The FireGL 3 graphics card is a high end graphics accelerator intended for the CAD and animation markets. The primary feature
of the FireGL range of cards is good quality, hardware accelerated OpenGL graphics support.
A feature which was not present in the FireGL 2 display adapter was full screen anti-aliasing. On the FireGL 3, when this feature
is enabled, it is implemented in 4 steps, moving from a minimal level of anti-aliasing to a more extensive level.
Anti-aliasing is the technique where diagonal lines are filtered in such a way as to make them appear smoother than they would normally be.
On testing this feature there was a noticeable impact on the real time replay performance of the card. As a guide, with full (level
4) anti-aliasing selected, a particular reference real time application ran at 9 frames per second. With slightly less (level
3) anti-aliasing set, a replay performance of 14 frames per second was achieved. At level 2, 12 frames per second was
achieved, and at level 1 (the minimum setting), I recorded 17 frames per second. With fullscreen anti-aliasing switched off, the
test application ran at 25 frames per second, though the code itself was written to not exceed this value, regardless of the
hardware performance available. The reason for the inversion of the values for level 2 and 3 was not known, though the values
themselves were double-checked just to make sure no timing errors had crept in.
All this testing was performed using a full screen display, set to a resolution of 1280 x 1024 pixels, in 32-bit colour. With
such a high resolution, I found turning off the anti-aliasing did not result in an unacceptable level of image degradation, so the
anti-aliasing feature is probably better suited to either real-time applications using much lower screen resolutions, or non-real-time applications, such as CAD.
Benchmarks
When I get the opportunity to test an OpenGL graphics accelerator I like to run a couple of my own benchmarking programs. While the
more standard graphics benchmark program which exist probably give a more direct comparison between the performance of
different graphics chipsets, my test programs are real applications and I find they give me a better feel for what a
particular graphics card can offer in a real application.
For a card like the FireGL 3, the key features it offers are OpenGL support and real-time display capability. Some time ago I
obtained a demonstration program called "Vampire" which was a real-time OpenGL simulation. It also had the added feature of
displaying the instantaneous frame rate while it was running. For the FireGL 3, this ran at about 70 frames per second, ranging
between 55 and 90, depending on the image complexity at any particular time. This was for a full-screen replay, at 1280 x
1024 pixels resolution. From a prior review of the FireGL 2 card, this was about the same performance achieved by that card.
A second test, based on a real-time 3D replay program for displaying vehicle motion paths, showed that when set to run at
25 frames per second the card did exactly that, maintaining a very stable 25 frame per second replay. Although a qualitative
assessment, the cards image quality appeared noticeably better than that displayed on lower cost Open GL cards. Raising the
replay rate to 50 frames per second also displayed stable real-time performance, though rates higher than this started to show
some degradation in the program's ability to maintain the replay at a real time rate.
Although the FireGL 3 supports 3DS MAX and similar rendering programs, it is important to realise that simply using such a
display card will not improve the general rendering performance of these programs. It is necessary to appreciate just what sort
of applications a card like the FireGL 3 will benefit. In the case of a program like 3DS MAX, the actual gain will be made in
the real-time performance of the on-screen viewports and not in final rendering operation. That is, the quality of the on-screen
display during editing will be improved, as will the display response, but final rendering, which is not a display dependent
operation, will be unaffected.
Similarly, many 3D CAD programs offering support for OpenGL hardware acceleration will see the benefits in the ability to
manipulate complex 3D models on screen in real-time. The key gains will be a general improvement in image quality and response
on-screen and the ability to employ dual monitors, which can be a real advantage in applications employing busy menus and multiple
pop-up dialogue boxes.
Conclusion
The FireGL 3 card and the other cards in the family are primarily aimed at providing extremely high performance in real-time
display applications and the viewing of 3D geometry. These functions they perform well, though at a price, as the cards are
quite expensive. Nevertheless, these cards offer very good real-time OpenGL performance and image quality for appropriate
applications.
