Show HN: LLM Simulation – Experience TTFT and tokens/SEC before investing (llmsimulation.ht-x.com)

Show HN: LLM Simulation is a small benchmarking/simulation tool that predicts inference latency and throughput on your hardware using 207 real-world benchmarks. It reports two practical metrics — TTFT (time-to-first-token) and tokens/SEC (throughput) — so you can estimate how a model will behave before buying GPUs or committing to a deployment. TTFT is calculated for a typical 500‑token prompt, and the tool models the linear growth of latency with prompt length using the formula TTFT = 50ms + (tokens / prompt_speed). It also highlights that longer contexts increase KV‑cache memory use, a key constraint for large context models. For AI/ML engineers and infra planners this is useful for cost/latency trade-offs, hardware selection, and prompt engineering: tokens/SEC helps compare raw throughput across quantizations and accelerators, while TTFT exposes startup latency that matters for interactive apps. Because the estimates are grounded in 207 actual benchmarks, you get empirical guidance on whether your target latency and memory footprint are achievable, how much context will blow up RAM, and where to tune (model size, quantization, batch size, or prompt length) to hit production SLAs.