arXiv:2611.12435v1 Announce Type: cross
Abstract: Non-linear functions are an important class of operations in modern large language models. However, chip vendors building for AI have so far relied on traditional vector or scalar processing units for non-linear functions. The performance of these units has not kept pace with matrix multiplication units (MMUs), which can be found in both high-performance and edge-computing devices from almost every vendor. Reconfigurable units such as FPGAs and CGRAs have been proposed in the literature, but have not been systematically compared with traditional units backed by MMUs. In this paper, we build two SoCs: one using an RVV vector processor and the other using a latency-insensitive (\emph{elastic}) CGRA, both coupled with an MMU and driven by a single-core CPU. We then propose mappings for common feed-forward networks on our SoCs and find that the CGRA reaches a maximum speedup in clock cycles of 7.12x over RVV on the activation stage of simple point-wise non-linear functions like SiLU, whereas the RVV reaches a maximum speedup of \textbf{9.8x} over the CGRA on complex non-linear functions such as $\mathrm{softmax}$. Over the enclosing feed-forward block the corresponding maxima are 3.16x and 4.24x.
arxiv
RVV vs. CGRA SoCs for Non-Linear Functions in AI
This paper presents a systematic comparison of two SoC designs for AI workloads: one using an RVV vector processor and one using an elastic CGRA, both paired with a matrix multiplication unit. The authors propose mappings for feed-forward networks and report that the CGRA achieves up to 7.12x speedup over RVV for simple point-wise activations like SiLU, while RVV achieves up to 9.8x speedup over CGRA for complex functions like softmax. The results highlight that no single processing approach dominates across all non-linear operations in LLMs.
The article provides concrete, recent hardware research comparing two emerging AI accelerator approaches, with quantitative speedup results useful for chip architects and AI systems designers.
Comments & Questions (0)
No comments yet
Be the first to comment!