Supported Hardware#
The OpenFold2 NIM runs on a single GPU, and supports input sequences with length less than (or equal to) 2048 AAs.
GPU#
The OpenFold2 NIM is supported to run on the GPU configurations below.
GPU |
GPU Memory (GB) |
Precision |
# of GPUs |
|---|---|---|---|
NVIDIA A100 80GB PCIe |
80 |
BF16 |
[1] |
NVIDIA A100 40GB PCIe |
40 |
BF16 |
[1] |
NVIDIA A100-SXM4-80GB |
80 |
BF16 |
[1] |
NVIDIA B200 |
180 |
BF16 |
[1] |
NVIDIA B300 |
288 |
BF16 |
[1] |
NVIDIA GB10 (DGX Spark) |
128 |
BF16 |
[1] |
NVIDIA GB200 |
186 |
BF16 |
[1] |
NVIDIA GB300 |
288 |
BF16 |
[1] |
NVIDIA GH200 144GB |
144 |
BF16 |
[1] |
NVIDIA H100 80GB HBM3 |
80 |
BF16 |
[1] |
NVIDIA H100 80GB PCIe |
80 |
BF16 |
[1] |
NVIDIA H100 NVL |
94 |
BF16 |
[1] |
NVIDIA H200 |
141 |
BF16 |
[1] |
NVIDIA H200 NVL |
141 |
BF16 |
[1] |
NVIDIA L40S |
48 |
BF16 |
[1] |
NVIDIA RTX 6000 Ada Generation |
48 |
BF16 |
[1] |
NVIDIA RTX PRO 6000 Blackwell Server Edition |
96 |
BF16 |
[1] |
NVIDIA RTX PRO 6000 Blackwell Workstation Edition |
96 |
BF16 |
[1] |
Note
The GB10 features 128GB of LPDDR5x unified memory shared between CPU and GPU, providing seamless memory access across the Grace Blackwell Superchip architecture.
Storage#
NVIDIA recommends at least 80GB of disk space. At startup, the NIM downloads model parameters (roughly 5.2GB) into the NIM cache if they are not present on the local filesystem.
CPU and System Memory#
At least 64 GB of RAM
A CPU with at least 8 available cores