Nemotron-4 340B (NVIDIA) and PaLM 2 (Google) are both frontier AI models. Nemotron-4 340B was published in June 2024 and PaLM 2 in May 2023.
Nemotron-4 340B was trained on 1.8 x 10^25 FLOP, about 2.5x the compute of PaLM 2 at 7.3 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 | Nemotron-4 340B | PaLM 2 |
|---|---|---|
| Organization | NVIDIA | |
| Published | Jun 14, 2024 | May 10, 2023 |
| Training compute | 1.8 x 10^25 FLOP | 7.3 x 10^24 FLOP |
| Parameters | 340B | 340B |
| Dataset size | 9T | 3.6T |
| Training hardware | NVIDIA H100 SXM5 80GB | Google TPU v4 |
| Chips used | 6,144 | — |
| Training time | 2.2K h | — |
| Training cost (2023 USD) | $21M | $5M |
| Training power draw | 8.5 MW | — |
| Accessibility | Open weights (unrestricted) | API access |
| Open weights | Yes | No |
| Country | United States | United States |
