GPUDatacenterReleased Mar 22, 2022

NVIDIA H100 SXM5 80GB

80 GB HBM3 · 5120-bit · 16,896 cores

The SXM5 module version of NVIDIA's Hopper data center accelerator, built for HGX and DGX systems in which GPUs are linked over NVLink and NVSwitch. Hopper introduced the Transformer Engine, which switches between FP8 and FP16 to speed up large language models, along with DPX instructions and confidential computing. It has no display outputs, RT Cores or video encoder.

General

Manufacturer: NVIDIA
Model: H100 SXM5 80GB
Segment: Datacenter
GPU chip: GH100
Architecture: Hopper
Release date: Mar 22, 2022
Price: Search on Amazon

Memory

Memory size: 80 GB
Memory type: HBM3
Bus width: 5120-bit
Bandwidth: 3,350 GB/s
Memory clock: 2,617 MHz

Clocks & silicon

GPU clock: —
Boost clock: 1,980 MHz
Shader units: 16,896 CUDA Cores
Tensor / AI cores: 528
RT cores: —
Transistors: 80 billion
Technology (nm): 5 nm
TSMC 4N (5 nm-class)
TDP / TBP: 700 W

Interface & APIs

Bus interface: SXM5 (PCIe 5.0 x16 host)
SLI / multi-GPU: NVLink 4.0 (900 GB/s)
DirectX: N/A (compute only)
OpenGL: N/A (compute only)
Vulkan: N/A
Other APIs: CUDA (compute capability 9.0), OpenCL 3.0

AI support

Precisions: FP8FP16BF16INT8INT4TF32FP64
AI accelerator: 4th Gen Tensor Cores (528) + Transformer Engine
Peak AI TOPS: 3,958 TOPS
Notes: Fourth-generation Tensor Cores with a Transformer Engine. Peak throughput with structured sparsity: FP8 and INT8 3,958, FP16 and BFLOAT16 1,979 TFLOPS, TF32 989 TFLOPS; dense rates are half. FP64 is 34 TFLOPS and FP64 Tensor Core 67 TFLOPS. NVIDIA does not claim INT4 Tensor Core support on Hopper. Multi-Instance GPU partitions the card into up to seven instances.

All NVIDIA GPUs →

Datacenter gpus from NVIDIA announced around the same time as the NVIDIA H100 SXM5 80GB. Add any of them to your compare list to see the specs side by side.

gpu