GPUDatacenterReleased Mar 22, 2022

NVIDIA H100 CNX

80 GB HBM2e · 5120-bit · 14,592 cores

The H100 CNX is a converged accelerator that puts a Hopper H100 GPU and a ConnectX-7 SmartNIC on a single dual-slot PCIe board, joined by an integrated PCIe Gen5 switch so network traffic reaches GPU memory via GPUDirect RDMA without crossing the host. Announced at GTC in March 2022, it targets multi-node AI training, 5G vRAN and AI-on-5G, and offers one 400 Gb/s or two 200 Gb/s ports for Ethernet or InfiniBand. It supports Multi-Instance GPU partitioning into up to seven instances.

General

Manufacturer: NVIDIA
Model: H100 CNX
Segment: Datacenter
GPU chip: GH100
Architecture: Hopper
Release date: Mar 22, 2022
Price: Search on Amazon

Memory

Memory size: 80 GB
Memory type: HBM2e
Bus width: 5120-bit
Bandwidth: 2,000 GB/s
Memory clock: —

Clocks & silicon

GPU clock: —
Boost clock: —
Shader units: 14,592 CUDA Cores
Tensor / AI cores: 456
RT cores: —
Transistors: 80 billion
Technology (nm): 4 nm
TSMC 4N
TDP / TBP: 350 W

Interface & APIs

Bus interface: PCIe 5.0 x16
SLI / multi-GPU: NVLink Bridge (two-way)
DirectX: N/A (compute only)
OpenGL: N/A (compute only)
Vulkan: N/A
Other APIs: CUDA 9.0, OpenCL 3.0

AI support

Precisions: FP8FP16BF16INT8INT4TF32FP64
AI accelerator: 4th Gen Tensor Cores (456) + Transformer Engine
Peak AI TOPS: —
Notes: Fourth-generation Tensor Cores with the Transformer Engine, supporting FP8, FP16, BF16, TF32, FP64 and INT8. INT4 is not supported on Hopper. NVIDIA does not publish separate peak throughput figures for the H100 CNX.

All NVIDIA GPUs →

Datacenter gpus from NVIDIA announced around the same time as the NVIDIA H100 CNX. Add any of them to your compare list to see the specs side by side.

gpu