Current shape
--
----INFERENCE / HOST CAPACITY
Loading Math.js...
Size the orchestration layer
Keep the workload moving.
Translate a real inference request rate into CPU demand, safe throughput, and the headroom a denser host shape could create.
Scenario model, not benchmark
The source makes forecasts about Venice and agentic AI. This lab uses only your workload assumptions and does not verify a chip, SKU, GPU, or investment claim.
Workload and host assumptions
CPU time is host orchestration time per request. Safe utilization is the ceiling you are willing to operate at.
Canonical result
No capacity run yet
The useful sizing decision appears before export.
--
CPU cores demanded by this workload
--current safe requests / sec
--candidate safe requests / sec
--current safe cores
--candidate safe cores
--candidate headroom cores
--safe utilization input
Candidate shape
--
----How the capacity movesDemanded cores = requests/sec x CPU ms/request / 1000. Safe cores = hosts x cores/host x utilization. Safe request capacity = safe cores x 1000 / CPU ms/request.
Interpretation: This estimates the CPU orchestration layer only. It excludes memory bandwidth, GPU service time, queueing, retries, tail latency, power, prices, VM/SKU behavior, and model quality. Use measured traces before a production or procurement decision.