TSR-GEMM: Tile-Selective Precision Recovery for Robust Mixed-Precision Matrix Multiplication on GPU Tensor Cores
International Scientific Journal of Engineering and Management · 20.04.2026 · 05 (04) · 1-8
DOI: 10.55041/isjem06493 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Scientific Journal of Engineering and Management, online erschienen am 20.04.2026. Verfasst von Abhinav Rai und 4 weiteren Autor:innen.
Abstract (Originalsprache)
Modern deep learning frameworks increasingly rely on mixed-precision general matrix multiplication (GEMM) to exploit the throughput advantages of half-precision (FP16) Tensor Cores on NVIDIA GPUs. While FP16 GEMM delivers substan-tial speedups over FP32 computation, it introduces numerical errors that are spatially non-uniform across the output matrix—concentrated in tiles whose input sub-blocks exhibit high condi-tion numbers or significant cancellation. Existing recovery mech-anisms, such as iterative refinement, operate at matrix-global granularity and therefore cannot exploit this spatial locality. We introduce TSR-GEMM (Tile-Selective Residual GEMM), a three-phase mixed-precision GEMM pipeline that (1) performs the bulk computation in FP16 using Tensor Cores while simultaneously accumulating per-tile norm statistics, (2) evaluates a lightweight instability score for each output tile based on input panel and output tile norms, and (3) selectively re-computes only flagged tiles in FP32 via cuBLAS. TSR-GEMM exposes a single tunable threshold τ that governs the precision–performance trade-off. On an NVIDIA RTX 3050 Ti GPU across matrix dimensions from 512×512 to 4096×4096, TSR-GEMM achieves FP32-comparable accuracy (5.4 × 10−8 relative error) at full recovery, while at 70% tile recovery it reduces error by 8× over pure FP16 with only a 12% throughput reduction relative to the no-recovery baseline. The τ sweep reveals a smooth, well-behaved Pareto frontier, confirming the instability score as a reliable predictor of per-tile numerical risk. Index Terms—mixed-precision arithmetic, GEMM, Tensor Cores, CUDA, numerical accuracy, tile-selective recovery, GPU computing
Zitieren
APA
Rai, A., L, D. P., Chauhan, V. S., Singh, A., & Shrivastava, V. (2026). TSR-GEMM: Tile-Selective Precision Recovery for Robust Mixed-Precision Matrix Multiplication on GPU Tensor Cores. International Scientific Journal of Engineering and Management, 05(04), 1-8. https://doi.org/10.55041/isjem06493
Vancouver
Rai A, L DP, Chauhan VS, Singh A, Shrivastava V. TSR-GEMM: Tile-Selective Precision Recovery for Robust Mixed-Precision Matrix Multiplication on GPU Tensor Cores. International Scientific Journal of Engineering and Management. 2026;05(04):1-8. doi:10.55041/isjem06493
BibTeX
@article{Rai2026,
author = {Rai, Abhinav and L, Dr. Pavithra and Chauhan, Vedant Singh and Singh, Ananya and Shrivastava, Vinayak},
title = {TSR-GEMM: Tile-Selective Precision Recovery for Robust Mixed-Precision Matrix Multiplication on GPU Tensor Cores},
journal = {International Scientific Journal of Engineering and Management},
year = {2026},
volume = {05},
number = {04},
pages = {1--8},
doi = {10.55041/isjem06493}
}
EUVIMED