3. NVIDIA Geforce Go 6800 Details

NVIDIA's new mobile chip is nothing more than a brother of the NV40, running with "only" 12 pixel shaders and five vertex shaders. (Actually, since the GeForce Go 6800 is a PCI Express device, we cannot help but thinking that it may in fact be closer to the NV45 than the NV40.) The GPU also offers a bunch of energy-saving features, collectively termed "Powermizer".

NVIDIA's engineers incorporated lots of technology and a massive 190 million transistors into the new chip. Due to the high transistor count, thermal loss is of course much higher than with the chip's predecessor. Despite that, the thermal envelope, at 35 watts, is still acceptable when this chip's performance is taken into account. Marketing people like to describe this architecture as revolutionary in terms of performance per watt.

The GeForce Go 6800's Block Diagram.

Please check out our full review of the desktop graphics card GeForce 6800 Ultra if you are interested in all the technical details of NVIDIA's new GPU. Here you will find a discussion of features such as Ultrashadow II, Purevideo and Displacement Mapping, a look at the integrated programmable video processor, and an explanation of how the pixel and vertex shader actually work. NVIDIA offers several whitepapers on their technical internals, too, so we will concentrate on talking about the specific features of the GeForce Go 6800.

Two versions of the GPU will be made available. The first one is the single chip without memory, while the second is based on the MXM graphics module that was initiated by NVIDIA. MXM is just a standardized PCI Express module that can be easily exchanged; it was created for notebook makers who need to have a way of using one notebook design line that differs mainly in the graphics board used. In theory, MXM could also be an interesting option for the end user, enabling the upgrade of the mobile graphics unit. However, it might take many more months until this scenario becomes reality, if it ever becomes the case at all.

Geforce Go6800 MXM

The memory interface of NVIDIA's GeForce Go 6800 is 256 bits wide. Thus, the new chip offers twice the memory bandwidth of its predecessor, the GeForce FX Go 5700. According to NVIDIA, GeForce Go 6800 supports DDRI, DDR2 and GDDR3 graphics memory, even though DDR2 has not yet been used for mobile applications. We expect DDRI and GDDR3 to dominate Go 6800 mobile solutions.

GDDR3 memory comes with on-die termination, resulting in much higher power consumption at "low" clock speeds (300 MHz) when compared with DDRI memory. As soon as higher clock speeds are required, even DDRI needs to be terminated properly (onboard, not on-die) in order to avoid signal reflection. In this case, using DDR3 memory may be the smarter choice from an energy point of view. Another efficiency advantage here is the much lower voltage of DDR3 parts. Finally, deploying DDR3 memory will double memory bandwidth.

Since NVIDIA is a fabless company, IBM takes care of producing the 130 nm chips. Both the GPU and its graphics memory are supposed to run at 300 MHz (DDR1 only), while there is the usual clocking margin for notebook builders to use. The GeForce Go 6800 in Sager's test notebook was clocked slightly slower, at 275 MHz chip and 300 MHz memory clock.

NVIDIA is already talking about a high-end version of their new mobile flagship for 2005, likely to run at 450 and 600 MHz. We expect desktop replacement notebooks equipped with this "hardcore" version to be paired with DDR3 memory, of course.