Am Freitag, 27. Oktober 2000 01:07 schrieb Gareth Hughes:
> Dieter Nützel wrote:
> > The new ISS and 3Dnow! code did not run on my Athlon (3DNow!).
> > If I use the DRI libGL.so (hardware accelerated) it works fine.
> Dieter, can you try again? Perhaps include a backtrace as well? It
> runs fine on my Duron system.
Did it again, but with 'cvs -qz3 update -dP -D 2000-10-22'
(rollback to the 22th October)
Linux kernel is 2.4.0-test10-pre5
Will try second time with the new 3DNOw!/SSE stuff.
> Part of what I'm working on at the moment is a comprehensive
> test/benchmark suite for the fastpath vertex routines in Mesa, which
> will involve expanding the available accelerated functionality in core
> Mesa. I'm comparing all the results of all of the functions (xform,
> normal xform and now fastpath xform/clip/project) with reference C code
> and haven't seen any errors.
> -- Gareth
> PS - My 600MHz Duron (KT133 w/ PC133 SDRAM) is flogging my 700MHz PIII
> (I assume 100MHz FSB w/ PC100 SDRAM) laptop. It's regularly up to twice
> as fast executing the same assembly code, and 3DNow! pretty well rocks.
> I'm doing a thorough review of the SSE code to try and improve things a
> bit on that platform.
This sounds great.
Send as much input for testing as you can along to me 'cause I have an AMD
Athlon T-Bird 900MHz slot A (yes, really slot A:-), here for testing.
I will try the DRI trunk and the latest tdfx-2-1-branch (tdfx-3-0-branch is
very slow for textures) with my V5 5500 AGP on Q3 and UT 428a.
Should I tell you that I saw astonishing speed with the VTK
'sphere-bench.tcl' benchmark (tdfx-3-0-branch)?
The single VSA-100 paired with the Athlon T-Bird 900 outperform a GeForce2 on
a dual PIII 733 (Windows200 and Linux)!!!
Even 3DLabs Oxygen GMX2000 look not that great...
Best results, summary :
- i686 running Linux 2.4.0-test10-pre5
- VTK 3.1.2 (rev: 1.668, 2000/10/26 00:02:16)
- Visual is 1280x1024, truecolor/truecolor/24
- Tcl/Tk 8.0.5
- window was 400 x 400
1) 512x512 : 1735.2 kpolys/s : [stripper]
2) 512x512 : 787.2 kpolys/s : [transparency]
3) 512x512 : 787.0 kpolys/s : 
4) 512x512 : 786.0 kpolys/s : [small_sphere]
5) 512x512 : 624.1 kpolys/s : [texture]
6) 512x512 : 621.2 kpolys/s : [texture, transparency]
7) 512x512 : 503.8 kpolys/s : [wireframe]
------- DONE (in 303 sec.) -------
BTW When will we see SLI? Couldn't resist :-)
Graduate Student, Computer Science
University of Hamburg
Department of Computer Science
Cognitive Systems Group
D-22527 Hamburg, Germany