NVIDIA · Hopper

NVIDIA Hopper

Proven previous-generation accelerators, widely available new and refurbished.

In plain English

Hopper GPUs (H100 and H200) remain excellent for training and inference and are usually the best value per unit of AI performance on the secondary market.

Who it is for

  • · Cost-conscious AI teams
  • · Universities
  • · Buyers scaling existing Hopper estates

Typical workloads

  • · LLM training
  • · Inference at scale
  • · HPC and simulation

Generations

Previous:

Next: nvidia-blackwell

Products in this family

Data Centre AI Accelerators
Representative

NVIDIA H200 NVL 141 GB

A very capable data centre GPU in card form, so it can be fitted into common rack servers rather than requiring a specialist chassis.

Memory

141 GB

Deployment

Rack server

DatacentreHigh-Memory

£24,950

ex VAT

Low stock

Refurbished AI Infrastructure
Representative

NVIDIA H100 80 GB PCIe — Refurbished

A used but professionally tested H100. The same silicon as new stock, at a lower price, with a warranty and a documented test report.

Memory

80 GB

Deployment

Rack server

RefurbishedHigh-MemoryProduction Inference

£15,750

ex VAT

In stock

Refurbished AI Infrastructure
Representative

NVIDIA A100 80 GB PCIe — Refurbished

An older data centre GPU that still performs well for many workloads and is one of the cheapest routes into 80 GB of GPU memory.

Memory

80 GB

Deployment

Rack server

RefurbishedHigh-MemoryProduction Inference

£6,450

ex VAT

In stock