I will set up, validate and troubleshoot your gpu cluster and ai infrastructure

You have GPUs. You need them to behave like a reliable compute platform.

I build and run GPU and HPC infrastructure in production: provisioning, driver and CUDA stack validation, scheduling, high-speed networking and monitoring. I work at the infrastructure layer. I am not an ML engineer and will not pretend to be one.

WHAT I DO

GPU node bring-up: NVIDIA driver, CUDA toolkit and runtime alignment, persistence mode, MIG

Health validation: DCGM diagnostics, Xid and ECC error triage, thermal and power checks

Multi-node GPU communication: nccl-tests all-reduce benchmarking, NCCL tuning, network path validation

Scheduling: SLURM GPU partitions and GRES, cgroups and binding, or Kubernetes with the NVIDIA device plugin and GPU Operator

Monitoring: Prometheus, DCGM exporter, Grafana dashboards

Automation: Ansible roles so the build is repeatable

BEFORE YOU ORDER

Message me with node count, GPU model, interconnect and what you are seeing. If it is not something I can genuinely fix, I will say so.

You get working configuration, benchmark output showing what changed, and written documentation.