Flan-PaLM 540B (Google) and Nemotron-4 340B (NVIDIA) are both frontier AI models. Flan-PaLM 540B was published in October 2022 and Nemotron-4 340B in June 2024.
Nemotron-4 340B was trained on 1.8 x 10^25 FLOP, about 7.1x the compute of Flan-PaLM 540B at 2.5 x 10^24 FLOP. Training compute is the closest available proxy for how much was invested in a model, though it says nothing on its own about how well that compute was spent.
These two models share no benchmark on which both have been scored, so no direct performance comparison is possible here. The specification table below is a comparison of inputs, not of results.
| Field | Flan-PaLM 540B | Nemotron-4 340B |
|---|---|---|
| Organization | NVIDIA | |
| Published | Oct 20, 2022 | Jun 14, 2024 |
| Training compute | 2.5 x 10^24 FLOP | 1.8 x 10^25 FLOP |
| Parameters | 540B | 340B |
| Dataset size | 1.4B | 9T |
| Training hardware | Google TPU v4 | NVIDIA H100 SXM5 80GB |
| Chips used | 512 | 6,144 |
| Training time | 37 h | 2.2K h |
| Training cost (2023 USD) | — | $21M |
| Training power draw | 348.3 kW | 8.5 MW |
| Accessibility | Unreleased | Open weights (unrestricted) |
| Open weights | No | Yes |
| Country | United States | United States |
