DECA: A Near-Core LLM Decompression Accelerator Grounded on a 3D Roofline Model

TL;DR

This paper introduces DECA, a near-core decompression accelerator for large language models that improves inference performance by offloading decompression tasks, supported by a 3D Roofline performance model.

Contribution

It presents a novel analytical 3D Roofline model for understanding GeMM performance and a dedicated hardware accelerator with ISA extensions for efficient decompression.

Findings

DECA accelerates compressed GeMMs by up to 4x.

Reduces next-token generation time for Llama2-70B and OPT-66B by 1.6x-2.6x.

Provides insights into memory, vector, and hardware interactions for LLM inference.

Abstract

To alleviate the memory bandwidth bottleneck in Large Language Model (LLM) inference workloads, weight matrices are stored in memory in quantized and sparsified formats. Hence, before tiles of these matrices can be processed by in-core generalized matrix multiplication (GeMM) hardware engines, they need to be dequantized and de-sparsified. This is currently performed in software with vector operations. Unfortunately, this approach delivers only modest performance. Moreover, it is hard to understand how to improve the system, as the overall GeMM performance depends on the interaction between memory resources, vector units, and hardware matrix engines. To improve the performance of LLM inference in advanced platforms equipped with in-core GeMM engines and HBM, this paper makes three main contributions. First, it develops an analytical performance model with a 3D visual representation that provides insights into how memory resources, vector units, and hardware matrix engines interact to deliver compressed GeMM performance. Second, it proposes DECA, a new near-core ML-model decompression accelerator. DECA offloads tile de-sparsification and dequantization from the CPU, producing ready-to-use tiles for in-core GeMM engines. Third, it introduces a new ISA extension that enables out-of-order invocation of the near-core accelerator. With this extension, accelerator and core computations can interleave and overlap with high-performance. Our evaluation shows that, in a simulated 56-core Xeon 4 server with HBM, DECA accelerates the execution of compressed GeMMs by up to 4x over the use of optimized Intel software kernels. Further, DECA reduces the next-token generation time of Llama2-70B and OPT-66B by 1.6x-2.6x.