NVIDIA's FLAGSHIP Tegra K1 processor could find its way into servers, the company's CEO Jen-Hsun Huang said in a conference call.
Nvidia announced its 64-bit Tegra K1 processor at CES in Las Vegas in January, claiming it can the performance of the Playstation 4 (PS4) and Xbox One games consoles.
Described by the firm as a "super chip" that can bridge the gap between mobile computing and supercomputing, the Nvidia Tegra K1, which replaced the Tegra 4, is based on the firm's Kepler architecture.
Despite Nvidia claiming that the chip would be capable of bringing next generation PC gaming to mobile devices at launch, Huang said that much of the interest in the chip sees it in microservers.
"I think we're seeing a lot of interest in putting something like Tegra in microservers, but one step at a time," Huang said in the conference call.
If this was to happen, it would put Nvidia in competition with ARM processor makers in that market. Huang said what makes the Tegra K1 appeal is that the software stack is ready and would easily slip into microservers. And because the software is written on Nvidia's CUDA parallel programming tool for the firm's graphics core, the stack has the computing power of both GPUs and CPUs in order to boost application speed.
However, it wouldn't be all plain sailing for Nvidia if it were to release its 192-core chip into the server market. The Tegra K1 is based on custom ARM CPU, but high performance applications like databases are designed to work mostly on x86 and RISC processors as opposed to ARM.
There's also plenty of competition in that market from Samsung and AMD, which are also planning to launch 64-bit ARM chips.
The Nvidia Tegra K1 is set to hit the market sometime later this year. The 32-bit version of the Tegra K1 is available now for pre-order via the company's Jetson mini supercomputer, which was unveiled at the GPU Technology Conference (GTC) in March.
Claiming to be "the world's first mobile supercomputer", the Jetson TK1 kit is built for embedded systems to aid the development of computers attempting to simulate human recognition of physical objects, such as robots and self-driving cars.
With a total performance of 326 GFLOPS, the Jetson TK1 should be more powerful than the Raspberry Pi board, which delivers just 24 GFLOPS, but retails for much more, costing $192 in the US, or $1 per core. µ