updated 03:44 pm EDT, Wed June 25, 2014
Denver to start as mobile Tegra chip, Tesla GPUs used to supplement ARM-based servers
In 2011, Nvidia announced that it would be developing a 64-bit processor with the "Denver" project code name for servers. This week, it appears that the company's view of servers in the future changed, causing the company to look toward ARM chips. Rather than Denver being built for server use, the company says that the chips will make an appearance in the mobile line of Tegra processors.
The Wall Street Journal says that the change was made sometime in the past, but hasn't been abandoned, according to an Nvidia spokesperson. The project is still expected to yield a high-performance chip, but instead it will begin as a part of the Tegra line for mobile. Vice President of Accelerated Computing Ian Buck told >A href="http://www.pcworld.com/article/2366480/nvidia-abandons-64bit-denver-chip-for-servers.html">PC World that Denver server processors are "not something we're doing today."
"Our Project Denver architecture is initially focused on the mobile market, and our Tegra processors are already enabling unique computing solutions for auto infotainment systems, smart phones, tablets and embedded applications," said Buck.
Nvidia doesn't indicate that it is abandoning the server world entirely, as the spokesperson said that the company would be keeping their options for the future. In the meantime, Nvidia is expecting "great things" from the project. The company says that it is possible that the Tegra K1 chips, which hit the market later this year, could make it into microservers.
Rather than building processors, Nvidia plans to supplement ARM processor-based servers with GPUs. Looking at the future of low-power servers outside of low impact services, Nvidia has been approached by ARM server manufactures to help with high performance computing (HPC).
On its own, an ARM chip doesn't possess the requirements to fuel HPC demands. By adding GPUs like Nvidia's Tesla K20 accelerators, that feature Nvidia's Cuda 6.5 parallel programming platform, the ARM systems could gain the computer power required to handle larger workloads.
"Nvidia has built the industry's most comprehensive accelerated computing platform -- including servers, software, development tools, processors and related technologies -- optimized for the HPC industry," said Buck. "GPUs are the enabling technology that allow server vendors to build HPC-class systems around flexible ARM 64 processors. The result is new, highly-innovative computing solutions for HPC."
So far, three companies have indicated they are building systems utilizing ARM and Nvidia's GPUs. Cirrascale and E4 Computer Engineering will have development servers ready in July that feature two of the Tesla K20 GPUs. Production versions, along with a box from Eurotech, are expected later in the year.