# GH200

> NVIDIA · Hopper

Canonical: https://gpuninja.com/chips/gh200
Data as of: 2026-10-11T17:23:10Z
Path: Home › GPU chips
- Specs verified: 11 Oct 2026, 17:23 UTC

## Specifications

- Vendor: NVIDIA (Verified · source download.nvidia.com · retrieved 11 Oct 2026, 16:50 UTC · checked 11 Oct 2026, 17:23 UTC)
- Architecture: Hopper (Verified · source docs.nvidia.com · retrieved 11 Oct 2026, 16:56 UTC · checked 11 Oct 2026, 17:23 UTC)
- Die: GH100 (Verified · source docs.nvidia.com · retrieved 11 Oct 2026, 16:56 UTC · checked 11 Oct 2026, 17:23 UTC)
- Process: TSMC 4N (Verified · source nvidia.com · retrieved 11 Oct 2026, 16:58 UTC · checked 11 Oct 2026, 17:23 UTC)
- Shader cores: 16,896 (Unverified · source developer.nvidia.com · retrieved 11 Oct 2026, 17:22 UTC · checked 11 Oct 2026, 17:23 UTC)
- Tensor cores: 528 (Unverified · source developer.nvidia.com · retrieved 11 Oct 2026, 17:22 UTC · checked 11 Oct 2026, 17:23 UTC)
- RT cores: None
- Memory bus: Unknown
- Max VRAM: 96 GB (Verified · source docs.nvidia.com · retrieved 11 Oct 2026, 16:56 UTC · checked 11 Oct 2026, 17:23 UTC)
- Reference boost clock: 1,980 MHz (Verified · source en.wikipedia.org · retrieved 11 Oct 2026, 17:22 UTC · checked 11 Oct 2026, 17:23 UTC)
- Reference TDP: Unknown
- Host interface: Not recorded
- Launch MSRP: No public MSRP
- Launch date: 28 May 2023 (Verified · source nvidianews.nvidia.com · retrieved 11 Oct 2026, 16:57 UTC · checked 11 Oct 2026, 17:23 UTC)
- Segment: Data center (Verified · source nvidia.com · retrieved 11 Oct 2026, 16:56 UTC · checked 11 Oct 2026, 17:23 UTC)
- PCI IDs: 10de:2342 (Verified · source download.nvidia.com · retrieved 11 Oct 2026, 16:50 UTC · checked 11 Oct 2026, 17:23 UTC)
- Status: Active (Verified · source nvidia.com · retrieved 11 Oct 2026, 16:56 UTC · checked 11 Oct 2026, 17:23 UTC)

## Performance

- Memory bandwidth: Unknown
- FP32 compute: 67 TFLOPS (Verified · source glennklockwood.com · retrieved 11 Oct 2026, 16:57 UTC · checked 11 Oct 2026, 17:23 UTC)
- FP16 compute: 990 TFLOPS (Verified · source glennklockwood.com · retrieved 11 Oct 2026, 16:57 UTC · checked 11 Oct 2026, 17:23 UTC)
- BF16 compute: 990 TFLOPS (Verified · source glennklockwood.com · retrieved 11 Oct 2026, 16:57 UTC · checked 11 Oct 2026, 17:23 UTC)
- FP8 compute: 1,979 TFLOPS (Verified · source glennklockwood.com · retrieved 11 Oct 2026, 16:57 UTC · checked 11 Oct 2026, 17:23 UTC)
- FP4 compute: Not supported
- INT8 compute: 1,979 TOPS (Verified · source glennklockwood.com · retrieved 11 Oct 2026, 16:57 UTC · checked 11 Oct 2026, 17:23 UTC)

## Boards built on this chip

Board, Chip, VRAM, TDP, Length, Slots, Price, Stock

No boards recorded for this chip.

## Benchmarks

No benchmark results recorded.

## Sources

- https://download.nvidia.com/XFree86/Linux-x86_64/580.95.05/README/supportedchips.html — Vendor, PCI IDs; last verified 11 Oct 2026, 17:23 UTC
- https://docs.nvidia.com/datacenter/tesla/mig-user-guide/latest/supported-gpus.html — Architecture, Die, Max VRAM; last verified 11 Oct 2026, 17:23 UTC
- https://www.nvidia.com/en-us/data-center/technologies/hopper-architecture/ — Process; last verified 11 Oct 2026, 17:23 UTC
- https://developer.nvidia.com/blog/nvidia-hopper-architecture-in-depth/ — Shader cores, Tensor cores; last verified 11 Oct 2026, 17:23 UTC
- https://en.wikipedia.org/wiki/Hopper_(microarchitecture) — Reference boost clock; last verified 11 Oct 2026, 17:23 UTC
- https://nvidianews.nvidia.com/news/nvidia-grace-hopper-superchips-designed-for-accelerated-generative-ai-enter-full-production — Launch date; last verified 11 Oct 2026, 17:23 UTC
- https://www.nvidia.com/en-us/data-center/grace-hopper-superchip/ — Segment, Status; last verified 11 Oct 2026, 17:23 UTC
- https://glennklockwood.com/garden/processors/gh200 — FP32 compute, FP16 compute, BF16 compute, FP8 compute, INT8 compute; last verified 11 Oct 2026, 17:23 UTC