The Quadro FX3000 is based on the same NV35 core that powers the latest GeForce FX 5900 cards, but with a few new additions to cater to the professional market. Compared to the older Quadro FX2000, which was based on the NV30, or GeForce FX5800, the FX3000 has double the memory, at 256MB, as well as a 256-bit memory interface, giving double the theoretical graphics bandwidth, hitting just over 27GB/s. It also supports other features from the NV35, including a 128-bit floating-point precision graphics pipeline, 128-bit colour, 12-bit sub-pixel accuracy, 16x full-scene anti-aliasing, 3D volumetric textures, and of course support for all the latest OpenGL, DirectX and NVIDIA's own Cg graphics language versions.

NVIDIA's whole attack for the FX3000 has been a slight side-step from its usual 'bigger, better, faster' approach to GPU marketing. In the consumer marketplace, ATI has a slight edge over NVIDIA in terms of image quality, and in the professional space, NVIDIA has taken some flak from the likes of 3dlabs for its previous cap on memory of 128MB, compared to the 3dlabs Wildcat VP990, with its 512MB. The result was NVIDIA upping the memory to 256MB on its new Quadro card.
So, marketing aside, what does this all mean for us?

Getting real

This review aims to cut through much of the hype and make sense of the theoretical specifications. In the professional, high-pressure production-intensive world of 3D design and animation, results, efficiency, reliability, stability and, of course, true 3D application speed are what make or break the job. If we have to wait for a quarter of a million polygons to spin around in 3dsmax or reboot again because the drivers are unstable, then the client has to wait for delivery.

The Quadro FX3000 is here being pitted against the previous model, the FX2000, and NVIDIA's latest workstation entry-level card, the FX500. We've also run these professional cards up against the ATI RADEON 9800 PRO to give you an idea of how the latest consumer-level cards compare in this realm.

SPECviewperf (www.spec.org) is the industry standard 3D performance benchmark. It simulates several different working situations, such as high polygon models, heavy texture usage, viewport animation and lighting set-ups via various professional 3D application tests, such as 3ds max, Pro-Engineer and Unigraphics.

It's a great test environment, but to further emphasise the real world performance of these cards, and avoid tweaked architecture, five different 3ds max 5.1 full screen animation scenes were used. Viewport playback rates were measured in frames per second and averaged out over several tests. Again, these scenes were designed either by discreet to test out the display features and performance of 3dsmax in conjunction with card manufacturers, or by ourselves in real world 3D animation jobs or specific tests we created that push the capabilities of graphics card and 3ds max performance.

Results and impressions

Looking at the benchmark graphs, it's clear that NVIDIA has done a good job in performance terms. The FX3000 outperforms all the other cards in virtually every test, although the margin is slim with the FX2000 in some cases.

However, a very important point to note, especially for 3D professionals, is driver implementation. The FX3000, and even the FX500, have normal NVIDIA drivers, but don't yet have NVIDIA-optimised 3ds max and Maxtreme drivers, unlike the FX2000. As a result the FX2000 delivers results that are 25-30 percent higher than when using the generic NVIDIA Quadro drivers v.44.71. We used the best performing drivers available for each card in these tests, because in reality, in a working environment, that's exactly what you would do.

In general usage, the FX3000 was great: smooth, reliable and definitely faster than the older FX2000 in all 3ds max areas. Massive polygon models, fully shaded modes, wireframes, complex IK rigs in character animation as well as general display quality and stability were all superb to work with. It's a shame 3ds max optimised drivers are not available yet, but when they are, the FX3000 will be even better.

In addition, the FX2000 still performs very well indeed in 3ds max 5.1. It's very fast and smooth, producing good quality displays in any resolution. The maturation of the certified drivers really helps in general usage as it significantly boosts performance but also aids stability, which could be improved with the later model cards.

The FX500 is a good, modern, entry-level workstation card, a lot cheaper than the high-end cards, and it performs well, all things considered. Compared to the performance you get when working with its bigger brothers, the FX500 seems rather slow and laboured. However, we are talking a mere fraction of the price, and for general basic 3D work it should get the job done.

Our consumer-level reference card, the RADEON 9800 PRO, overall performed terribly. Using 3ds max5.1 was hard work. The reason for this is that games are a low polygon, texture and effects-heavy, real-time 3D environment. For this you don't need a throbbing geometry engine, but you do need massive rendering power. Professional modelling, on the other hand, is far more  geometry-intensive, and doesn't require as much in terms of real-time rendering horsepower. As such, the 9800 PRO performed reasonably well in the few low-polygon, texture-rich, 3ds max benchmark tests. Unless working in 3D is a very low priority or you are forced to spend less than $1,000 we would not recommend these games cards to any serious 3D professional, especially when there are the likes of the Quadro FX series available. Drivers and real 3D application performance is why workstation cards cost so much more. No contest.