Degenerate Tri-Strips vs Tri-Lists Again

[V]

After reading Nvidia's presentations from the GameDeveloperConference, i have picked up conflicting information. In Richard Huddy's presentation he recommends using Degenerate Tri-Strips (via indexing) while the OpenGL optimizations indicate to *never* use Degenerate Tri-Strips. Perhaps that indication is for non-indexed strips?

It seems that if your mesh data is already arranged as a batch of Tri-Strips, then connecting those batches via a degenerate tri format reduces index-bandwidth and gives the driver less to fetch per-triangle into the cache-

Any clarifications or recommendations would be helpful- ATI? Matrox?

[jwatte]

How about writing a test case and timing it? I'd be very interested in seeing what results you'd come up with.

I also believe the main penalty is in the set-up overhead per call to DrawElements(), rather than the extra two vertexes to start a new strip, at least for hardware transform and lighting.