Wave GPU Roadmap

v1 — Observability & Waste Detection (in development)

• GPU inventory and device metadata (DCGM-based)
• Real-time utilization, memory, power, temperature, and ECC metrics
• Per-namespace and per-label-group cost aggregation
• 5 rule-based waste detection patterns (idle, underutilized, unassigned, GPU over-allocation, memory over-allocation)

v2 — Smart GPU Sizing (planned)

Bring Wave's right-sizing discipline to GPU workloads. Recommend optimal per-workload GPU allocation based on real utilization patterns, not static requests.

v3 — GPU Sharing via HAMi (planned)

Integrate with the HAMi project to enable fractional GPU allocation and multi-tenant sharing. Split expensive GPUs across jobs with predictable isolation and scheduling.

Have Input?

Reach out — we're actively shaping priorities.