Orchestration
NeuroHash supports multiple orchestration solutions to manage your GPU workloads. Choose the tool that best fits your workflow.
Kubernetes
Industry-standard container orchestration, optimized for GPU workloads.
Options
| Tier | Description | Best For |
|---|---|---|
| Managed Kubernetes | Fully managed control plane, automatic updates, monitoring | Teams that want zero ops overhead |
| Pre-installed Kubernetes | K8s pre-configured on your cluster, you manage workloads | Teams with existing K8s expertise |
| Self-installed | Bare metal access, install your own K8s distribution | Teams with custom requirements |
Features
- NVIDIA GPU Operator pre-configured
- GPU scheduling and resource management
- Helm charts for common ML frameworks
- Horizontal pod autoscaling for inference
- Persistent volume support for datasets
Slurm
High-performance computing (HPC) workload manager, the standard in academic and research environments.
Features
- Native GPU-aware job scheduling
- Multi-node distributed training support
- Job queuing with priority management
- Integration with MPI for distributed workloads
- Familiar to researchers coming from university HPC clusters
Use Cases
- Large-scale distributed training jobs
- Batch processing of training runs
- Multi-user shared cluster environments
- Research workloads with complex dependencies
dstack
AI-native, open-source orchestration platform built specifically for ML workflows.
Features
- Define training runs as code (YAML configs)
- Automatic GPU provisioning
- Built-in experiment tracking
- Cost optimization and spot instance support
- Native integration with popular ML frameworks
Example Configuration
type: task
python: "3.11"
commands:
- pip install -r requirements.txt
- python train.py --epochs 100 --batch-size 32
resources:
gpu: 1
gpu_name: H100
Choosing an Orchestrator
| Criteria | Kubernetes | Slurm | dstack |
|---|---|---|---|
| Container-native | Yes | Limited | Yes |
| HPC/batch jobs | Limited | Excellent | Good |
| ML-specific features | With add-ons | Limited | Built-in |
| Learning curve | Moderate | Low (for HPC) | Low |
| Multi-tenant | Yes | Yes | Limited |
| Best for | Production ML ops | Research/HPC | ML experimentation |
Getting Started
Orchestration options are available with 1-Click Clusters and Private Cloud deployments. Contact us to discuss the best setup for your team.