Faster Inference of LLMs using FP8 on the Intel Gaudi

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Lee, Joonhyung, Markovich-Golan, Shmulik, Ohayon, Daniel, Hanani, Yair, Park, Gunho, Kim, Byeongwook, Karnieli, Asaf, Livne, Uri, Shen, Haihao, Huang, Tai, Kwon, Se Jung, Lee, Dongsoo
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916653312245760
author Lee, Joonhyung
Markovich-Golan, Shmulik
Ohayon, Daniel
Hanani, Yair
Park, Gunho
Kim, Byeongwook
Karnieli, Asaf
Livne, Uri
Shen, Haihao
Huang, Tai
Kwon, Se Jung
Lee, Dongsoo
author_facet Lee, Joonhyung
Markovich-Golan, Shmulik
Ohayon, Daniel
Hanani, Yair
Park, Gunho
Kim, Byeongwook
Karnieli, Asaf
Livne, Uri
Shen, Haihao
Huang, Tai
Kwon, Se Jung
Lee, Dongsoo
contents Low-precision data types are essential in modern neural networks during both training and inference as they enhance throughput and computational capacity by better exploiting available hardware resources. Despite the incorporation of FP8 in commercially available neural network accelerators, a comprehensive exposition of its underlying mechanisms, along with rigorous performance and accuracy evaluations, is still lacking. In this work, we contribute in three significant ways. First, we analyze the implementation details and quantization options associated with FP8 for inference on the Intel Gaudi AI accelerator. Second, we empirically quantify the throughput improvements afforded by the use of FP8 at both the operator level and in end-to-end scenarios. Third, we assess the accuracy impact of various FP8 quantization methods. Our experimental results indicate that the Intel Gaudi 2 accelerator consistently achieves high computational unit utilization, frequently exceeding 90% MFU, while incurring an accuracy degradation of less than 1%.
format Preprint
id arxiv_https___arxiv_org_abs_2503_09975
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Faster Inference of LLMs using FP8 on the Intel Gaudi
Lee, Joonhyung
Markovich-Golan, Shmulik
Ohayon, Daniel
Hanani, Yair
Park, Gunho
Kim, Byeongwook
Karnieli, Asaf
Livne, Uri
Shen, Haihao
Huang, Tai
Kwon, Se Jung
Lee, Dongsoo
Hardware Architecture
Low-precision data types are essential in modern neural networks during both training and inference as they enhance throughput and computational capacity by better exploiting available hardware resources. Despite the incorporation of FP8 in commercially available neural network accelerators, a comprehensive exposition of its underlying mechanisms, along with rigorous performance and accuracy evaluations, is still lacking. In this work, we contribute in three significant ways. First, we analyze the implementation details and quantization options associated with FP8 for inference on the Intel Gaudi AI accelerator. Second, we empirically quantify the throughput improvements afforded by the use of FP8 at both the operator level and in end-to-end scenarios. Third, we assess the accuracy impact of various FP8 quantization methods. Our experimental results indicate that the Intel Gaudi 2 accelerator consistently achieves high computational unit utilization, frequently exceeding 90% MFU, while incurring an accuracy degradation of less than 1%.
title Faster Inference of LLMs using FP8 on the Intel Gaudi
topic Hardware Architecture
url https://arxiv.org/abs/2503.09975