Single page Print

NVIDIA
Like last year, NVIDIA announced a new mobile graphics chip at Comdex 2001. This product doesn't yet have a name beyond NV17-M, and laptops based on it won't hit store shelves until next spring. Nonetheless, NV17-M does promise to make life a lot more fun for road warriors.

The NV17-M is an odd combination of GeForce2 and GeForce3 technologies with some new non-3D features thrown in. We talked briefly with Bill Henry, NVIDIA's Director of Mobile Product Management, about what exactly went into the NV17-M. We gathered several important facts from the conversation:

  • First, the NV17-M is very much a new chip. It has a revamped video processing engine for low CPU overhead during DVD playback, and it supports a variety of outputs natively, including LCDs and televisions.


    The NV17-M incorporates a host of mobile-ready features into one chip

  • The NV17-M's memory interface is similar to that of the GeForce3: it's a crossbar-style controller, which NVIDIA has dubbed its "Lightspeed memory architecture." The chip also includes some of the GeForce3's bandwidth-saving techniques, like Z occlusion culling.


    NVIDIA's new NV17-M mobile chip is packaged together with two RAM chips


    A major improvement in memory bandwidth

  • NV17-M can deliver GeForce3-style multi-sampled anti-aliasing. With most laptop displays, the chip should be able to provide high performance while delivering a much-needed reduction of jaggies on extra-crisp LCDs.

  • The chip's main 3D rendering core is very similar to a GeForce2 MX. It has two pixel pipelines capable of laying down two textures per pixel. There is no vertex shader, so the chip will have to offload DirectX 8-style vertex calculations to the CPU, as does the GeForce2. NV17-M also lacks a full DirectX 8 pixel shader implementation; it has only register combiners very similar to the GeForce2. Because pixel shaders can't be appropriately emulated in hardware, NV17-M will not be a true DX8-capable part.
And that's about it. I should emphasize something here. NVIDIA showed a familiar GeForce3 demo at the NV17-M launch: the Chameleon. Folks oohed and aahed over the Chameleon demo, but it was clear to us that this wasn't the real GeForce3 version of the demo. Most obviously, not all the gloss and bump maps were being applied to the chameleon model. Also, the model looked rough (lower poly) at important vertex junctions, like where the legs and tail attached to the torso. Finally, when we saw the NV17-M Chameleon demo at NVIDIA's Comdex display, the demo exhibited serious stuttering problems whenever the chameleon changed skins, as if the chip couldn't keep up with the texture transfers. We didn't see this stuttering problem in the press conference, so it may have just been a bug, but it's hard to say.

Although they were reasonably forthcoming when we pressed them—incisively, as we tend to do—on this point, NVIDIA is playing a dangerous game, giving folks the impression the NV17-M has a GeForce3-class renderer at its core. NV17-M doesn't necessarily need to have full vertex and pixel shaders in order to succeed in the mobile market; NVIDIA knows that, and they ought to be willing to say so.

Beyond the new mobile chip, NVIDIA was showing off Microsoft's Xbox, which uses NVIDIA graphics and I/O chips. They even had one ripped open in order to show exactly where the NVIDIA chips are.


The Xbox cracked open


The Xbox motherboard with XGPU and MCP chips