Skip to content

Hardware Specifications

Criteria: Supports NVIDIA GPUs of generations newer than Tesla, provided that at least 320 GB total GPU VRAM is available to each MLNode container. Any combination of GPUs is allowed, as long as the system can host the LLMs approved by network governance and participate in PoC.

NVIDIA GPU Release Date VRAM Architecture
B300 March 2025 288 GB HBM3e Blackwell Ultra
B200 2024 192 GB HBM3e Blackwell
RTX PRO 6000 March 2025 96 GB GDDR7 ECC Blackwell
H200 2024 141 GB HBM3e Hopper
H100 May 2022 80 GB HBM3 Hopper
A100 80GB November 2020 80 GB HBM2e Ampere