SAN JOSE, CA. NVIDIA has announced that its deep learning compute platform, the DGX-2, provides 10x the performance compared with the previous generation, which was released six months ago. The DGX-2 is capable of delivering two petaflops of computational power for executing deep learning workloads in a single server, supported by a 2x memory boost in NVIDIA Tesla V100 GPUs and the NVIDIA NVSwitch GPU interconnect fabric, which allows DGX-2 GPUs to communicate at up to 2.4 terabytes per second.
The DGX-2 can be outfitted with up to 16, 32 GB Tesla V100 GPUs that share a common memory space for high-performance computing (HPC) applications, while NVSwitch extends NVIDIA’s NVLink offering to deliver 5x the bandwidth of leading PCIe switches. The result is a platform with the processing capabilities of 300 servers, but 60x smaller and 18x more power efficient.
The DGX-2 is also supported by updates the NVIDIA’s deep learning and HPC software stack, including new versions of CUDA, TensorRT, NCCL, cuDNN, and a new Isaac software developer kit. These updates are available free of charge for members of the NVIDIA developer community.
For more information, including technical details, visit nvda.ws/2IRilLe.eletter-03-27-2018
About the Author
Brandon is responsible for Embedded Computing Design’s IoT Design, Automotive Embedded Systems, Security by Design, and Industrial Embedded Systems brands, where he drives content strategy, positioning, and community engagement. He is also Embedded Computing Design’s IoT Insider columnist, and enjoys covering topics that range from development kits and tools to cyber security and technology business models. Brandon received a BA in English Literature from Arizona State University, where he graduated cum laude. He can be reached by email at email@example.com.Follow on Twitter More Content by Brandon Lewis