The company is really bringing out its guns at GTC 2013

Mar 20, 2013 09:08 GMT  ·  By

NVIDIA's GPU Technology Conference is always a time of revelations, and anyone who thought it strange that the Tegra business got an update when the main GPU business didn't can breathe a sigh of relief.

It turns out that NVIDIA has, in fact, updated the product roadmap for the graphics processing unit business segment.

The move was similar to how the corporation revealed its intention to release the Logan Tegra 4 in 2014 and the Parker in 2015.

While 2013 will be the second year of the Kepler, 2014 will be the time of arrival for the Maxwell, if we are reading the chart right.

We suppose it is possible for the Santa Clara company to only release the chip in very late 2014 or early 2015, but then the successor would have to come even later.

Needless to say, it would be suspicious for the GPU line to get refreshed more slowly than the Tegra chip collection.

That said, after Maxwell, will come the Volta. This is the first time the chip's name is shown.

In addition to CUDA, FPM, Dynamic Parallellism and unified virtual memory (which were included progressively in Tesla, Fermi, Kepler and Maxwell generations, respectively), Volta will add stacked RAM to the feature pack.

Stacked DRAM would allow for bandwidths of up to 1 TB/s, which will be a level of magnitude above the 300+ GB/s of the Radeon HD 7970 and NVIDIA's own GeForce GTX Titan, but still within the realm of possibility.

“In Volta, due after Maxwell, memory modules will be piled directly atop Volta’s GPU cores,” NVIDIA explains.

In the words of NVIDIA's CEO Jen-Hsun Huang, 1 TB/s is enough to transfer a Blu-ray disk's worth of data through the chip in just a fiftieth of a second. A goal definitely worth striving towards.

Photo Gallery (2 Images)

NVIDIA GPU Roadmap
NVIDIA Volta GPU
Open gallery