AMD 矩阵核心的精确模型
Accurate Models of AMD Matrix Cores

原始链接: https://arxiv.org/abs/2609.14845

本文探讨了现代 GPU 矩阵乘法器中缺乏 IEEE 754 标准合规性的问题,并指出舍入方式、累加器宽度和非规格化数处理等未公开的实现细节,导致了不同硬件间数值结果无法复现。 研究人员对三种 AMD GPU 架构(CDNA 1、CDNA 2 和 CDNA 3)的矩阵核心数值行为进行了表征。通过设计特定的测试向量并采用迭代式的随机细化过程,研究人员为每种架构开发了高保真的 MATLAB 软件模型。这些模型通过 1000 万组随机输入数据进行了验证,确保了与硬件层面的位级精度一致。 该研究最终利用这些模型对比了 AMD 矩阵核心与 NVIDIA 张量核心的精度,展示了这些模型在实验研究和跨厂商数值分析中的应用价值。

抱歉。
相关文章

原文

View a PDF of the paper titled Accurate Models of AMD Matrix Cores, by Faizan A Khattak and Mantas Mikaitis and Carlo J. Graziani

View PDF HTML (experimental)
Abstract:Matrix multipliers available on recent GPUs do not conform with the IEEE 754 floating point standard. Features of matrix multipliers differ across vendors and architectures of the same vendor, such as accumulator width, rounding behaviour, normalisation points, intermediate underflow and overflow logic, the handling of subnormals, and the treatment of special inputs. As a result, reproducibility of small matrix multiplier results across devices is not possible and cannot be achieved by software control. Implementation details of matrix multipliers are not documented, making it difficult to interpret discrepancies in the computed results. We characterise the numerical behaviour of matrix multipliers across three AMD GPU architectures: CDNA 1, CDNA 2, and CDNA 3, using the MI100, MI210/250, and MI300A/300X GPUs, respectively. We design test vectors to target numerical features for all supported input formats and provide the derivation and the reasoning for why each vector allows to determine a particular numerical feature based on the outputs of the devices. MATLAB-based software models of the matrix multipliers are then developed for each architecture and validated for bit-level reproducibility against hardware using a randomized test suite consisting of 10 million sets of random input vectors. To achieve this, we applied a previously developed technique to iteratively refine the accuracy of the models in a loop, by randomized testing followed by test-refinement until the model matches the hardware for every test case. Finally, as a proof of concept for what experimental research can be done with the models, we have utilised them in two demonstrative numerical applications, quantifying application-level accuracy differences between AMD matrix cores and the NVIDIA tensor cores.
From: Faizan Ahmad Khattak [view email]
[v1] Sun, 13 Sep 2026 23:34:01 UTC (1,297 KB)
[v2] Tue, 15 Sep 2026 10:33:08 UTC (1,297 KB)
联系我们 contact @ memedata.com