From The Register comes a very interesting interview with Steve Scot, CTO of Tesla products, a division of Nvidia, which talks about his company’s plans for the high-performance computing (HPC) and supercomputers segment.
Nvidia believes the HPC (High Performance Computing) market is still too small, a situation that will change in the coming years, and the company is preparing for that future with its future Tesla product line, initially born as a variant of its successful line GeForce product for the consumer sector. But this situation will change in the future where Tesla will increasingly have more in common with Tegra (family of ARM SoCs) with GeForce, Nvidia plans to turn it into a product even more capable than it is today, as integrated ARM Cortex-A57 cores customized by Nvidia itself to offer superior performance to that of any comparable SoC.
Both Tesla and Tegra aim to provide high performance and the best possible consumption per watt factor; that will determine the success of both products both on the market of portable devices (Tegra) and the server segment and supercomputers (Tesla). For this last segment, Nvidia is working on an interconnect technology (LAN router type) across multiple GPUs, which will be integrated into future Tesla products, while working on similar technologies integrated in Tegra to Stacked DRAM with their future GPUs Volta.
Nvidia is confident ARM will prevail over x86, so that in the future ARM will be in more types of devices, including those segments such, as PC gaming, currently dominated by x86, supplanting x86 microprocessors, although currently Nvidia agrees to continue cooperating with x86 microprocessors from Intel and AMD.
X86 may disappear over time, not because it is lower than ARM, but it imposes many brakes on innovation, whereas ARM incorporates an open architecture that is increasingly used for its highly customizable, thanks to which it represents a great opportunity for many companies.