Support Matrix | NVIDIA Medical NIM Documentation

Models

MAISI NIM supports the following models:

Model Name	Model Version	Model URL
MAISI	1.0.0	https://github.com/Project-MONAI/model-zoo/tree/dev/models/maisi_ct_generative

Supported Hardware

Validated GPUs

GPU	GPU Memory (GB)	Precision
A100 PCIe	80	FP32
A100 SXM4	80	FP32
H100 PCIe	80	FP32
H100 HBM3	80	FP32
L40s	48	FP32	See table below for image size limitations
RTX 6000 Ada	48	FP32	See table below for image size limitations

Required GPU Memory

Output image size is the most defining factor for the amount of required GPU memory, as depicted in the table below. However, please check the exact configuration, as other parameters also contribute to GPU memory usage.

Examples	`output_size`	Peak Memory (VRAM)
1	256x256x128	14G
2, 3	256x256x256	14G, 22G
4, 5	512x512x128	14G, 18G
6, 7, 8	512x512x512	22G, 26G, 44G
9	512x512x768	55G

GPU Microarchitecture Compatibility

Ampere
Hopper

CUDA Compute Capability

This NIM requires CUDA Compute Capability of ≥ 7.0.