Groovy graphics guru Ivon Smith has taken the new Quadro for a spin. Try to excuse the less than complete windshield and missing headlight.
Everyone gets very excited when a new graphics chipset is comes out. We drool at spec sheets and performance promises, along with all the new features that will make our lives, as 3D artists and gamers, much more fun, productive and creative.
Arguably one of the most anticipated GPU releases for ages has been NVIDIA’s FX series (NV30). The gaming version, GeForce FX, was released first among much hoo-hah about real-time texture rendering, cloth and skin deformation and claimed massive amounts of polygon-pushing and unprecedented frame rates. Nice try! As most of us have seen in tests rife throughout the media, the GeForce FX didn’t kick butt as much as most of us were hoping, or lead to believe it would. And one of NV30’s claims to fame, DirectX 9 rendering effects, also will not be available in games for a long time yet.
Slightly later in the release timeline, with perhaps as much industry-targeted hoo-hah and a much inflated price entry point is the professional 3D DCC version of the NV30, the Quadro FX. Until the release of ATI’s FireGL X1, NVIDIA’s Quadro series (headed by the 980 XGL) were pretty much the most powerful OpenGL cards on the market for CAD, 3D modelling and animation, 3D visualisation, film and TV work, trouncing all the competition in benchmarks as well. Very good stuff!
Enter Quadro FX. 128MB 400MHz DDR-II memory; 400MHz chip speed; AGP 8x; 128-bit floating-point precision pipeline; 128-bit colour; 125 million transistors; 12-bit subpixel precision; 16x FSAA; CineFX shading architecture; up to 3,840x2,400 display resolution; dual-DVI digital outputs and 3D volumetric texture support for medical visualisation and solid-modelling applications. NVIDIA’s own programming language for OpenGL and DirectX display effects, Cg, is supported and enhanced in the Quadro FX family, promising realtime photorealistic effects within 3D design environments. Working with the top 3D application manufacturers, NVIDIA has developed the Quadro FX to offer advanced realtime shader effects, to help speed production pipelines so game developers can see their in-game textures and effects in the creation application, or 3D production designers to visualise their engineering products more easily.
CAD applications, such as AutoCAD, ProEngineer, SolidWorks and Unigraphics should benefit from enhanced geometry handling capabilities, and DCC applications (used extensively in the games, TV and film industries) such as 3dsmax, Maya, Softimage|XSI and Houdini will all be able to display better quality, more effects-laden and complex models, improving the design process in demanding areas such as 3D character animation.
So, how did the FX2000 actually stack-up against the competition in benchmarking? And even more importantly, how did the card (and drivers) perform in real-world applications? The test 3D workstation was a dual AMD MP2000+, with 1.5GB DDR266, ASUS A7M266-D motherboard, dual-18in Samsung LCDs, and Windows XP SP 1.
The FX2000 ranked alongside its archrival, the FIREGL X1, and my usual 3D workstation card, the slightly more modest 3DLabs WildcatVP870. Industry standard benchmark tests were performed, but the cards were also compared for power and stability in a general working 3D environment and my own custom 3D test scenes within 3dsmax 5.1.
The VP870 is usually stable, reliable and fairly smooth in its 3D usage, Photoshop and even the general Windows environment. We were pleasantly surprised and quite impressed by the workflow capabilities, features, speed and stability of ATI’s offering as well, producing naught but the odd artifact and wireframe glitch in many days of 3dsmax 5.1 production work.
When it came to the FX2000 and the Detonator drivers, it was, at first, ‘Houston, we have a problem.’ The card and the standard NVIDIA drivers seemed to have an issue with my workstation and its Windows XP environment. Not only did the card operate very sluggishly, creating glitches and instabilities in 3dsmax and Windows, but eventually it refused to work at all, totally corrupting the system, necessitating an entire Windows reinstall! Ouch.
Following technical advice from NVIDIA and Australian manufacturers of the FX2000, Leadtek, Windows was reinstalled, and the latest certified drivers for 3dsmax were used. Notoriously, the driver certification process is one of the main things we, as 3D designers, pay for when buying OpenGL cards, the FX2000 not giving much change out of $4,000. Why this is the case was apparent once I used these certified 3dsmax drivers.
Stability was restored and we were in the land of super smooth! In 3dsmax 5.1 the speed and power of the card became evident, ripping through hours of production work with ease. For example, my custom benchmark scene, 3dhouse.max, is an architectural model of 200,000 polys. The FX2000 ate this alive, smoothly navigating the scene, fully textured in realtime! Very nice. I cannot overemphasise the importance of actual performance in 3D applications over benchmarks which can be focused on specific strong or weak points inherent in a particular OpenGL card architecture.
Having said that, the relative performance in SPECviewperf V7.0 was pretty darned impressive too, with the FX2000 besting the FIREGL X1 in all tests by between 20% and 200%+. Note that the specific values are limited by overall system performance, but the score shows how the FX2000 performed in relation to the other cards.
The 3dsmax 5.1 test scenes showed a mixed bag of results, the FX2000 easily showing its superiority in shaded polygon geometry handling features, whereas the FIREGL X1 demonstrated that it is still easily worth its more modest asking price in handling lights and complex texture maps.
The FIREGL X1 was a great card to work with. The Quadro FX2000, once stabilised, proved to be a stunning card to work with! It’ll be on my Christmas list, along with that Ferrari again!
Issue: 133 | February, 2012