Confirm condition, inspection evidence and commercial terms in the quote.
🖥️

NVIDIA

Photo Coming Soon

AI/GPU Compute

NVIDIA

Used NVIDIA H100 80GB PCIe/SXM GPU

Price on request

Ask for a quote tied to the required configuration and condition.

Contact for availability. Stock and condition require confirmation for a specific unit.

Place a Bid or Ask

Talk to a specialist · parts@caladansemi.com

Get a quote for this partPricing and availability to be confirmed

✓Request unit-specific inspection and test records
✓Confirm included documentation and accessories
✓Agree delivery, payment and return terms in the quote

Block 1:

How Much Does Used NVIDIA H100 80GB PCIe/SXM GPU Cost in 2026?

Used units range from $12,000 to $25,000 in 2026, while refurbished models sit at $18,000–$30,000, depending on memory health and PCIe 5.0 compliance. This represents 30–60% savings versus new H100s (priced ~$35,000+ in 2026). Caveat: GPUs with >3 years of continuous HBM2e memory usage may show >10% performance degradation in FP16 workloads due to wear on high-bandwidth memory modules. Always request a NVIDIA System Management Interface (nvidia-smi) stress test log to validate stability.

Block 2:

NVIDIA H100 80GB PCIe/SXM GPU Specifications and Compatible Tools

Key specs: Hopper H100 GPU architecture, 80GB HBM2e memory (2TB/s bandwidth), 3 TFLOPS FP16, 1.5 TFLOPS FP64, PCIe 5.0 x16 (or SXM5 variant), 300W TDP. Compatible with semiconductor tools like Applied Materials Endura ClusterTools (2022–2024 models), Lam Research Centura RS2 (2023+), and ASML Twinscan NXT:2000 (with NVIDIA AI-powered inspection modules). Exclusion: Do not install in systems using NVIDIA V100-based firmware (Part #NV-V100-FW-2.1), as Hopper’s SM 9.0 architecture requires updated driver stacks (535+).

Block 3:

What to Check Before Buying a Used NVIDIA H100 80GB PCIe/SXM GPU in 2026

  1. HBM2e memory integrity: Run nvidia-smi --query-gpu=memory.error.count --format=csv to confirm zero uncorrectable errors.
  2. PCIe 5.0 link stability: Use a PCIe 5.0 protocol analyzer to verify 32 GT/s signal integrity under load.
  3. Thermal throttling history: Check sensor logs for >1,000 cumulative hours above 80°C (reduces lifespan by 20–30%).
  4. Firmware lifecycle: Reject parts with firmware older than **NVIDIA GPU Driver 530

Quick Info

ConditionConfirm in quote
CategoryAI/GPU Compute
Secondary Market

Place a Bid or Ask

Submit your buying or selling requirements for Used NVIDIA H100 80GB PCIe/SXM GPU.

Want to Buy
Want to Sell

Confirm export requirements and payment terms before any transaction · parts@caladansemi.com