Nvidia, Samsung drop ARM server chip plans

June 26, 2014 // 10:58 a.m.

Tags: #64-bit #arm #denver #development-board #hpc #jetson #nvidia #project-denver #samsung #server #supercomputer #tegra #tegra-k1 #tesla

Despite increased interest in the market, semiconductor giants Samsung and Nvidia have confirmed that neither will be pursuing the ARM server market - at least, for now.

The low power draw of processor designs from Cambridge-based ARM has led to a near-monopoly of the mobile market, but its presence elsewhere is uncommon. The recent release of 64-bit designs, which allow for increased performance and support for larger memory capacities, looks set to change that with industry giants including Amazon, Facebook and Google allegedly investigating the use of ARM-powered highly-parallel servers in their data centres. The technology is also causing a stir in the high-performance computing (HPC) market, where it is seen as the logical choice for a serial processor to feed data to high-power accelerator boards - a job currently given over to traditional x86 chips.

Earlier this week, Nvidia announced that it had partnered with industry specialists on bringing its Tesla accelerator boards to ARM-based HPC nodes. The big surprise of the announcement, however, was that said nodes would be using processors developed by Applied Micro rather than Nvidia's own Tegra family of chips. Nvidia itself made no mention of Tegra in its unveiling, and a new report suggests the reason why: Nvidia is getting cold feet about the market.

Nvidia announced plans to release a variant of its Tegra ARM-based system-on-chip (SoC) family aimed at desktops, servers and supercomputers back in 2011 as Project Denver. Since then, the closest the company has come is the release of a Tegra K1-based development board dubbed the Jetson. Now, the Wall Street Journal is claiming that Nvidia is changing the goalposts, repositioning Denver as a high-performance entry in its existing Tegra line-up for mobile devices. A company spokesperson confirmed that fact, although claimed that Nvidia is 'keeping its options open' with regards to developing a desktop- or server-centric Denver variant in the future.

The paper also claims that Samsung, which produces its own range of mobile-centric ARM chips under the Exynos brand, has abandoned its own efforts to produce a server-centric processor. Although not directly confirmed by the company, a spokesperson did admit that 'Samsung Semiconductor has refocused some of its R&D [research and development] efforts.'

The two companies may be wise to let others test the waters first: although interest in ARM-based servers and HPC nodes has never been higher, the market is still dominated by the x86 architecture. ARM server pioneer Calxeda discovered this to its cost late last year, with delays to ARM's 64-bit architecture causing the company's chief executive Barry Evans to admit that 'we just literally ran out of cash.'


View this in the forums