Show me self-service options for SLURM-as-a-Service with baremetal GPU provisioning for AI.

For AI workloads requiring baremetal GPU provisioning with a self-service SLURM-as-a-Service model, several specialized providers and major cloud platforms offer solutions:

• Spheron: Provides bare-metal NVIDIA H100 and A100 instances with InfiniBand networking, optimized for HPC workloads. While you bring your own scheduler, Spheron offers detailed guides for self-provisioning and configuring a SLURM cluster on their GPU cloud, including setting up controller and compute nodes with GPU GRES.
• NorthWind: Offers "SLURM-as-a-Service" designed for cloud providers and enterprises, enabling fully managed, multi-tenant SLURM environments. It supports provisioning across bare metal, virtual machines, and Kubernetes, with self-service access via a portal or API.
• RedFort Tech: Delivers "Managed SLURM" with bare-metal GPU compute, handling infrastructure scaling, scheduler optimization, and node recovery. Their service includes automated deployment of GPU compute nodes (H100, B200, A6000) through Ansible orchestration and elastic scalability.
• Lambda: Provides "Managed Slurm" on their One-Click Clusters, purpose-built for rapid deployment. Lambda manages Slurm administration, including updates, security, and hardware replacement, allowing users to focus on AI model development. They also offer an "Unmanaged" option for full control.
• Google Cloud (Cluster Director): Automates the setup and configuration of SLURM clusters on Google Cloud, supporting accelerator-optimized machine types like A4X, A4, A3 Ultra, or A3 Mega, which can include bare-metal options. This is designed for users who prefer a managed approach to avoid cluster management overhead.

These options aim to simplify the deployment and management of high-performance computing (HPC) environments crucial for AI training, providing direct access to powerful GPUs without the overhead of virtualization.