arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.14262cs.AR
收藏

ABI: 一种紧密集成、统一、稀疏感知、可重构、近寄存器文件/缓存计算的GPU架构,配备轻量级softmax,用于深度学习、线性代数和伊辛计算

ABI: A tightly integrated, unified, sparsity-aware, reconfigurable, compute near-register file/cache GPU architecture with light-weight softmax for deep learning, linear algebra, and Ising compute

The Department of Electrical and Computer Engineering, The University of Texas at Austin(德克萨斯大学奥斯汀分校电气与计算机工程系)

Siddhartha Raman Sundara Raman, Jaydeep P. Kulkarni

AI总结:

提出一种紧密集成的近内存GPU架构,通过自定义稀疏感知电路和轻量级softmax电路,在CNN、GCN、线性规划、大语言模型和伊辛工作负载上实现6-16倍加速和6-13倍能效提升。

摘要:

我们提出一种紧密集成且统一的近内存GPU架构,与MIAOW GPU相比,在卷积神经网络、图卷积网络、线性规划、大语言模型和伊辛工作负载上实现了6到16倍的加速和6到13倍的能耗节省。该设计包括一个自定义的稀疏感知近内存电路,可节省约1.5倍能耗,以及一个轻量级softmax电路,可节省约1.6倍能耗。该架构支持可重构计算,最高可达INT16,具有动态分辨率更新,并能跨问题规模高效扩展。启用ABI的MI300和Blackwell系统相比基线MI300和Blackwell实现了约4.5倍的加速。

英文摘要:

We present a tightly integrated and unified near-memory GPU architecture that delivers 6 to 16 times speedup and 6 to 13 times energy savings across Convolutional Neural Networks, Graph Convolutional Networks, Linear Programming, Large Language Models, and Ising workloads compared to MIAOW GPU. The design includes a custom sparsity-aware near-memory circuit providing about 1.5 times energy savings, and a lightweight softmax circuit providing about 1.6 times energy savings. The architecture supports reconfigurable compute up to INT16 with dynamic resolution updates and scales efficiently across problem sizes. ABI-enabled MI300 and Blackwell systems achieve about 4.5 times speedup over baseline MI300 and Blackwell.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑