Whether you prefer to stack greens or reds in your sandwiches, both the cabbage and tomato farmers have been busy working to make superior chomp-chump for enthusiast. After the considerably successful launch of the ATi HD3870 X2, our friends at NVIDIA have been busy trying to come up with a rendering monster that would have made even Hillel himself proud.

The NVIDIA 9800 GX2 is what I'd term a SLI Package involving some very nifty engineering workaround to tackle cumbersome politics within the computer hardware industry. Based on the G92 GPU, the GX2 design puts multi-GPU in any system with less pushy support, the same way the HD3870 X2 and NVIDIA GeForce 7950 GX2 did.

The G92 still rules the show.

"Politics" here refers to the perrenial rivalry between the camps of red, green and blue, which have resulted in certain graphics drivers refusing to provide multi-GPU support on specific motherboard chipset platforms. What these GPU "packages" do, is market multi-GPU rendering as a single Add-In-Card which pleases just about everybody. That in return, means working on just about every platform.

The sheer adrenaline rush.

When the guys at Innovision swing by with two huge black boxes, it got the journalists real curious. On to the pictures!

The Inno3D NVIDIA 9800GX2 depends on the reference PCB and thermal design from NVIDIA. Clad in a gloss black metal jacket, this graphics accelerator is actually electrically shielded. HDMI and Dual DVI ports are the escape routes for frisky pixels, while a SLI port is present for you to run quadruple GPUs. Naturally, such a design takes up two expansion slots.

Consisting two physical circuit boards, the NVIDIA 9800 GX2 falls into the heavyweight category, measuring a lengthy 10.5 inches. Found along the GPU rim are 1ns Samsung memory that runs on a 256 bit memory controller, totalling 1024MB.

As lengthy as the modules itself is the history of the chipset linking up the two G92s, known as the BR-04. PCIe and SLI lanes pushes data about via the twin flexible printed circuit jumpers.

Squishing a pair of GPU chipsets together to form a single dual-slot graphics accelerator is no mean feat, especially if you think of it as an act of melting a pair of 8800 GTS 512MB into one. Melting, is exactly what will happen if the design thermals are not taken care of.

Then again, heat isn't going to be a tripping stone for NVIDIA this time around.

Back when the NVIDIA GeForce 7950 GX2 was released, the front-to-back design proved to be an engineering challege for thermal management. Enough surface area had to to be garnered from a heatsink half the width of an expansion slot, while noise had to be kept in check when attempting to reduce turbulence from drawing air through a small gap.

A literal nightmare for overlclockers without the creative touch indeed. The old GX2 made its mark only in the hands of Italian overclocking artisans and those who engaged custom 'water' blocks (Shamino).

Today, we see a totally new approach to the design of the "stack." The 9800 GX2 now has its PCB halves back-to-back, clamping two GPU faces and associated components onto a common heatsink. Heatpipes are a notable feature, showing up through the black paint that has been applied to the assembly.

This heatsink engages an axial blower fan from Delta to pass air through the diagonally-angled fins which throws hot air through the top and side of the assembly.

PWM control is utilised on the 9800 GX2, with the blower being a 4wire. Regular NVIDIA fare conducts the heat from component to assembly - grey goop and soaked canvas as thermal interface material.

It was apparent that the new approach brought about routing difficulties for PCB designers, but in exchange was a subsequently beefier heatsink design that can run considerably cooler than rival attempts. We shall investigate the temperatures for the 9800 GX2 to see how it fares.

Power consumption of the two modules combined takes a fair few amperes, which has to be supplied via the dual 12V PCIe power sockets. This apporach is not unlike the ATi HD3870 X2.

3 phase power is available for both G92 chipsets, based around a Volterra power system. At its heart is the Volterra VT1165 power management controller, which has already been seen on a few generation of ATi graphics accelerators. Intersil's ISL6269 takes care of memory power this time around. Power delivery is replicated on both modules.

For those who can't get wait to get their hands dirty at voltage modifications, we've already uploaded a preliminary guide here. Try it at your own risk, however, for we have yet to finalize testing.