Pick a model size, quantization, and node count. Estimates are calibrated against real measured results from our 8× GB10 cluster — including the multi-node overhead the spec-sheet math ignores.
Enter your model above.
This is a rough floor — the single-stream decode rate the memory bandwidth allows, before tuning. A well-tuned deployment (graph capture, speculative decoding / multi-token prediction) runs meaningfully above the floor; the out-of-box state runs 2–5× below it. Treat as an order-of-magnitude sanity check, not a promise (±40%). Multi-node figures assume a correctly configured 100 Gbps+ fabric; a misconfigured one costs more than any number here (see the 13 Gbps trap).
That gap is exactly what we fix. The 1-day diagnostic measures where your real number is, why, and what's recoverable — in writing, or it's free.
Book a 1-day diagnostic →