AvailableHopper • 2022
GH100
The GH100 is NVIDIA's Hopper architecture GPU silicon — the foundational chip that powers all H100 variants (SXM5, PCIe, NVL). Built on TSMC's custom 4N process with 80 billion transistors across an 814 mm² die, it introduced the Transformer Engine for accelerated LLM training and inference. The GH100 die is the largest and most complex GPU NVIDIA had shipped at its 2022 launch.
Performance vs Previous Gen
Up to 6× faster AI training than A100 (Volta→Hopper generational leap)
Architecture Highlights
Transformer Engine4th-gen Tensor CoresFP8 supportHBM3DPXNVLink 4
Technical Specifications
Architecture | Hopper |
Release Year | 2022 |
Process Node | TSMC 4N (4nm) |
Form Factor | SXM5 / PCIe Gen5 |
VRAM | 80 GB HBM3 |
Memory Bandwidth | 3.35 TB/s |
Interconnect | NVLink 4 (900 GB/s) |
FP16 Performance | 1,979 TFLOPS (with sparsity) |
FP32 Performance | 67 TFLOPS |
FP64 Performance | 34 TFLOPS |
TDP | 700W (SXM) / 350W (PCIe) |
CUDA Cores | 16,896 |
Tensor Cores | 528 |
Transistors | 80 billion |
Die Size | 814 mm² |
Best Use Cases
- Foundation for H100 SXM/PCIe/NVL — AI training
- LLM training (70B+ params)
- HPC
- inference
Quick Facts
FamilyHopper
ArchitectureHopper
Released2022
ProcessTSMC 4N (4nm)
Form FactorSXM5 / PCIe Gen5
Related Hopper GPUs
GH100 — Frequently Asked Questions
What is the GH100 GPU used for?
The GH100 is best suited for: Foundation for H100 SXM/PCIe/NVL — AI training, LLM training (70B+ params), HPC, inference.
How much VRAM does the GH100 have?
The GH100 has 80 GB HBM3 of VRAM.
What is the memory bandwidth of the GH100?
The GH100 has a memory bandwidth of 3.35 TB/s.
What interconnect does the GH100 use?
The GH100 uses NVLink 4 (900 GB/s).
When was the GH100 released?
The GH100 was released in 2022.