# Prefill Latency on Llama-3-8B 500 Mbps

Source: [Prefill Latency on Llama-3-8B 500 Mbps](<https://www.sota2.com/research/sota/prefill-latency-on-llama-3-8b-500-mbps>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-05-28T07:31:29.821Z

Evaluation of the prefill latency performance for the Llama-3-8B model operating under a 500 Mbps constraint.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Prefill Latency |
| Dataset (including any stated split) | Llama-3-8B 500 Mbps |
| Metric | Prefill Latency (s) |
| Unit | TIME |
| Direction | Lower is better |
| Results loaded | 8 of 8 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [Prefill Latency (s)](<https://www.sota2.com/research/sota/prefill-latency-on-llama-3-8b-500-mbps.md?metric=Prefill%20Latency%20(s)>): Lower is better; unit: TIME.

## Recorded results

| Method | Prefill Latency (s) | Result date | Report paper |
| --- | --- | --- | --- |
| BP | 1.485 | Not recorded | 68fa78c01e209131cea7a48b |
| ASTRA | 1.54 | Not recorded | 68fa78c01e209131cea7a48b |
| ASTRA | 1.548 | Not recorded | 68fa78c01e209131cea7a48b |
| BP | 1.561 | Not recorded | 68fa78c01e209131cea7a48b |
| ASTRA | 1.583 | Not recorded | 68fa78c01e209131cea7a48b |
| SP | 2.052 | Not recorded | 68fa78c01e209131cea7a48b |
| Llama-3-8B (Single-device) | 4.578 | Not recorded | 68fa78c01e209131cea7a48b |
| TP | 10.14 | Not recorded | 68fa78c01e209131cea7a48b |

### BP

Prefill Latency (s): 1.485. Result ID: 6a17ee5dfbc373592c114a45.

[Reporting paper](<https://www.sota2.com/research/paper/68fa78c01e209131cea7a48b.md>)

Recorded evaluation configuration: {"Strategy":"Block Parallelism","Nb":"4"}. This record does not assert independent reproduction.

### ASTRA

Prefill Latency (s): 1.54. Result ID: 6a17ee5dfbc373592c114a47.

[Reporting paper](<https://www.sota2.com/research/paper/68fa78c01e209131cea7a48b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78c01e209131cea7a48b.md>)

Recorded evaluation configuration: {"G":"1"}. This record does not assert independent reproduction.

### ASTRA

Prefill Latency (s): 1.548. Result ID: 6a17ee5dfbc373592c114a48.

[Reporting paper](<https://www.sota2.com/research/paper/68fa78c01e209131cea7a48b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78c01e209131cea7a48b.md>)

Recorded evaluation configuration: {"G":"16"}. This record does not assert independent reproduction.

### BP

Prefill Latency (s): 1.561. Result ID: 6a17ee5dfbc373592c114a46.

[Reporting paper](<https://www.sota2.com/research/paper/68fa78c01e209131cea7a48b.md>)

Recorded evaluation configuration: {"Strategy":"Block Parallelism","Nb":"8"}. This record does not assert independent reproduction.

### ASTRA

Prefill Latency (s): 1.583. Result ID: 6a17ee5dfbc373592c114a49.

[Reporting paper](<https://www.sota2.com/research/paper/68fa78c01e209131cea7a48b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa78c01e209131cea7a48b.md>)

Recorded evaluation configuration: {"G":"32"}. This record does not assert independent reproduction.

### SP

Prefill Latency (s): 2.052. Result ID: 6a17ee5dfbc373592c114a44.

[Reporting paper](<https://www.sota2.com/research/paper/68fa78c01e209131cea7a48b.md>)

Recorded evaluation configuration: {"Strategy":"Sequence Parallelism"}. This record does not assert independent reproduction.

### Llama-3-8B (Single-device)

Prefill Latency (s): 4.578. Result ID: 6a17ee5dfbc373592c114a42.

[Reporting paper](<https://www.sota2.com/research/paper/68fa78c01e209131cea7a48b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa75d11e209131cea6a028.md>)

[Implementation](<https://github.com/meta-llama/llama-agentic-system>)

Recorded evaluation configuration: {"Device":"Single-device"}. This record does not assert independent reproduction.

### TP

Prefill Latency (s): 10.14. Result ID: 6a17ee5dfbc373592c114a43.

[Reporting paper](<https://www.sota2.com/research/paper/68fa78c01e209131cea7a48b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa67331e209131cea14ad7.md>)

[Implementation](<https://github.com/NVIDIA/Megatron-LM>)

Recorded evaluation configuration: {"Strategy":"Tensor Parallelism"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.