In addition to its Hopper GH100 GPU, NVIDIA is also unveiling its next-generation Grace CPU design which will be accommodated by two brand new Superchip platforms. These include the Grace CPU and the Grace Hopper platforms, designed for HPC and AI ecosystems.
NVIDIA's SUPERCHIP Platforms Powered By Grace CPU & Hopper GPU Architecture Unveiled, Up To 600 GB of Memory, 144 ARM Neoverse CPU Cores, & 2x Perf/Watt Versus Existing Servers
NVIDIA has devised a 3 Chip Data Center Strategy where it wants its Hopper GPU, Grace CPU, and a combination of both its IPs to leverage the whole AI and data center ecosystem. For the more non-traditional workloads, NVIDIA has two platforms cooked up. These are labeled as 'SUPERCHIPS' and will be available in CPU only and GPU+CPU combinations.
First up, NVIDIA has the Grace Hopper Superchip, a platform that is designed for giant-scale AI and HPC use. This Grace CPU and Hopper GPU combo are stacked with 600 GB of GPU memory, a 900 GB per second coherent NVLINK interface and runs the NVIDIA Computing Stack software suite. All of this offers a 30x system memory bandwidth to the GPU in the servers equipped with SUPERCHIP products and offers an immense increase in performance. The Grace Hopper Superchip platform is expected to be available in the first half of 2023.
Grace Hopper Superchip
The Grace Hopper Superchip combines the Grace and Hopper architectures using NVIDIA NVLink-C2C to deliver a CPU+GPU coherent memory model for accelerated AI and high-performance computing (HPC) applications.
- CPU+GPU designed for giant-scale AI and HPC
- New 900 gigabytes per second (GB/s) coherent interface, 7X faster than PCIe Gen 5
- 30X higher aggregate system memory bandwidth to GPU compared to DGX A100
- Runs all NVIDIA software stacks and platforms, including NVIDIA HPC, NVIDIA AI, and NVIDIA Omniverse
The second Superchip platform is the Grace CPU which features the high-performance Grace ARM CPU cores aimed at HPC and AI infrastructure. The computing platform is composed of 144 cores based on the ARM Neoverse architecture that delivers 740 SPECrate@2017_int_base performance (estimated). The previous estimates were 300 in the same performance metric so this is an increase of almost 2.5x from its first unveiling.
Grace CPU Superchip
NVIDIA Grace CPU Superchip uses the NVLink-C2C technology to deliver 144 Arm v9 cores and 1 TB/s of memory bandwidth.
- High-performance CPU for HPC and cloud computing
- Super chip design with up to 144 Arm v9 CPU cores
- World’s first LPDDR5x with ECC Memory, 1TB/s total bandwidth
- SPECrate2017_int_base over 740 (estimated)
- 900 GB/s coherent interface, 7X faster than PCIe Gen 5
- 2X the packaging density of DIMM-based solutions
- 2X the performance per watt of today’s leading CPU
- Runs all NVIDIA software stacks and platforms, including RTX, HPC, AI, and Omniverse
It is also the first platform to feature LPDRR5X memory with ECC enablement, offering up to 1 TB/s of memory bandwidth. The platform also runs on NVIDIA's Computing Stack software suite and delivers 2x the performance per watt over traditional servers from Intel and AMD. The platform is also expected to unveil in 1H of 2023. The Grace SUPERCHIPs are expected to be rated at 500W.
NVIDIA states that its Grace is a highly specialized processor targeting workloads such as training next-generation NLP models that have more than 1 trillion parameters. When tightly coupled with NVIDIA GPUs, a Grace CPU-based system will deliver 10x faster performance than today’s state-of-the-art NVIDIA DGX-based systems, which run on x86 CPUs.
NVIDIA built Grace by leveraging the incredible flexibility of Arm’s data center architecture. By introducing a new server-class CPU, NVIDIA is advancing the goal of technology diversity in AI and HPC communities, where choice is key to delivering the innovation needed to solve the world’s most pressing problems.
