M tri/sec on the glean
polygon-performance test with 50-pixel triangles and 32-bit
color format.
Car: The Porsche pulls 0.98G on a dry asphalt skid-pad using
new OEM Pirelli tires and stock suspension.
Card: The GeForce Ti4600 has a blend-function accuracy of
0.48 LSB on the glean blend-function test for 8-bit color
channels.
...
In our case, I think we can do a little better than the typical
reviewer, because any user is allowed to contribute a test that checks
exactly what's most important *to that user*.
Allen
_______________________________________________________________
Hundreds of nodes, one monster rendering program.
Now that's a super model! Visit http://clustering.foundries.sf.net/
_______________________________________________
Dri-devel mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/dri-devel
