Supported Hardware#
GenMol NIM v2.0.0 uses a single generic FP32 profile and runs on one GPU per container instance.
Model Profile#
Precision |
Disk Space |
Profile ID |
|---|---|---|
FP32 |
340MB |
9a60e7446868a0e75a97f5aa7f9b5312bfc8f221d6855b16a54dd0064ffd77b3 |
GPU#
The v2.0.0 validation matrix in genmol/tests/accuracy.yaml covers the GPU configurations below.
GPU |
GPU Memory |
# of GPUs |
|---|---|---|
NVIDIA RTX 6000 Ada Generation |
48GB |
1 |
NVIDIA L40S |
48GB |
1 |
NVIDIA A10G |
24GB |
1 |
NVIDIA A100 |
40GB or 80GB |
1 |
NVIDIA H100 |
80GB |
1 |
NVIDIA H200 |
141GB |
1 |
NVIDIA B200 |
180GB |
1 |
NVIDIA GB200 |
186GB |
1 |
NVIDIA B300 |
288GB |
1 |
NVIDIA GB300 |
288GB |
1 |
NVIDIA GB10 (DGX Spark) |
128GB |
1 |
NVIDIA GH200 144GB |
144GB |
1 |
NVIDIA RTX PRO 6000 Blackwell Workstation Edition |
96GB |
1 |
Note
The GB10 entry corresponds to the DGX Spark platform and uses 128GB of unified memory shared across CPU and GPU.
Storage#
NVIDIA recommends at least 20GB of local disk space for the container image, model assets, and cache directory.
CUDA Compute Capability#
This NIM requires CUDA Compute Capability >= 7.0.
Supported Software#
Prerequisite |
Version |
Link |
|---|---|---|
NVIDIA GPU Driver |
580.126.20 or newer |
|
NVIDIA Container Toolkit |
latest |
https://docs.nvidia.com/datacenter/cloud-native/container-toolkit/latest/install-guide.html |
Docker |
latest |
Ubuntu, CentOS, Debian: https://docs.docker.com/engine/install/ |