bit-tech.net

NVIDIA's next high-end GPU to support TurboCache

NVIDIA's next high-end GPU to support TurboCache

NVIDIA's G70, featuring TurboCache technology...?

Yesterday, we saw Sony lift the curtain on its next-generation Playstation 3 console in a press conference held on the eve of the Electronic Entertainment Expo (E3) festival. Along with this, we also saw Jen-Hsun Huang, CEO and President of NVIDIA, demonstrate the graphics engine behind the Playstation 3.

Under the cross licensing agreement undertaken by NVIDIA and Sony, we expect NVIDIA's next high-end chip will utilise TurboCache technology in a similar manner to the Playstation 3.

The graphics chip inside the PS3 is co-designed by NVIDIA and Sony, and is known as the RSX Reality Synthesizer. The chip is clocked at 550MHz and is said to be twice as fast as a GeForce 6800 Ultra, which means that it will be faster than NVIDIA's current bleeding edge solution: Dual GeForce 6800 Ultra's operating in SLI. It's also manufactured on a 90nm FAB process, and we suggested yesterday that it might feature 24 pixel pipelines under the hood.

We understand that the RSX has a total of 136 shader ops per clock cycle. Based on our calculations, we get the impression that the chip features 24 pixel pipelines, along with 8 vertex shaders. It's possible that each of the pixel pipelines are more efficient than the current pipeline configuration found in a GeForce 6800 Ultra pipeline thanks to an increased number of operations per clock.

The RSX also features 256MB of local frame buffer in the shape of GDDR3, clocked at 700MHz, while having access to the Playstation 3's other 256MB of Rambus XDR memory, clocked at 3.2GHz, via the Cell processor at 20GB/s and 15GB/s when writing/reading to or from the Cell processor's buffer.

If we go back to the announcement that NVIDIA published back in December, revealing that NVIDIA and Sony were working together on the graphics chip for the Sony's Next-Generation Computer Entertainment System, as it was referred to back then, we take the following quote from the press release:

"Both companies are jointly developing a custom graphics processing unit (GPU) incorporating NVIDIA’s next-generation GeForce and SCEI’s system solutions for next-generation computer entertainment systems featuring the Cell processor."

This is closely followed by another interesting statement, "This collaboration is made under a broad, multi-year, royalty-bearing agreement."

Based on those quotes, we believe that NVIDIA's next-generation desktop graphics processing unit will be of similar specification to the RSX Reality Synthesizer GPU. However, we understand that the next-generation high-end GPU, which we have referred to as G70 in the past, is manufactured on TSMC's 110nm FAB process, rather than the 90nm Sony/OTSS FAB process that has been used for the RSX GPU.

We believe that 'G70' will have support for TurboCache memory sharing technologies, while video cards based on this GPU will also have a large local frame buffer of either 256MB or 512MB (or both iterations), allowing for larger frames to overflow in to system memory should it be required.

In saying this, we believe that there will be a second GPU released later to bridge back to AGP, as TurboCache cannot be supported on the AGP interface. It's also possible that, due to the power implementations of NVIDIA's next GPU, we may not see a bleeding edge AGP card because the power requirements are too high.

Discuss this news in our forums.