The performance gains obtained by large language models (LLMs) are closely linked to their substantial computational and memory requirements. Quantized LLMs offer significant advantages with extremely quantized models, motivating the development of specialized architectures to accelerate their workloads. This paper proposes D-Legion, a novel scalable many-core architecture, designed using many adaptive-precision systolic array cores, to accelerate matrix multiplication in quantized LLMs. The proposed architecture consists of a set of Legions where each Legion has a group of adaptive-precision systolic arrays. D-Legion supports multiple computation modes, including quantized sparse and dense matrix multiplications. The block structured sparsity is exploited within a fully-sparse, or partially-sparse windows. In addition, memory accesses of partial summations (psums) are spatially reduced through parallel accumulators. Furthermore, data reuse is maximized through optimized scheduling techniques by multicasting matrix tiles across the Legions. A comprehensive design space exploration is performed in terms of Legion/core granularity to determine the optimal Legion configuration. Moreover, D-Legion is evaluated on attention workloads from two BitNet models, delivering up to 8.2$\times$ lower latency, up to 3.8$\times$ higher memory savings, and up to 3$\times$ higher psum memory savings compared to state-of-the-art work. D-Legion, with eight Legions and 64 total cores, achieves a peak throughput of 135.68 TOPS at a frequency of 1 GHz. A scaled version of D-Legion, with 32 Legions, is compared to Google TPUv4i, achieving up to 2.5$\times$ lower total latency, up to 2.3$\times$ higher total throughput, and up to 2.7$\times$ higher total memory savings.
Large-scale floating-point matrix multiplication is a fundamental kernel in many scientific and engineering applications. Most existing work only focus on accelerating matrix multiplication on FPGA by adopting a linear systolic array. This paper towards the extension of this architecture by proposing a scalable and highly configurable multi-array architecture. In addition, we propose a work-stealing scheme to ensure the equality in the workload partition among multiple linear arrays. Furthermore, an analytical model is developed to determine the optimal design parameters. Experiments on a real…
The emerging memristive Memory Processing Unit (mMPU) overcomes the memory wall through memristive devices that unite storage and logic for real processing-in-memory (PIM) systems. At the core of the mMPU is stateful logic, which is accelerated with memristive partitions to enable logic with massive inherent parallelism within crossbar arrays. This paper vastly accelerates the fundamental operations of matrix-vector multiplication and convolution in the mMPU, with either full-precision or binary elements. These proposed algorithms establish an efficient foundation for large-scale mMPU applicat…
General sparse matrix-matrix multiplication (SpGEMM) is an integral part of many scientific computing, high-performance computing (HPC), and graph analytic applications. This paper presents a new compressed sparse vector (CSV) format for representing sparse matrices and FSpGEMM, an OpenCL-based HPC framework for accelerating general sparse matrix-matrix multiplication on FPGAs. The proposed FSpGEMM framework includes an FPGA kernel implementing a throughput-optimized hardware architecture based on Gustavson's algorithm and a host program implementing pre-processing functions for converting inp…
Matrix-multiplication units (MXUs) are now prevalent in every computing platform. The key attribute that makes MXUs so successful is the semiring structure, which allows tiling for both parallelism and data reuse. Nonetheless, matrix-multiplication is not the only algorithm with such attributes. We find that many algorithms share the same structure and differ in only the core operation; for example, using add-minimum instead of multiply-add. Algorithms with a semiring-like structure therefore have potential to be accelerated by a general-purpose matrix operation architecture, instead of common…
Accelerators for sparse matrix multiplication are important components in emerging systems. In this paper, we study the main challenges of accelerating Sparse Matrix Multiplication (SpMM). For the situations that data is not stored in the desired order (row/column order), we propose a compact high performance sparse format, which allows for random access to a dataset with low memory access overhead. We show that using this format results in a 14-49 times speedup for SpMM. Next, we propose a high performance systolic architecture for SpMM, which uses a mesh of comparators to locate the useful (…
Homomorphic Encryption (HE) enables users to securely outsource both the storage and computation of sensitive data to untrusted servers. Not only does HE offer an attractive solution for security in cloud systems, but lattice-based HE systems are also believed to be resistant to attacks by quantum computers. However, current HE implementations suffer from prohibitively high latency. For lattice-based HE to become viable for real-world systems, it is necessary for the key bottlenecks - particularly polynomial multiplication - to be highly efficient. In this paper, we present a characterization …
School of Electrical and Computer Engineering, Purdue University（普渡大学电气与计算机工程学院）
Deep Neural Networks (DNNs) have transformed the field of machine learning and are widely deployed in many applications involving image, video, speech and natural language processing. The increasing compute demands of DNNs have been widely addressed through Graphics Processing Units (GPUs) and specialized accelerators. However, as model sizes grow, these von Neumann architectures require very high memory bandwidth to keep the processing elements utilized as a majority of the data resides in the main memory. Processing in memory has been proposed as a promising solution for the memory wall bott…
FusionCIM: Accelerating LLM Inference with Fusion-Driven Computing-in-Memory Architecture
AI Chip Center for Emerging Smart Systems (ACCESS), InnoHK（新兴智能系统人工智能芯片中心（ACCESS），创新香港）；The Hong Kong University of Science and Technology（香港科技大学）；Huazhong University of Science and Technology（华中科技大学）
DL inference queries play an important role in diverse internet services and a large fraction of datacenter cycles are spent on processing DL inference queries. Specifically, the matrix-matrix multiplication (GEMM) operations of fully-connected MLP layers dominate many inference tasks. We find that the GEMM operations for datacenter DL inference tasks are memory bandwidth bound, contrary to common assumptions: (1) strict query latency constraints force small-batch operation, which limits reuse and increases bandwidth demands; and (2) large and colocated models require reading the large weight …
This paper makes a case for accelerating lattice-based post quantum cryptography (PQC) with memristor based crossbars, and shows that these inherently error-tolerant algorithms are a good fit for noisy analog MAC operations in crossbars. We compare different NIST round-3 lattice-based candidates for PQC, and identify that SABER is not only a front-runner when executing on traditional systems, but it is also amenable to acceleration with crossbars. SABER is a module-LWR based approach, which performs modular polynomial multiplications with rounding. We map the polynomial multiplications in SABE…
GFormer: Accelerating Large Language Models with Optimized Transformers on Gaudi Processors
University of Houston（休斯顿大学）；Indiana University（印第安纳大学）；Stevens Institute of Technology（史蒂文斯理工学院）；Argonne National Laboratory（阿贡国家实验室）；Binghamton University（宾汉姆顿大学）
This paper presents GRAPHR, the first ReRAM-based graph processing accelerator. GRAPHR follows the principle of near-data processing and explores the opportunity of performing massive parallel analog operations with low hardware and energy cost. The analog computation is suit- able for graph processing because: 1) The algorithms are iterative and could inherently tolerate the imprecision; 2) Both probability calculation (e.g., PageRank and Collaborative Filtering) and typical graph algorithms involving integers (e.g., BFS/SSSP) are resilient to errors. The key insight of GRAPHR is that if a ve…
Neural networks (NNs) are growing in importance and complexity. A neural network's performance (and energy efficiency) can be bound either by computation or memory resources. The processing-in-memory (PIM) paradigm, where computation is placed near or within memory arrays, is a viable solution to accelerate memory-bound NNs. However, PIM architectures vary in form, where different PIM approaches lead to different trade-offs. Our goal is to analyze, discuss, and contrast DRAM-based PIM architectures for NN performance and energy efficiency. To do so, we analyze three state-of-the-art PIM archit…
The second-order training methods can converge much faster than first-order optimizers in DNN training. This is because the second-order training utilizes the inversion of the second-order information (SOI) matrix to find a more accurate descent direction and step size. However, the huge SOI matrices bring significant computational and memory overheads in the traditional architectures like GPU and CPU. On the other side, the ReRAM-based process-in-memory (PIM) technology is suitable for the second-order training because of the following three reasons: First, PIM's computation happens in memory…
