Beyond GEMM-Centric NPUs: Enabling Efficient Diffusion LLM Sampling

• URL: http://arxiv.org/abs/2601.20706v1
• Date: Wed, 28 Jan 2026 15:37:50 GMT
• Title: Beyond GEMM-Centric NPUs: Enabling Efficient Diffusion LLM Sampling
• Authors: Binglei Lou, Haoran Wu, Yao Lai, Jiayi Nie, Can Xiao, Xuan Guo, Rika Antonova, Robert Mullins, Aaron Zhao,
• Abstract summary: Diffusion Large Language Models (dLLMs) introduce iterative denoising to enable parallel token generation.<n>Our design employs lightweight non-GEMM vector primitives, in-place memory reuse strategies, and a decoupled mixed-precision memory hierarchy.
• Score: 14.471123653746275
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Diffusion Large Language Models (dLLMs) introduce iterative denoising to enable parallel token generation, but their sampling phase displays fundamentally different characteristics compared to GEMM-centric transformer layers. Profiling on modern GPUs reveals that sampling can account for up to 70% of total model inference latency-primarily due to substantial memory loads and writes from vocabulary-wide logits, reduction-based token selection, and iterative masked updates. These processes demand large on-chip SRAM and involve irregular memory accesses that conventional NPUs struggle to handle efficiently. To address this, we identify a set of critical instructions that an NPU architecture must specifically optimize for dLLM sampling. Our design employs lightweight non-GEMM vector primitives, in-place memory reuse strategies, and a decoupled mixed-precision memory hierarchy. Together, these optimizations deliver up to a 2.53x speedup over the NVIDIA RTX A6000 GPU under an equivalent nm technology node. We also open-source our cycle-accurate simulation and post-synthesis RTL verification code, confirming functional equivalence with current dLLM PyTorch implementations.

Related papers

• LUT-LLM: Efficient Large Language Model Inference with Memory-based Computations on FPGAs [14.676146518251185]
  We present LUT-LLM, the first FPGA accelerator enabling 1B+ LLM inference via vector-quantized memory operations.<n>LUT-LLM achieves 1.66x lower latency than AMD MI210 and 1.72x higher energy efficiency than NVIDIA A100, scaling to 32B models with 2.16x efficiency gain over A100.
  arXiv  Detail & Related papers  (2025-11-09T01:17:08Z)
• Eliminating Multi-GPU Performance Taxes: A Systems Approach to Efficient Distributed LLMs [61.953548065938385]
  We introduce the ''Three Taxes'' (Bulk Synchronous, Inter- Kernel Data Locality, and Kernel Launch Overhead) as an analytical framework.<n>We propose moving beyond the rigid BSP model to address key inefficiencies in distributed GPU execution.<n>We observe a 10-20% speedup in end-to-end latency over BSP-based approaches.
  arXiv  Detail & Related papers  (2025-11-04T01:15:44Z)
• Optimizing LLM Inference: Fluid-Guided Online Scheduling with Memory Constraints [14.341123057506827]
  Large Language Models (LLMs) are indispensable in today's applications, but their inference procedure demands significant computational resources.<n>This paper formulates LLM inference optimization as a multi-stage online scheduling problem.<n>We develop a fluid dynamics approximation to provide a tractable benchmark that guides algorithm design.
  arXiv  Detail & Related papers  (2025-04-15T16:00:21Z)
• Towards On-Device Learning and Reconfigurable Hardware Implementation for Encoded Single-Photon Signal Processing [0.0]
  We propose an online training algorithm based on a One-Sided Jacobi rotation-based Online Sequential Extreme Learning Machine (OSOS-ELM)<n>We fully exploit parallelism in executing OSOS-ELM on a heterogeneous FPGA with integrated ARM cores.<n>We validate our approach through three case studies involving single-photon signal analysis.
  arXiv  Detail & Related papers  (2025-04-12T00:58:52Z)
• EPS-MoE: Expert Pipeline Scheduler for Cost-Efficient MoE Inference [49.94169109038806]
  This paper introduces EPS-MoE, a novel expert pipeline scheduler for MoE that surpasses the existing parallelism schemes.<n>Our results demonstrate at most 52.4% improvement in prefill throughput compared to existing parallel inference methods.
  arXiv  Detail & Related papers  (2024-10-16T05:17:49Z)
• Efficient Arbitrary Precision Acceleration for Large Language Models on GPU Tensor Cores [3.6385567224218556]
  Large language models (LLMs) have been widely applied but face challenges in efficient inference. We introduce a novel bipolar-INT data format that facilitates parallel computing and supports symmetric quantization. We implement an arbitrary precision matrix multiplication scheme that decomposes and recovers at the bit level, enabling flexible precision.
  arXiv  Detail & Related papers  (2024-09-26T14:17:58Z)
• UIO-LLMs: Unbiased Incremental Optimization for Long-Context LLMs [111.05657299071648]
  UIO-LLMs is an incremental optimization approach for memory-enhanced transformers under long-context settings.<n>We refine the training process using the Truncated Backpropagation Through Time (TBPTT) algorithm.<n>UIO-LLMs successfully handle long context, such as extending the context window of Llama2-7b-chat from 4K to 100K tokens with minimal 2% additional parameters.
  arXiv  Detail & Related papers  (2024-06-26T08:44:36Z)
• Tensor Slicing and Optimization for Multicore NPUs [2.670309629218727]
  This paper proposes a compiler optimization pass for Multicore NPUs, called Slicing Optimization (TSO) TSO identifies the best tensor slicing that minimizes execution time for a set of CNN models. Results show that TSO is capable of identifying the best tensor slicing that minimizes execution time for a set of CNN models.
  arXiv  Detail & Related papers  (2023-04-06T12:03:03Z)
• Energy-efficient Task Adaptation for NLP Edge Inference Leveraging Heterogeneous Memory Architectures [68.91874045918112]
  adapter-ALBERT is an efficient model optimization for maximal data reuse across different tasks. We demonstrate the advantage of mapping the model to a heterogeneous on-chip memory architecture by performing simulations on a validated NLP edge accelerator.
  arXiv  Detail & Related papers  (2023-03-25T14:40:59Z)
• NumS: Scalable Array Programming for the Cloud [82.827921577004]
  We present NumS, an array programming library which optimize NumPy-like expressions on task-based distributed systems. This is achieved through a novel scheduler called Load Simulated Hierarchical Scheduling (LSHS) We show that LSHS enhances performance on Ray by decreasing network load by a factor of 2x, requiring 4x less memory, and reducing execution time by 10x on the logistic regression problem.
  arXiv  Detail & Related papers  (2022-06-28T20:13:40Z)
• SreaMRAK a Streaming Multi-Resolution Adaptive Kernel Algorithm [60.61943386819384]
  Existing implementations of KRR require that all the data is stored in the main memory. We propose StreaMRAK - a streaming version of KRR. We present a showcase study on two synthetic problems and the prediction of the trajectory of a double pendulum.
  arXiv  Detail & Related papers  (2021-08-23T21:03:09Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.