Quick Answer
Key takeaways
- Browse live GPU capacity across 181 offerings, 30 GPU types, and 9 regions before deployment
- Choose GPU instances for custom environments or serverless inference API for supported models
- Authorize compute from wallet balance before requesting GPU resources
- Deploy LLM inference, computer vision, image generation, and video generation workloads
- Access SSH-ready instances with NVIDIA GPU configurations for persistent inference services
- Use scoped API keys and idempotency validation for serverless inference requests
Two Paths for AI Inference on Hostnot GPU
Hostnot GPU offers GPU instances for full control and serverless APIs for request-response workflows.
| Feature | GPU Instances | Serverless AI API |
|---|---|---|
| Use case | Custom models, persistent services, self-managed inference | Supported models, chat/image/video requests |
| Access method | SSH with public key | API endpoint with bearer token |
| Environment | Bring containers, dependencies, model artifacts | Pre-configured, model catalog discovery |
| Authorization | Wallet reserves rate before provisioning | Wallet authorization per request |
| Best for | Fine-tuned models, multi-step pipelines, experiments | Standard inference, API integration, validation |
GPU Instance Inference Capabilities
- LLM inference: serve or evaluate language models with required memory and context length
- Computer vision: process image and video workloads on appropriately sized GPUs
- Image generation: run Stable Diffusion and other generative models with persistent environments
- Video generation: deploy video synthesis models with GPU memory matching model requirements
- Custom environments: bring compatible dependencies, containers, and model artifacts
- SSH access: connect with registered public key, verify host fingerprint, keep private key local
- Review live marketplace for GPU count, memory, storage, region, capacity class, and rate
Serverless Inference API Workflow
- Discover current model catalog through customer-safe API endpoint
- Validate model capabilities before sending inference requests
- Send chat, image, or video requests with scoped API credentials
- Include unique idempotency keys for request deduplication
- Wallet authorization and usage visibility handled server-side
- Provider credentials and internal routing stay behind platform boundary
- Access 60 serverless models across supported inference types
Deployment Preparation for Inference Workloads
- Create account and fund wallet before requesting compute resources
- Register SSH public key for GPU instance access (never upload private key)
- Review GPU memory requirements for target model size and batch configuration
- Check model license and runtime dependencies before custom deployment
- Match GPU count, memory, and region to inference latency and throughput needs
- Confirm authorized rate displays before deployment; rates synchronize with capacity
- Save workload outputs before terminating instances; terminal closure is not lifecycle action
Frequently asked questions
What inference workloads run on Hostnot GPU?
Hostnot GPU supports large language model inference, computer vision, generative AI for images and video, and other machine learning inference applications using GPU instances or the serverless API.
How do I choose between GPU instances and serverless inference?
Use GPU instances for custom models, persistent services, or self-managed environments; use the serverless API for supported models with standard chat, image, or video request workflows.
Can I serve my own fine-tuned models?
Yes. Deploy fine-tuned models on GPU instances by bringing compatible environments, containers, and model artifacts that match the selected GPU memory and configuration.
What GPU types are available for inference?
Hostnot GPU provides 30 GPU types across 181 offerings, including NVIDIA-based configurations suitable for AI inference workloads where available.
How does wallet authorization work for inference?
Review the displayed rate before deployment; available wallet balance excludes active reservations, and usage is recorded after the billable state or request outcome.
Where can I see current GPU availability?
The live GPU Marketplace shows current customer-visible capacity, configurations, regions, and rates across all 181 offerings before you request deployment.
- Best for: AI developers, startups, and businesses deploying inference workloads for LLMs, computer vision, or generative AI with flexible GPU access
- Skip if: you need managed model hosting with zero infrastructure decisions or prefer vendor-locked inference endpoints
- Next step: review live GPU capacity and model catalog, fund your wallet, and deploy your first inference workload
Compare configurations and deployment options on Hostnot GPU on Launch Llama before starting your first inference deployment.