Nvidia has outlined a new 192 core "super chip" called the Tegra K1 which is designed to bridge the gap between mobile computing and supercomputing. But Nvidia has a lot way to go before it can convince hardware makers that it has what it takes.
Nvidia plans to offer the Tegra K1 in two flavors – a quad-core A15 part based on the ARM architecture good up to 2.3GHz, and a 64-bit dual-core part based on the Denver architecture good up to 2.5GHz. Both will feature 192 Cuda cores and are based on the Kepler architecture.
The CES announcement for the Tegra K1 by Nvidia CEO Jen-Hsun Huang made it sound good, but then everything announced at CES sounds good. There's a huge gulf between an announcement at CES and real-world success, and if Nvidia is to make the Tegra K1 work, it has a lot of work to do.
First, the parts actually need to be released, and fast. The current schedule has the 32-bit version of the Tegra K1 appearing in devices before mid-2014, and the 64-bit version by the end of the year. Given that Apple already has 64-bit processors inside hardware, and Qualcomm isn't that far behind, this is a long time, and it gives Nvidia's competitors plenty of time to gain more ground.
And this assumes that Nvidia doesn't fall behind schedule.
Another issue will be power efficiency. While the Tegra K1 might stack up well compared to the AMD silicon inside the new Xbox One and PlayStation 4 (5W versus 100W), it's how the architecture stacks up against mobile silicon in the real world that matters.
Then there's the matter of how easily hardware makers can integrate the chip into designs. Qualcomm in particular has worked hard to produce reference designs that hardware makers can drop into their hardware, and has even come up with simple solutions to integrate cellular and LTE into devices.
Then there's price. The mobile market is especially cut throat, and even a few cents can sway a hardware maker.
Nvidia has been talking up Tegra solutions for years now, but their impact on the real-worked has been limited, as hardware makers have been turning to Qualcomm for silicon. If the Tegra K1 is going to be different, Nvidia needs to do more than just come up with the silicon. It needs to put in place the groundwork to make the Tegra K1 compelling to hardware makers.