ConfigIQ
LLM inference sizing, GPU comparison, and cost modeling for engineers and infrastructure teams.
A free community service, provided as-is without warranty. All results are best-effort estimates for planning, not guarantees of real-world performance.- What performance will I get?
- Predict performance takes a model, GPU, and parallelism you already have and estimates TTFT, TPOT, and throughput.
- What should I buy?
- Recommend sizing starts from latency and throughput targets and searches for GPU count, tensor and pipeline parallelism, and replicas.
- Will the model fit?
- KV cache calculator estimates memory and KV headroom for a model and workload shape.
A complete comparison, including REST endpoints and MCP tools, is in User scenarios.