Supercomputing Compatibility Listings
SuperCompat – structured, authoritative compatibility matrices for the full AI/HPC software stack.
Resolve kernel, driver, and framework version mismatches before they break your supercomputing cluster.
CUDA
NVIDIA CUDA Toolkit ↔ GPU Driver ↔ Linux Kernel compatibility matrix.
View Matrix →ROCm
AMD ROCm ↔ AMDGPU Driver ↔ Kernel ↔ GPU architecture compatibility.
View Matrix →OFED / MLNX_OFED
OpenFabrics OFED ↔ Kernel ↔ RDMA/InfiniBand/RoCE protocol compatibility.
View Matrix →Open MPI
Open MPI ↔ UCX ↔ OFED ↔ CUDA/ROCm GPU-aware MPI compatibility.
View Matrix →Hardware Platforms
NVIDIA (Blackwell, Hopper, Volta, Ampere) and AMD (MI300, MI250, MI100) GPU platform requirements.
View Platforms →🤖 Machine-Readable API
All data is available as structured JSON or YAML for CI/CD pipelines, Ansible, and automation tools:
# JSON (default)
curl https://hicain.io/api/v1/cuda
# YAML
curl -H "Accept: application/yaml" https://hicain.io/api/v1/cuda/12.4
# Compatibility check
curl -X POST https://hicain.io/api/v1/check \
-H "Content-Type: application/json" \
-d '{"kernel":"5.15","cuda":"12.4","driver":"550.54.14","ofed":"23.10-0.5.5.0"}'Common Mismatches This Service Solves
- 🔴 CUDA / Driver mismatch – wrong NVIDIA driver for the installed CUDA toolkit
- 🔴 Kernel too new/old for CUDA or ROCm – DKMS module won't build
- 🔴 OFED incompatible with running kernel – InfiniBand/RoCE link down after OS update
- 🔴 Open MPI built against wrong UCX or OFED – silent MPI collective failures
- 🟡 EOL software in production – running unsupported stack versions in HPC clusters
- 🟡 GPU-aware MPI misconfiguration – CUDA/ROCm version outside Open MPI's tested range