Hardware Specifications
Criteria: Nvidia GPUs belonging to generations after Tesla, with a minimum of 16 GB VRAM per GPU. Each MLNode container must work with at least 40GB VRAM total.
| NVIDIA GPU | Release Date | VRAM | Architecture | Generation |
|---|---|---|---|---|
| H200 | 2024 | 141 GB of HBM3e | Hopper | H100 (Data Center) |
| H100 | May 2022 | 80 GB HBM3 | Hopper | H100 (Data Center) |
| A100 | May 2020 | 40 GB or 80 GB HBM2e | Ampere | A100 |
| RTX 6000 Ada Gen | December 2022 | 48 GB GDDR6 | Ada Lovelace | RTX 40 Series |
| RTX A6000 | December 2020 | 48 GB GDDR6 | Ampere | RTX 30 Series |
| L40 | 2022 | 48 GB GDDR6 | Ada Lovelace | L-Series (Data Center) |
| A40 | 2021 | 48 GB GDDR6 | Ampere | A40 |
| RTX 4090 (>= 2 per MLNode) | October 2022 | 24 GB GDDR6X | Ada Lovelace | RTX 40 Series |
| RTX 3090 (>= 2 per MLNode) | September 2020 | 24 GB GDDR6X | Ampere | RTX 30 Series |
| L4 (>= 2 per MLNode) | March 2023 | 24 GB GDDR6 | Ada Lovelace | L-Series (Data Center) |