Compare · Two sizes · H100 and B200
Llama 3.3 70B Instruct vs Llama 3.1 405B Instruct FP8: GPU energy per million output tokens
On the index rows, Llama 3.1 405B Instruct FP8 uses 4.88× the GPU energy per output token of Llama 3.3 70B Instruct on the H100 (572.9 against 117.5 Wh/M, on eight cards and four cards at running batches of 255 and 724) and 5.25× on the B200 (318.5 against 60.7 Wh/M, on four cards and two cards at running batches of 510 and 1,022), GPU only, each at the most energy-efficient configuration the ML.ENERGY Leaderboard v3.0 measured for it (data compiled 16 February 2026). Llama 3.3 70B Instruct has 70 B parameters; Llama 3.1 405B Instruct FP8 has 405 B parameters. Part of each gap is the sweep rather than the model: the two rows of a pair differ in card count and running batch, and the table states both.
Unit watt-hours per million output tokens (Wh/M). GPU only, busy server. Every row is a floor. Data llama-3-3-70b-instruct.json, llama-3-1-405b-instruct-fp8.json. Updated 19 October 2026.
Cite this page: Lead Green, a Katonic project. Llama 3.3 70B Instruct vs Llama 3.1 405B Instruct FP8: GPU energy per million output tokens. https://leadgreen.ai/compare/llama-3-3-70b-instruct-vs-llama-3-1-405b-instruct-fp8/, release 2026-10-19, retrieved [date]. CC BY 4.0.
Each ratio compares the two models' index rows on the same GPU: the lowest-energy configuration the leaderboard measured for each, at or under 500 ms median inter-token latency. The ratio is the second model's energy per token over the first's, so a figure under 1 means the second model's row is the lower one. Card count and running batch differ between the rows and are shown in the table.
The rows
Index rows on every GPU both models share
| GPU | Model | Cards | Precision | Running batch | GPU W | Tokens / s | Wh / M | Output tokens | Wh per answer | kg CO2e / bn tokens, world grid |
|---|---|---|---|---|---|---|---|---|---|---|
| H100 | Llama 3.3 70B Instruct | 4 × H100 | BF16 | 724 | 2,546 | 6,018 | 117.5 | 413 | 0.049 | 55.6 |
| H100 | Llama 3.1 405B Instruct FP8 | 8 × H100 | FP8 | 255 | 4,875 | 2,363 | 572.9 | 367 | 0.210 | 271.0 |
| B200 | Llama 3.3 70B Instruct | 2 × B200 | BF16 | 1,022 | 1,935 | 8,854 | 60.7 | 409 | 0.025 | 28.7 |
| B200 | Llama 3.1 405B Instruct FP8 | 4 × B200 | FP8 | 510 | 3,883 | 3,386 | 318.5 | 380 | 0.121 | 150.7 |
Running batch is the average number of requests the server was decoding at once over the steady-state window (vLLM's num_requests_running). GPU W is the measured average power summed across the cards. Wh per answer is Wh/M times the row's own average output length. Carbon uses the world average grid, 473 gCO2e per kWh (Ember, 2024); the grid pages price every model on each listed grid. Model names link to every measured configuration.
What explains it
Two sizes
The second model has 5.8× the parameters of the first, and a dense model reads every parameter for every output token, so the weights moved per token scale with size and so does the memory needed to hold them. On the H100 the rows sit on four cards (Llama 3.3 70B Instruct) and eight cards (Llama 3.1 405B Instruct FP8), at running batches of 724 (Llama 3.3 70B Instruct) and 255 (Llama 3.1 405B Instruct FP8). The energy gap on the H100 (4.88×) is smaller than the parameter gap (5.8×), because each row runs at its own card count and running batch, where its fixed draw is spread over its own number of tokens. Size is the largest lever after load; the question for a buyer is whether the task needs the larger model's answers.
Load still moves the number more than either model does: Qwen 3 32B on 2 × H100 spans 7.1× between a running batch of 8 and 256 on the same cards, which is wider than the gap on this page. Pick the model for the task, then run it busy. The running batch and floor entries define the terms.
Questions
Asked about Llama 3.3 70B Instruct against Llama 3.1 405B Instruct FP8
Is Llama 3.3 70B Instruct more energy-efficient than Llama 3.1 405B Instruct FP8?
Yes, on these rows: on the ML.ENERGY Leaderboard's index rows, Llama 3.1 405B Instruct FP8 uses 4.88× the GPU energy per output token of Llama 3.3 70B Instruct on the H100 (572.9 against 117.5 Wh per million output tokens) and 5.25× the GPU energy per output token of Llama 3.3 70B Instruct on the B200 (318.5 against 60.7 Wh per million output tokens). These are floors at high load, GPU only, and the two rows of a pair differ in card count and running batch.
How much more energy does Llama 3.1 405B Instruct FP8 use per million tokens than Llama 3.3 70B Instruct?
The index rows differ by 455.4 Wh/M on the H100 (455.4 kWh per billion tokens) and 257.8 Wh/M on the B200 (257.8 kWh per billion tokens). Per answer, at each row's own average output length, Llama 3.3 70B Instruct costs 0.049 Wh on the H100 and 0.025 Wh on the B200 and Llama 3.1 405B Instruct FP8 0.210 Wh on the H100 and 0.121 Wh on the B200.
Does the bigger model cost more per answer?
Yes, at these rows: on the H100, 0.049 Wh for Llama 3.3 70B Instruct against 0.210 Wh for Llama 3.1 405B Instruct FP8, using each row's own average output length (413 and 367 tokens). The larger model also needs more cards to hold its weights (4 against 8 on the H100). Whether its answers are worth the difference is a question about the task, not the index.
Why are the two rows not like for like?
The leaderboard sweeps each model separately: card count, the cap on requests in flight and the precision are chosen per model, and the running batch actually reached differs. On the H100 the rows here ran on 4 and 8 cards at running batches of 724 and 255. The ratio compares the best measured row on each side, not a matched run; every configuration is on the model pages and in the API, so any pair can be re-cut at a matched cap or running batch.
Where do these figures come from?
The ML.ENERGY Leaderboard v3.0 (data compiled 16 February 2026, retrieved 26 September 2026), Text Conversation task, GPU energy measured with Zeus under vLLM on H100 80GB HBM3 and B200 servers; each row is the lowest-energy configuration for its model and GPU at or under 500 ms median inter-token latency. Carbon per billion tokens multiplies the row by the world average grid intensity, 473 gCO2e per kWh (Ember, 2024).
Sources
Where every figure came from
ML.ENERGY Leaderboard version 3.0 (released 1 December 2025; data compiled 16 February 2026; retrieved 26 September 2026), The ML.ENERGY Initiative, started at SymbioticLab, University of Michigan. Text Conversation task, 328 stable configurations. Energy per token is steady-state GPU energy divided by the output tokens generated in the window, measured with Zeus under vLLM 0.11.1 on H100 80GB HBM3 and B200 servers; energy per request is that figure times the average output length. Each index row is the lowest-energy configuration for its model and GPU at or under a 500 ms median inter-token latency, the leaderboard's own default ceiling, which no configuration reaches. Apache 2.0 per the ml-energy/benchmark-v3 dataset card. Cite Chung et al., The ML.ENERGY Benchmark, NeurIPS 2025 Datasets and Benchmarks.
Parameters and architecture as the leaderboard records them (total and activated parameters in billions). Carbon: Ember, Yearly electricity data, October 2026 release, world average 2024, CC BY 4.0, retrieved 17 October 2026.
Cite as. Lead Green, a Katonic project. Green Tokens: GPU energy per million output tokens. https://leadgreen.ai/index/, release 2026-10-19, retrieved [date]. © 2026 Katonic AI Technology Ltd. Licensed CC BY 4.0, offered as-is and without warranty; state whether you changed any rows or columns. Underlying data: ML.ENERGY Leaderboard v3.0 (The ML.ENERGY Initiative; Apache 2.0 per the ml-energy/benchmark-v3 dataset card; cite Chung et al., The ML.ENERGY Benchmark, NeurIPS 2025 Datasets and Benchmarks, arXiv 2505.06371); Ember, Yearly electricity data, October 2026 release (CC BY 4.0); AI Energy Score (Apache-2.0 as tagged on its Space; results reproduced as facts with attribution); NVIDIA figures reproduced as facts from the documents named in the sources ledger.