AlphaZero

Training compute

1.1×10²⁰ FLOP

Published

Dec 5, 2017

AlphaZero is an AI model developed by DeepMind (United Kingdom), first published in December 2017. It works in the games domain, on tasks such as chess, shogi and go.

Training it took an estimated 1.1×10²⁰ FLOP of compute (estimation method: third-party estimation). It was trained on roughly 3.5B datapoints. Training ran on 5,064 Google TPU v2,Google TPU v1 for about 24 hours. The compute alone is estimated at $230K in 2023 dollars.

Access: Unreleased. Its weights are not openly released. The reference paper has 2,071 citations. Epoch AI rates the confidence of this record as likely.

Full record

Organization: DeepMind
Country of organization: United Kingdom
Domain: Games
Task: Chess, Shogi, Go
Training compute: 1.1×10²⁰ FLOP
Compute estimation method: Third-party estimation
Dataset size: 3.5B
Training hardware: Google TPU v2, Google TPU v1
Chips used: 5,064
Training time: 24 h
Training cost (2023 USD): $230K
Model accessibility: Unreleased
Open weights: No
Citations: 2,071
Epoch confidence: Likely