NVIDIA’s GPUs to work on ARM processors
NVIDIA (NVDA) is known for its GPU (graphics processing unit) technology, which was initially used to enhance PC graphics. The company expanded the use of its GPUs in data centers to accelerate the computing performance of Intel’s Xeon and IBM’s Power CPUs (central processing unit) to handle AI (artificial intelligence) workloads. Now it is targeting ARM architecture.
At the International Supercomputing Conference in Germany, NVIDIA announced that it will bring its GPU accelerators to ARM processors by the end of the year. By 2020, NVIDIA will bring its full software stack of AI and HPC (high-performance computing) to the ARM ecosystem. The stack includes NVIDIA’s CUDA-X AI, HPC libraries, GPU-accelerated AI frameworks, software development kits like PGI compilers with OpenACC support and profilers.
NVIDIA’s support in ARM, which designs the world’s most energy-efficient CPU architecture, will give way to energy-efficient, AI-enabled exascale supercomputers. NVIDIA has collaborated with ARM before to bring deep learning inferencing to mobile and Internet of Things devices.
ARM opens a new door for NVIDIA
Japan’s SoftBank Group-owned ARM designs energy-efficient CPU architecture and licenses the design to chipmakers like Qualcomm, Cavium, and Ampere. These chipmakers develop their processor on ARM’s CPU architecture. ARM chips dominate the smartphone processor market because of their energy efficiency.
NVIDIA’s vice president of accelerated computing unit Ian Buck stated that building a supercomputer with ARM-based CPUs would be a technical challenge, but NVIDIA undertook this challenge as researchers in Europe and Japan want a third option beyond Intel and IBM. Japan’s scientific research institute Riken is developing the next-generation, ARM-based exascale Fugaku supercomputer.
ARM has been looking to take some share from Intel in the data center market but gained little success. NVIDIA’s support on its architecture will help it gain some market share.
For NVIDIA, the collaboration with ARM is another step to tap the HPC market. NVIDIA is also acquiring high-speed networking chipmaker Mellanox Technologies to expand its data center offering.