Hybrid
Run high-density video encoding/decoding/transcoding inside your own data center using Quadra VPUs without changing ingest paths, codecs, or downstream workflows.
Use this architecture when:
- Existing on-prem infrastructure is capacity-constrained
- Bursty or seasonal demand requires elastic scaling
- Cloud costs must be controlled while preserving flexibility
- Regulatory or data-sovereignty requirements limit full cloud migration
This architecture is optimized for cost arbitration, operational flexibility, and incremental expansion.
What changes
- Encoding workloads are dynamically placed based on cost and capacity
- Cloud is used surgically, not continuously
- Overall cost per stream decreases while peak capacity increases
What doesn’t
- Ingest sources, codecs, or output destinations
- FFmpeg/GStreamer-based workflows
- Operational ownership or pipeline logic
VPU placement
- Quadra VPUs operate in both on-prem servers and cloud instances
- Encoding layers are consistent across environments
- No dependency on GPU sharing or cloud-specific acceleration services
Scaling model
- Baseline capacity handled on-prem
- Burst capacity scales via cloud VPU instances
- Linear performance scaling without cross-environment contention
Prerequisites
- Quadra-enabled on-prem servers
- Access to VPU-enabled cloud instances
- Unified orchestration layer (Bitstreams or existing scheduler)
- Network connectivity between environments
Validation path
- Establish baseline performance on-prem
- Introduce cloud VPU instances for overflow testing
- Compare cost per stream across environments
- Gradually tune workload distribution rules
What this is not
- Not an active-active mirror deployment
- Not a cloud-first mandate
- Not a GPU fallback strategy
- Not a complex orchestration rewrite
Outcome
Lower total cost per stream, elastic capacity on demand, and unified operations across on-prem and cloud without duplicating pipelines.
Supported by the VPU Ecosystem, partners operating this architecture in production today.