Garden's

Almanac of Matter Models

A guide to Machine Learned Interatomic Potentials pre-installed on shared HPC clusters. Rootstock lets you run any of them through a drop-in ASE calculator.

Polaris
ALCF
GPU types
NVIDIA A100
● 32 verified ○ 8 lapsed // 17 not applicable
Sophia
ALCF
GPU types
NVIDIA A100
● 0 verified ○ 43 lapsed // 14 not applicable
Perlmutter
NERSC
GPU types
NVIDIA A100
● 40 verified ○ 17 lapsed // 0 not applicable
Delta
NCSA
GPU types
NVIDIA A100 / A40
● 0 verified ○ 0 lapsed // 57 not applicable
Model family / Checkpoint Size
Polaris
ALCF
NVIDIA A100
Sophia
ALCF
NVIDIA A100
Perlmutter
NERSC
NVIDIA A100
Delta
NCSA
NVIDIA A100 / A40
AllScAIP 1 checkpoint ·Attention / Graph-Transformer
allscaip-md-conserving-all-omol —
allscaip-md-direct-all-omol —
ANI 3 checkpoints ·Vanilla / Direct-Prediction
ani-2x —
ani-1ccx —
ani-1x —
CHGNet 0 checkpoints ·Message-Passing Neural Networks
chgnet-default 0.4M
DimeNet++ 4 checkpoints ·1.8M ·Message-Passing Neural Networks
dimenet-plus-plus-s2ef-oc20-all 1.8M
dimenet-plus-plus-s2ef-oc20-20m 1.8M
dimenet-plus-plus-s2ef-oc20-2m 1.8M
dimenet-plus-plus-s2ef-oc20-200k 1.8M
EquiformerV2 3 checkpoints ·31M – 153M ·Equivariant Models
equiformer-v2-153m-s2ef-oc20-all-md 153M
equiformer-v2-31m-s2ef-oc20-all-md 31M
equiformer-v2-83m-s2ef-oc20-2m 83M
eqv2-31m-mp 31M
eqv2-31m-omat 31M
eqv2-86m-omat 86M
eqv2-153m-omat 153M
eqv2-31m-omat-mp-salex 31M
eqv2-86m-omat-mp-salex 86M
eqv2-153m-omat-mp-salex 153M
eqv2-dens-31m-mp 31M
eqv2-dens-86m-mp 86M
eqv2-dens-153m-mp 153M
eSCN 4 checkpoints ·200M ·Equivariant Models
escn-l6-m2-lay12-s2ef-oc20-all-md —
escn-l6-m3-lay20-s2ef-oc20-all-md 200M
escn-l6-m2-lay12-s2ef-oc20-2m —
escn-l4-m2-lay12-s2ef-oc20-2m —
eSEN 3 checkpoints ·6.3M – 50.7M ·Equivariant Models
esen-30m-mptrj 30M
esen-30m-oam 30M
esen-30m-omat 30M
esen-md-direct-all-omol 50.7M
esen-sm-conserving-all-omol 6.3M
esen-sm-direct-all-omol 6.3M
GemNet 4 checkpoints ·32M – 168M ·Message-Passing Neural Networks
gemnet-oc-large-s2ef-oc20-all-md 168M
gemnet-oc-s2ef-oc20-all-md 39M
gemnet-oc-s2ef-oc20-all 39M
gemnet-dt-s2ef-oc20-all 32M
MACE-MP-0 3 checkpoints ·3.8M – 5.7M ·Atomic Cluster Expansion
mace-mp-0-small 3.8M
mace-mp-0-medium 4.7M
mace-mp-0-large 5.7M
MACE-OFF23 3 checkpoints ·4.7M ·Atomic Cluster Expansion
mace-off23-small —
mace-off23-medium 4.7M
mace-off23-large —
MACE-POLAR-1 3 checkpoints ·Atomic Cluster Expansion
mace-polar-1-s —
mace-polar-1-m —
mace-polar-1-l —
MatterSim 2 checkpoints ·1M – 5M ·Message-Passing Neural Networks
mattersim-v1-0-0-5m 5M
mattersim-v1-0-0-1m 1M
Orb-v2 3 checkpoints ·25M ·Vanilla / Direct-Prediction
orb-v2 25M
orb-d3-v2 25M
orb-mptraj-only-v2 25M
Orb-v3 8 checkpoints ·25.5M ·Vanilla / Direct-Prediction
orb-v3-conservative-20-omat 25.5M
orb-v3-conservative-20-mpa 25.5M
orb-v3-conservative-inf-omat 25.5M
orb-v3-conservative-inf-mpa 25.5M
orb-v3-direct-20-omat —
orb-v3-direct-20-mpa —
orb-v3-direct-inf-omat —
orb-v3-direct-inf-mpa —
PaiNN 1 checkpoint ·20.1M ·Equivariant Models
painn-s2ef-oc20-all 20.1M
SchNet 4 checkpoints ·9.1M ·Message-Passing Neural Networks
schnet-s2ef-oc20-all 9.1M
schnet-s2ef-oc20-20m —
schnet-s2ef-oc20-2m —
schnet-s2ef-oc20-200k —
SCN 3 checkpoints ·168M ·Equivariant Models
scn-s2ef-oc20-all-md 168M
scn-t4-b2-s2ef-oc20-2m —
scn-s2ef-oc20-2m —
TensorNet 1 checkpoint ·0.8M ·Equivariant Models
tensornet-matpes-pbe-2025-2 0.8M
TorchMD-Net 1 checkpoint ·Equivariant Models
torchmdnet-et-oc20 —
UMA 3 checkpoints ·150M – 1.4B ·Equivariant Models
uma-s-1 150M
uma-s-1p1 150M
uma-s-1p2 150M
uma-m-1p1 1.4B
57 checkpoints · 4 clusters · 72 verified installations, 68 lapsed, 88 not applicable verified lapsed not supported
How it works

Each model is pre-installed in its own isolated environment. Rootstock runs it as a subprocess on the same node and exchanges positions and forces over a local socket — so you can swap models without changing your dependencies. 

from ase.build import bulk
from rootstock import RootstockCalculator

atoms = bulk("Cu", "fcc", a=3.6) * (5, 5, 5)

with RootstockCalculator(
    cluster="perlmutter",
    checkpoint="mace-mp-0-medium",
    device="cuda",
) as calc:
    atoms.calc = calc
    print(atoms.get_potential_energy())