FP8 and INT4 Quantized Llama-3.1-70B Throughput and Accuracy on A100 and H100 GPUs

This report synthesises findings from 3 peer-reviewed papers addressing the following research question: What is the throughput difference between FP8 and INT4 quantized Llama-3.1-70B on HumanEval when deployed on A100 vs. H100 GPUs, and is the accuracy degradation consistent across both hardware. Large language models (LLMs) offer remarkable capabilities, yet their high inference costs restrict wider adoption. While increasing parameter counts improves accuracy, it also broadens the gap between state-of-the-art capabilities and practical deployability. 12 claims were extracted from source literature; 10 were independently verified against retrieved documents. An automated multi-reviewer quality assessment produced a score of 8.4/10. This report is a machine-generated literature synthesis and does not constitute original research.

Research goal: What is the throughput difference between FP8 and INT4 quantized Llama-3.1-70B on HumanEval when deployed on A100 vs. H100 GPUs, and is the accuracy degradation consistent across both hardware configurations?

Autonomous literature synthesis. Automated review score: 8.4/10. Full text and citation available at Assignee Research.