Rob Jenkins (robj++at++sgi.com)
Tue, 25 May 1999 12:49:23 +0100
In general, performance will increase with tri strip length, as a rule
of thumb consider strips of less than say 4 tris to be 'short'. GLperf
numbers are a good way to visualize this, I found a full listing
internally but the external page
http://www.specbench.org/gpc/opc/glperf_publish/index.html doesn't have
the latest and greatest numbers, there might be more up to date ones
around though.
Anyway, on iR ( esp with > 1 RM ) the display listed and immediate mode
numbers differ quite a bit. The increase once you get to large numbers
of tris can also tend to go up in steps, with some strip lengths having
similar numbers, then suddenly the rate goes up - related to strip
lengths of multiples of 4 or 8 or something ( allowing some optimal path
through the GE's I expect ).
I expect the OCTANE will some greater perf with longer strips but
probably the increase with greater length flattens out sooner than with
iR2.
The PF Prog Guide, chapter 19 has some reasonable rules of thumb for
this stuff, for a true feel for what is right for your app/data though
you should eval the performance as decribed in that section, knowing if
you're fill or geom limited can help you work out how best to fill the
pipe to the max.
On the subject of index vs non indexed, I *think* that indexed could
potentially be slower as you have an extra level of indirection on the
CPU but whether this is significant depends on what the rest of the pipe
is doing.
Cheers
Rob
Robert Stein wrote:
>
> Hi all,
>
> I have an application that draws large areas of adaptive terrain, and I'm
> interested in optimizing the drawing speed. We are currently using
> triangle strips to represent the terrain, I would like to know is there an
> optimum number of triangles per tristrip that would the best drawing
> performance. Do non-indexed tristrips draw faster than indexed? Also is
> there a similar optimum for the number of tristrip primitives per geoset?
> I typically develop on an Octane with EMXI graphics, and run on an Onyx
> with multiple IR boards... I don't know if these numbers vary with
> chipsets... Thanks in advance for your help...
>
> Sincerely,
>
> Robert Stein
>
> Robert J. Stein
> National Center For Supercomputing Applications
> 405 N. Matthews, Urbana, IL
> (217) 244-7584
> rstein++at++ncsa.uiuc.edu
> -----------------------------------------------------------------------
> List Archives, FAQ, FTP: http://www.sgi.com/software/performer/
> Submissions: info-performer++at++sgi.com
> Admin. requests: info-performer-request++at++sgi.com
This archive was generated by hypermail 2.0b2 on Tue May 25 1999 - 04:49:34 PDT