Closed Beta
Rough edges expected!AnalyzePGN
Welcome, NCM users! AnalyzePGN is a new tool from the creator of NCM. Paste a PGN and get deep Stockfish analysis of every position in your game.
Massively parallel
We scale out to as many AWS C7a cores (their fastest compute instances) as needed to analyze every position in parallel. Servers take about 15 seconds to spin up.
Resume analysis
Ran depth 30 but want to go deeper on the critical moves? Select a range and bump it to depth 45. We snapshot each running Stockfish process and store it for 48 hours, so analysis picks up exactly where it left off.
You can submit a full game for analysis, or click and drag a section of the move list to analyze or resume analysis on just that range.
Resume analysis on any selection — Stockfish picks up exactly where it left off.
Pricing
Still figuring this out. Target is about a penny per core per minute — so $0.60 for an hour of analysis time, delivered in seconds. During the beta there's no billing — I'll just be giving out credits. Pricing may change as I learn the real costs and find ways to drive them down.
Pay only for the milliseconds that Stockfish is actually running. For example, analyzing the mainline of Kasparov vs Topalov to depth 25 used 19m 24s of CPU time and finished in just over a minute from a cold start (including server spin-up time). That's about $0.19. (Would have been under a minute if not for 26. Qxd4+.)
NCM launched at $9/year in 2011 and is $19/year today after several hardware upgrades all the way up to Ryzen 7950X servers. Same mindset here.
How it actually works
We built a custom API with an autoscaler that launches EC2 instances and runs Stockfish 18 on them. When you submit a game (or a portion of one), we spin up as many cores as needed and each position gets its own dedicated C7a core — we never share cores.
The API is written in Elixir on the OTP and has been tested to a massive number of concurrent cores. It's rock solid. AnalyzePGN is a work in progress and the first client of the API — expect more to come.
Coming soon
192-core analysis — throw a full C7a instance at a single position for seconds, minutes, or even days. Our infrastructure is designed for it.