NVIDIA Announces Hopper H100, Grace CPU Superchips, Jetson Orin Developer Kit

Jensen Huang has just wrapped up his GTC Spring 2022 keynote and thus the embargo has lifted on several exciting announcements from NVIDIA. NVIDIA has a lot of interesting hardware and software to talk about at this "#1 AI developer conference" from the Hopper H100 to next year's Grace CPU Superchips to the Jetson Orin.

Hopper H100

NVIDIA announced the Hopper H100 as their most advanced GPU ever built at 80 billion transistors. The H100 is intended for AI infrastructure and is built on TSMC's 4N process. The H100 is rated for 4000 TFLOPS FP8, 2000 TFLOPS FP16, 1000 TFLOPS TF32, and 60 TFLOPS for FP64 performance. The HBM3 memory on the H100 can deliver 3 TB/s bandwidth and supports PCI Express 5.0 connectivity and 4th Gen NVLink. The TDP for the H100 SXM module is 700 Watts.

NVIDIA is also introducing the DGX H100 that is made up of eight H100 accelerators and the DGX SuperPOD with 32 DGX H100s for 1 Exaflops of AI performance.

NVIDIA Eos meanwhile is NVIDIA's new supercomputer to serve as a blueprint for OEM and cloud partners. The NVIDIA Eos design is made up of 576 DGX H100 systems for 18 Exaflops performance at FP8, 9 EFLOPS at FP16, and 275 PFLOPS at FP64.

The 144-Core Grace CPU Superchip

Coming in the first half of 2023 is the Grace Hopper Superchip as a CPU and GPU designed for giant-scale AI and HPC workloads. Also coming is the Grace CPU Superchip in H1'2023 that offers 144 cores, uses LPDDR5X system memory, and is rated to offer 2x performance per Watt over traditional servers.

The Grace CPU Superchip is based on Arm Neoverse CPU cores and aims to provide "twice the memory bandwidth and energy efficiency" against today's leading server processors. The Grace CPU Superchip has two CPU chips connected via NVLink-C2C while the Grace Hopper Superchip is their first CPU-GPU integrated module.

NVIDIA Jetson Orin Developer Kit

Also announced today is the Jetson Orin Developer Kit for robotics to edge AI purposes. The Jetson AGX Orin aims to deliver 8x the processing power of the Jetson AGX Xavier. The Jetson AGX Orin features a NVIDIA Ampere GPU, Arm Cortex A78AE CPU cores, and the latest accelerator IP. The Jetson AGX Orin developer kit is being made available at $1999 USD while the production models will start in Q4 at $399 USD.

On the software front they have introduced NVIDIA Triton as an open-source hyperscale model inference solution. They have also updated existing NVIDIA software packages including Riva 2.0, NeMo Megatron 0.9, Merlin 1.0, and updated NVIDIA Maxine as their audio/video quality enhancement SDK.

Those are the main highlights, stay tuned for more GTC Spring news this week on Phoronix.