Mind the Memory Gap: Unveiling GPU Bottlenecks in Large-Batch LLM Inference

• URL: http://arxiv.org/abs/2503.08311v1
• Date: Tue, 11 Mar 2025 11:21:35 GMT
• Title: Mind the Memory Gap: Unveiling GPU Bottlenecks in Large-Batch LLM Inference
• Authors: Pol G. Recasens, Ferran Agullo, Yue Zhu, Chen Wang, Eun Kyung Lee, Olivier Tardieu, Jordi Torres, Josep Ll. Berral,
• Abstract summary: Large language models have been widely adopted across different tasks, but their auto-regressive generation nature often leads to inefficient resource utilization during inference.<n>In this paper, through an in-depth GPU-level analysis, we reveal that large-batch inference remains memory-bound, with most GPU compute capabilities underutilized due to DRAM bandwidth saturation as the primary bottleneck.
• Score: 4.497936996651617
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Large language models have been widely adopted across different tasks, but their auto-regressive generation nature often leads to inefficient resource utilization during inference. While batching is commonly used to increase throughput, performance gains plateau beyond a certain batch size, especially with smaller models, a phenomenon that existing literature typically explains as a shift to the compute-bound regime. In this paper, through an in-depth GPU-level analysis, we reveal that large-batch inference remains memory-bound, with most GPU compute capabilities underutilized due to DRAM bandwidth saturation as the primary bottleneck. To address this, we propose a Batching Configuration Advisor (BCA) that optimizes memory allocation, reducing GPU memory requirements with minimal impact on throughput. The freed memory and underutilized GPU compute capabilities can then be leveraged by concurrent workloads. Specifically, we use model replication to improve serving throughput and GPU utilization. Our findings challenge conventional assumptions about LLM inference, offering new insights and practical strategies for improving resource utilization, particularly for smaller language models.

Related papers

• HeadInfer: Memory-Efficient LLM Inference by Head-wise Offloading [79.38548165722229]
  HEADINFER offloads the KV cache to CPU RAM while avoiding the need to fully store the KV cache for any transformer layer on the GPU.<n>We demonstrate HEADINFER maintains computational efficiency while significantly reducing memory footprint.
  arXiv  Detail & Related papers  (2025-02-18T06:26:05Z)
• Cost-Efficient Continual Learning with Sufficient Exemplar Memory [55.77835198580209]
  Continual learning (CL) research typically assumes highly constrained exemplar memory resources.<n>In this work, we investigate CL in a novel setting where exemplar memory is ample.<n>Our method achieves state-of-the-art performance while reducing the computational cost to a quarter or third of existing methods.
  arXiv  Detail & Related papers  (2025-02-11T05:40:52Z)
• DAOP: Data-Aware Offloading and Predictive Pre-Calculation for Efficient MoE Inference [14.676716521856813]
  Mixture-of-Experts (MoE) models face significant deployment challenges on memory-constrained devices. We presentP, an on-device MoE inference engine to optimize parallel GPU- CPU execution. P outperforms traditional expert caching and prefetching methods by up to 8.20x and offloading techniques by 1.35x while maintaining accuracy.
  arXiv  Detail & Related papers  (2024-12-16T07:59:21Z)
• Memory-Efficient Training for Deep Speaker Embedding Learning in Speaker Verification [50.596077598766975]
  We explore a memory-efficient training strategy for deep speaker embedding learning in resource-constrained scenarios.<n>For activations, we design two types of reversible neural networks which eliminate the need to store intermediate activations.<n>For states, we introduce a dynamic quantization approach that replaces the original 32-bit floating-point values with a dynamic tree-based 8-bit data type.
  arXiv  Detail & Related papers  (2024-12-02T06:57:46Z)
• MoE-Lightning: High-Throughput MoE Inference on Memory-constrained GPUs [55.95879347182669]
  MoE architecture is renowned for its ability to increase model capacity without a proportional increase in inference cost. MoE-Lightning introduces a novel CPU-GPU-I/O pipelining schedule, CGOPipe, with paged weights to achieve high resource utilization. MoE-Lightning can achieve up to 10.3x higher throughput than state-of-the-art offloading-enabled LLM inference systems for Mixtral 8x7B on a single T4 GPU (16GB)
  arXiv  Detail & Related papers  (2024-11-18T01:06:12Z)
• Deep Optimizer States: Towards Scalable Training of Transformer Models Using Interleaved Offloading [2.8231000588510757]
  Transformers and large language models(LLMs) have seen rapid adoption in all domains. Training of transformers is very expensive and often hits a memory wall'' We propose a novel technique to split the LLM into subgroups, whose update phase is scheduled on either the CPU or the GPU.
  arXiv  Detail & Related papers  (2024-10-26T00:43:59Z)
• MEFT: Memory-Efficient Fine-Tuning through Sparse Adapter [40.616849959987555]
  We introduce a novel mechanism that fine-tunes Large Language Models (LLMs) with adapters of larger size yet memory-efficient. This is achieved by leveraging the inherent activation sparsity in the Feed-Forward Networks (FFNs) of LLMs. We employ a Mixture of Experts (MoE)-like architecture to mitigate unnecessary CPU computations and reduce the communication volume between the GPU and CPU.
  arXiv  Detail & Related papers  (2024-06-07T14:49:22Z)
• CORM: Cache Optimization with Recent Message for Large Language Model Inference [57.109354287786154]
  We introduce an innovative method for optimizing the KV cache, which considerably minimizes its memory footprint. CORM, a KV cache eviction policy, dynamically retains essential key-value pairs for inference without the need for model fine-tuning. Our validation shows that CORM reduces the inference memory usage of KV cache by up to 70% with negligible performance degradation across six tasks in LongBench.
  arXiv  Detail & Related papers  (2024-04-24T16:11:54Z)
• LLMem: Estimating GPU Memory Usage for Fine-Tuning Pre-Trained LLMs [4.536118764799076]
  Fine-tuning pre-trained large language models with limited hardware presents challenges due to GPU memory constraints. We introduce LLMem, a solution that estimates the GPU memory consumption when applying distributed fine-tuning methods. We show that LLMem accurately estimates peak GPU memory usage on a single GPU, with error rates of up to 1.6%.
  arXiv  Detail & Related papers  (2024-04-16T22:11:35Z)
• JORA: JAX Tensor-Parallel LoRA Library for Retrieval Augmented Fine-Tuning [16.86356520836045]
  We introduce a novel framework for PEFT-compatible fine-tuning of Llama-2 models, leveraging distributed training. Our framework uniquely utilizes JAX's just-in-time (JIT) compilation and tensor-sharding for efficient resource management. Our experiments show more than 12x improvement in runtime compared to Hugging Face/DeepSpeed implementation with four GPU while consuming less than half the VRAM per GPU.
  arXiv  Detail & Related papers  (2024-03-17T23:02:04Z)
• SqueezeLLM: Dense-and-Sparse Quantization [80.32162537942138]
  Main bottleneck for generative inference with LLMs is memory bandwidth, rather than compute, for single batch inference. We introduce SqueezeLLM, a post-training quantization framework that enables lossless compression to ultra-low precisions of up to 3-bit. Our framework incorporates two novel ideas: (i) sensitivity-based non-uniform quantization, which searches for the optimal bit precision assignment based on second-order information; and (ii) the Dense-and-Sparse decomposition that stores outliers and sensitive weight values in an efficient sparse format.
  arXiv  Detail & Related papers  (2023-06-13T08:57:54Z)
• FlexGen: High-Throughput Generative Inference of Large Language Models with a Single GPU [89.2451963569343]
  FlexGen is a generation engine for running large language model (LLM) inference on a single commodity GPU. When running OPT-175B on a single 16GB GPU, FlexGen achieves significantly higher throughput compared to state-of-the-art offloading systems. On the HELM benchmark, FlexGen can benchmark a 30B model with a 16GB GPU on 7 representative sub-scenarios in 21 hours.
  arXiv  Detail & Related papers  (2023-03-13T05:19:28Z)
• Heterogeneous CPU+GPU Stochastic Gradient Descent Algorithms [1.3249453757295084]
  We study training algorithms for deep learning on heterogeneous CPU+GPU architectures. Our two-fold objective -- maximize convergence rate and resource utilization simultaneously -- makes the problem challenging. We show that the implementation of these algorithms achieves both faster convergence and higher resource utilization than on several real datasets.
  arXiv  Detail & Related papers  (2020-04-19T05:21:20Z)

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.