Uni-STC: Unified Sparse Tensor Core

Modern processors are increasingly adopting tensor cores as key computational units. Compared to existing designs for dense and structured sparsity, recent dual-side sparse tensor cores have evolved to support general sparsity. However, existing methods still face limitations on generality (incomplete sparse kernel support prevents broad applicability) and performance (outer-product/row-row schemes yield unsatisfactory hardware utilisation, data reuse, and energy efficiency). In this paper, we propose Uni-STC, a unified sparse tensor core that delivers high-performance dataflows for four key sparse kernels: sparse matrix-vector multiplication (SpMV), sparse matrixsparse vector multiplication (SpMSpV), sparse matrix-multiple vector multiplication (SpMM), and sparse general matrix-matrix multiplication (SpGEMM). To efficiently support these diverse sparse workloads, we first introduce BBC, a unified sparse format co-designed with Uni-STC's dataflow. We then design UniSTC's architecture supporting (1) fine-grained task partitioning to improve resource utilisation, (2) parallel sparse-tile processing to enhance data reuse, and (3) a dynamic network to reduce intermediate data movement and energy consumption. Evaluated across 2893 SuiteSparse and 302 DLMC matrices, Uni-STC demonstrates significant improvements, outperforming the state-of-the-art RM-STC with a $2.21 \times$ geomean speedup and $2.96 \times$ higher energy efficiency.

• Publication year

  2026

• Venue

  International Symposium on High-Performance Computer Architecture

• Publication date

  2026-01-31

• Fields of study

  Not labeled

• Identifiers
  DOI 10.1109/HPCA68181.2026.11408500
• External record

  Open on Semantic Scholar

• Source metadata

  Semantic Scholar

• No claims are published for this paper.

• No concepts are published for this paper.

• DSTC: Dual-Side Sparse Tensor Core for DNNs Acceleration on Modern GPU Architectures

  2025influential reference
• KAMI: Communication-Avoiding General Matrix Multiplication within a Single GPU

  2025cited by this paper
• BerryBees: Breadth First Search by Bit-Tensor-Cores

  2025cited by this paper
• TB-STC: Transposable Block-wise N:M Structured Sparse Tensor Core

  2025cited by this paper
• SpInfer: Leveraging Low-Level Sparsity for Efficient Large Language Model Inference on GPUs

  2025cited by this paper
• Avalanche: Optimizing Cache Utilization via Matrix Reordering for Sparse Matrix Multiplication Accelerator

  2025cited by this paper
• SOFA: A Compute-Memory Optimized Sparsity Accelerator via Cross-Stage Coordinated Tiling

  2024cited by this paper
• Cuper: Customized Dataflow and Perceptual Decoding for Sparse Matrix-Vector Multiplication on HBM-Equipped FPGAs

  2024cited by this paper
• FEATHER: A Reconfigurable Accelerator with Data Reordering Support for Low-Cost On-Chip Dataflow Switching

  2024cited by this paper
• FEASTA: A Flexible and Efficient Accelerator for Sparse Tensor Algebra in Machine Learning

  2024cited by this paper
• Leda: Leveraging Tiling Dataflow to Accelerate SpMM on HBM-Equipped FPGAs for GNNs

  2024cited by this paper
• AmgT: Algebraic Multigrid Solver on Tensor Cores

  2024influential reference
• ReCG: ReRAM-Accelerated Sparse Conjugate Gradient

  2024cited by this paper
• Trapezoid: A Versatile Accelerator for Dense and Sparse Matrix Multiplications

  2024influential reference
• pSyncPIM: Partially Synchronous Execution of Sparse Matrix Operations for All-Bank PIM Architectures

  2024cited by this paper
• ACES: Accelerating Sparse Matrix Multiplication with Adaptive Execution Flow and Concurrency-Aware Cache Optimizations

  2024cited by this paper
• NeuraChip: Accelerating GNN Computations with a Hash-based Decoupled Spatial Accelerator

  2024cited by this paper
• ASADI: Accelerating Sparse Attention Using Diagonal-based In-Situ Computing

  2024cited by this paper
• PruneGNN: Algorithm-Architecture Pruning Framework for Graph Neural Network Acceleration

  2024cited by this paper
• HotTiles: Accelerating SpMM with Heterogeneous Accelerator Architectures

  2024cited by this paper
• DPACS: Hardware Accelerated Dynamic Neural Network Pruning through Algorithm-Architecture Co-design

  2023cited by this paper
• RM-STC: Row-Merge Dataflow Inspired GPU Sparse Tensor Core for Energy-Efficient Sparse Acceleration

  2023influential reference
• A Tensor Marshaling Unit for Sparse Tensor Algebra on General-Purpose Processors

  2023cited by this paper
• Eureka: Efficient Tensor Cores for One-sided Unstructured Sparsity in DNN Inference

  2023cited by this paper
• HARP: Hardware-Based Pseudo-Tiling for Sparse Matrix Multiplication Accelerator

  2023cited by this paper
• HASpMV: Heterogeneity-Aware Sparse Matrix-Vector Multiplication on Modern Asymmetric Multicore Processors

  2023cited by this paper
• DASP: Specific Dense Matrix Multiply-Accumulate Units Accelerated General Sparse Matrix-Vector Multiplication

  2023cited by this paper
• Sparse-DySta: Sparsity-Aware Dynamic and Static Scheduling for Sparse Multi-DNN Workloads

  2023cited by this paper
• Tailors: Accelerating Sparse Tensor Algebra by Overbooking Buffer Capacity

  2023cited by this paper
• AmgR: Algebraic Multigrid Accelerated on ReRAM

  2023cited by this paper
• HASpGEMM: Heterogeneity-Aware Sparse General Matrix-Matrix Multiplication on Modern Asymmetric Multicore Processors

  2023cited by this paper
• Balancing Computation and Communication in Distributed Sparse Matrix-Vector Multiplication

  2023cited by this paper
• TaskFusion: An Efficient Transfer Learning Architecture with Dual Delta Sparsity for Multi-Task Natural Language Processing

  2023cited by this paper
• SPADE: A Flexible and Scalable Accelerator for SpMM and SDDMM

  2023cited by this paper
• HIRAC: A Hierarchical Accelerator with Sorting-based Packing for SpGEMMs in DNN Applications

  2023cited by this paper
• Accelerating Sparse Data Orchestration via Dynamic Reflexive Tiling

  2023cited by this paper
• SGCN: Exploiting Compressed-Sparse Features in Deep Graph Convolutional Network Accelerators

  2023cited by this paper
• VEGETA: Vertically-Integrated Extensions for Sparse/Dense GEMM Tile Acceleration on CPUs

  2023cited by this paper
• Flexagon: A Multi-dataflow Sparse-Sparse Matrix Multiplication Accelerator for Efficient DNN Processing

  2023cited by this paper
• HuffDuff: Stealing Pruned DNNs from Sparse Accelerators

  2023cited by this paper
• Spada: Accelerating Sparse Matrix Multiplication with Adaptive Dataflow

  2023cited by this paper
• Cascading structured pruning: enabling high data reuse for sparse DNN accelerators

  2022cited by this paper
• A Pattern-Based SpGEMM Library for Multi-Core and Many-Core Architectures

  2022cited by this paper
• GROW: A Row-Stationary Sparse-Dense GEMM Accelerator for Memory-Efficient Graph Convolutional Neural Networks

  2022cited by this paper
• CANDLES: Channel-Aware Novel Dataflow-Microarchitecture Co-Design for Low Energy Sparse Neural Network Acceleration

  2022cited by this paper
• TileSpGEMM: a tiled algorithm for parallel sparse general matrix-matrix multiplication on GPUs

  2022cited by this paper
• Sparseloop: An Analytical Approach To Sparse Tensor Accelerator Modeling

  2022cited by this paper
• Anticipating and eliminating redundant computations in accelerated sparse training

  2022cited by this paper
• AMD Instinct™ MI200 Series Accelerator and Node Architectures

  2022cited by this paper
• TileSpMSpV: A Tiled Algorithm for Sparse Matrix-Sparse Vector Multiplication on GPUs

  2022cited by this paper
• Gamma: leveraging Gustavson’s algorithm to accelerate sparse matrix multiplication

  2021cited by this paper
• Serpens: A High Bandwidth Memory Based Accelerator for General-Purpose Sparse Matrix-Vector Multiplication

  2021cited by this paper
• Griffin: Rethinking Sparse Optimization for Deep Learning Architectures

  2021cited by this paper
• Sapphire Rapids

  2021cited by this paper
• S2TA: Exploiting Structured Sparsity for Energy-Efficient Mobile CNN Acceleration

  2021cited by this paper
• TileSpMV: A Tiled Algorithm for Sparse Matrix-Vector Multiplication on GPUs

  2021cited by this paper
• Ten Lessons From Three Generations Shaped Google’s TPUv4i : Industrial Product

  2021cited by this paper
• GoSPA: An Energy-efficient High-performance Globally Optimized SParse Convolutional Neural Network Accelerator

  2021cited by this paper
• Sparsity-Aware and Re-configurable NPU Architecture for Samsung Flagship Mobile SoC

  2021cited by this paper
• Dual-side Sparse Tensor Core

  2021influential reference
• VIA: A Smart Scratchpad for Vector Units with Application to Sparse Matrix Computations

  2021cited by this paper
• SpaceA: Sparse Matrix Vector Multiplication on Processing-in-Memory Accelerator

  2021cited by this paper
• SPAGHETTI: Streaming Accelerators for Highly Sparse GEMM on FPGAs

  2021cited by this paper
• TensorDash: Exploiting Sparsity to Accelerate Deep Neural Network Training

  2020cited by this paper
• SpAtten: Efficient Sparse Attention Architecture with Cascade Token and Head Pruning

  2020cited by this paper
• GaaS-X: Graph Analytics Accelerator Supporting Sparse Data Representation using Crossbar Architectures

  2020cited by this paper
• Commutative Data Reordering: A New Technique to Reduce Data Movement Energy on Sparse Inference Workloads

  2020cited by this paper
• SpArch: Efficient Architecture for Sparse Matrix Multiplication

  2020cited by this paper
• SIGMA: A Sparse and Irregular GEMM Accelerator with Flexible Interconnects for DNN Training

  2020influential reference
• IA-SpGEMM: an input-aware auto-tuning framework for parallel sparse matrix-matrix multiplication

  2019cited by this paper
• Register-Aware Optimizations for Parallel Sparse Matrix–Matrix Multiplication

  2019cited by this paper
• SMASH: Co-designing Software Compression and Hardware-Accelerated Indexing for Efficient Sparse Matrix Operations

  2019cited by this paper
• Sparse ReRAM Engine: Joint Exploration of Activation and Weight Sparsity in Compressed Neural Networks

  2019cited by this paper
• Sparse Tensor Core: Algorithm and Hardware Co-Design for Vector-wise Sparse Neural Networks on Modern GPUs

  2019cited by this paper
• SparTen: A Sparse Tensor Accelerator for Convolutional Neural Networks

  2019cited by this paper
• ExTensor: An Accelerator for Sparse Tensor Algebra

  2019cited by this paper
• OuterSPACE: An Outer Product Based Sparse Matrix Multiplication Accelerator

  2018cited by this paper
• Accel-Sim: An Extensible Simulation Framework for Validated GPU Modeling

  2018cited by this paper
• Attention is All you Need

  2017cited by this paper
• SCNN: An accelerator for compressed-sparse convolutional neural networks

  2017cited by this paper
• Scalpel: Customizing DNN pruning to the underlying hardware parallelism

  2017cited by this paper
• In-datacenter performance analysis of a tensor processing unit

  2017cited by this paper
• CACTI 7

  2017cited by this paper
• ASCEnD-FreePDK45: An open source standard cell library for asynchronous design

  2016cited by this paper
• EIE: Efficient Inference Engine on Compressed Deep Neural Network

  2016cited by this paper
• A framework for general sparse matrix-matrix multiplication on GPUs and heterogeneous processors

  2015cited by this paper
• Speculative segmented sum for sparse matrix-vector multiplication on heterogeneous processors

  2015cited by this paper
• CSR5: An Efficient Storage Format for Cross-Platform Sparse Matrix-Vector Multiplication

  2015cited by this paper
• An Efficient GPU General Sparse Matrix-Matrix Multiplication for Irregular Data

  2014cited by this paper
• The university of Florida sparse matrix collection

  2011cited by this paper
• Reduced-Bandwidth Multithreaded Algorithms for Sparse Matrix-Vector Multiplication

  2011cited by this paper

• No citing papers are available for this paper.