Saved in:
Bibliographic Details
Main Authors: Ringlein, Burkhard, Parnell, Thomas, Stoica, Radu
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2505.03780
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915395977347072
author Ringlein, Burkhard
Parnell, Thomas
Stoica, Radu
author_facet Ringlein, Burkhard
Parnell, Thomas
Stoica, Radu
contents As LLMs grow in complexity, achieving state-of-the-art performance requires tight co-design across algorithms, software, and hardware. Today's reliance on a single dominant platform limits portability, creates vendor lock-in, and raises barriers for new AI hardware. In this work, we make the case for combining just-in-time (JIT) compilation with comprehensive kernel parameter autotuning to enable portable LLM inference with state-of-the-art performance without code changes. Focusing on performance-critical LLM kernels, we demonstrate that this approach explores up to 15x more kernel parameter configurations, produces significantly more diverse code across multiple dimensions, and even outperforms vendor-optimized implementations by up to 230%, all while reducing kernel code size by 70x and eliminating manual code optimizations. Our results highlight autotuning as a promising path to unlocking model portability across GPU vendors.
format Preprint
id arxiv_https___arxiv_org_abs_2505_03780
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle GPU Performance Portability needs Autotuning
Ringlein, Burkhard
Parnell, Thomas
Stoica, Radu
Hardware Architecture
Artificial Intelligence
Programming Languages
As LLMs grow in complexity, achieving state-of-the-art performance requires tight co-design across algorithms, software, and hardware. Today's reliance on a single dominant platform limits portability, creates vendor lock-in, and raises barriers for new AI hardware. In this work, we make the case for combining just-in-time (JIT) compilation with comprehensive kernel parameter autotuning to enable portable LLM inference with state-of-the-art performance without code changes. Focusing on performance-critical LLM kernels, we demonstrate that this approach explores up to 15x more kernel parameter configurations, produces significantly more diverse code across multiple dimensions, and even outperforms vendor-optimized implementations by up to 230%, all while reducing kernel code size by 70x and eliminating manual code optimizations. Our results highlight autotuning as a promising path to unlocking model portability across GPU vendors.
title GPU Performance Portability needs Autotuning
topic Hardware Architecture
Artificial Intelligence
Programming Languages
url https://arxiv.org/abs/2505.03780