Three years after officially launching its server Project Denver, Nvidia (NASDAQ:NVDA) is downplaying its plan to develop its own 64-bit ARM-based CPU processor for servers. The company initially planned to pair its Tegra chips with GPUs in servers, but is now focusing its latest 64-bit Tegra chips (Tegra K1) on smartphones, tablets, cars and other embedded devices. Nvidia said that though the mobile K1 chip could make it to microservers, it has scrapped any near-term plan to develop a specialized server CPU. Its exit leaves only four ARM server chip makers in the market – AMD (NYSE:AMD), AppliedMicro, Broadcom (NASDAQ:BRCM) and Cavium.
Rather than building its own server processors, Nvidia, which is known for its Graphics Processing Units (GPU), will supplement ARM processor-based servers with high-performance GPUs. Earlier this week, Nvidia announced its latest Tesla GPUs (Tesla K20) which features its Cuda 6.5 parallel programming platform. Several server vendors are leveraging the performance of Nvidia’s GPU accelerators to launch the world’s first 64-bit ARM development system for High Performance Computing (HPC), a market that includes Supercomputers. These systems will be available to individuals who design these often custom built computers.
ARM 64-bit server processors were primarily designed for micro-servers and web servers because of their extreme energy efficiency. However, a standalone ARM chip does not have the relevant requirements to fuel HPC demands. Adding a GPU equips the ARM systems with the computer power required to better handle HPC workloads, enabling them to build high-performance systems. Nvidia claims to have the industry’s most comprehensive accelerated computing platform — including servers, software, development tools, processors and related technologies — optimized for the HPC industry. Of course, Nvidia’s GPUs are already used as accelerators in Intel-based HPC systems.
The server market is currently dominated by the x86 processors, which account for well over 80% of total server shipments. When combined with a GPU, the x86 processors can do more work per watt than a standalone x86 processor. ARM processors can help cut down on power consumption as they are known for their low power processors, an important criteria given teh size and energy demands of most HPC systems. AMD believes that ARM CPUs have the potential to account for 20% of the server market by 2016 or 2017. 
The first GPU-accelerated ARM64 HPC development platforms will be available in July from Cirrascale Corp. and E4 Computer Engineering, with production systems expected to ship later this year.
Our price estimate of $18.43 for Nvidia is almost in line with the current price estimate.
- AMD reboots server technology strategy with first ARM chips, Tech World, June 18, 2012 [↩]