Prism: Unleashing GPU Sharing for Cost-Efficient Multi-LLM Serving

• URL: http://arxiv.org/abs/2505.04021v2
• Date: Mon, 12 May 2025 18:19:46 GMT
• Title: Prism: Unleashing GPU Sharing for Cost-Efficient Multi-LLM Serving
• Authors: Shan Yu, Jiarong Xing, Yifan Qiao, Mingyuan Ma, Yangmin Li, Yang Wang, Shuo Yang, Zhiqiang Xie, Shiyi Cao, Ke Bao, Ion Stoica, Harry Xu, Ying Sheng,
• Abstract summary: Serving large workload (LLMs) is expensive, especially for providers hosting language models.<n>The long-tail popularity of models and their long idle periods through sharing creates new opportunities and challenges for this task.<n>This paper presents Prism, a system that unleashes the potential of GPU sharing to achieve both efficiency and SLO attainment.
• Score: 28.19296680865433
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Serving large language models (LLMs) is expensive, especially for providers hosting many models, making cost reduction essential. The unique workload patterns of serving multiple LLMs (i.e., multi-LLM serving) create new opportunities and challenges for this task. The long-tail popularity of models and their long idle periods present opportunities to improve utilization through GPU sharing. However, existing GPU sharing systems lack the ability to adjust their resource allocation and sharing policies at runtime, making them ineffective at meeting latency service-level objectives (SLOs) under rapidly fluctuating workloads. This paper presents Prism, a multi-LLM serving system that unleashes the full potential of GPU sharing to achieve both cost efficiency and SLO attainment. At its core, Prism tackles a key limitation of existing systems$\unicode{x2014}$the lack of $\textit{cross-model memory coordination}$, which is essential for flexibly sharing GPU memory across models under dynamic workloads. Prism achieves this with two key designs. First, it supports on-demand memory allocation by dynamically mapping physical to virtual memory pages, allowing flexible memory redistribution among models that space- and time-share a GPU. Second, it improves memory efficiency through a two-level scheduling policy that dynamically adjusts sharing strategies based on models' runtime demands. Evaluations on real-world traces show that Prism achieves more than $2\times$ cost savings and $3.3\times$ SLO attainment compared to state-of-the-art systems.

Related papers

• Collaborative LLM Inference via Planning for Efficient Reasoning [50.04696654679751]
  We propose a test-time collaboration framework in which a planner model first generates a plan, defined as a distilled and high-level abstraction of the problem.<n>Small and large models take turns acting as planner and reasoner, exchanging plans in a multi-round cascade to collaboratively solve complex tasks.<n>Our method achieves accuracy comparable to strong proprietary models alone, while significantly reducing reliance on paid inference.
  arXiv  Detail & Related papers  (2025-06-13T08:35:50Z)
• Minute-Long Videos with Dual Parallelisms [57.22737565366549]
  Diffusion Transformer (DiT)-based video diffusion models generate high-quality videos at scale but incur prohibitive processing latency and memory costs for long videos.<n>We propose a novel distributed inference strategy, termed DualParal.<n>Instead of generating an entire video on a single GPU, we parallelize both temporal frames and model layers across GPUs.
  arXiv  Detail & Related papers  (2025-05-27T11:55:22Z)
• 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)
• Mélange: Cost Efficient Large Language Model Serving by Exploiting GPU Heterogeneity [27.87327662815485]
  Large language models (LLMs) are increasingly integrated into many online services, yet they remain cost-prohibitive to deploy. We introduce M'elange, a framework that automatically and efficiently derive the minimal-cost GPU allocation for a given LLM service. M'elange reduces deployment costs by up to 77% in conversational settings, 33% in document-based settings, and 51% in a mixed setting.
  arXiv  Detail & Related papers  (2024-04-22T18:56:18Z)
• 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)
• SpotServe: Serving Generative Large Language Models on Preemptible Instances [64.18638174004151]
  SpotServe is the first distributed large language models serving system on preemptible instances. We show that SpotServe can reduce the P99 tail latency by 2.4 - 9.1x compared with the best existing LLM serving systems. We also show that SpotServe can leverage the price advantage of preemptive instances, saving 54% monetary cost compared with only using on-demand instances.
  arXiv  Detail & Related papers  (2023-11-27T06:31:17Z)
• FusionAI: Decentralized Training and Deploying LLMs with Massive Consumer-Level GPUs [57.12856172329322]
  We envision a decentralized system unlocking the potential vast untapped consumer-level GPU. This system faces critical challenges, including limited CPU and GPU memory, low network bandwidth, the variability of peer and device heterogeneity.
  arXiv  Detail & Related papers  (2023-09-03T13:27:56Z)
• AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration [54.692405042065815]
  We propose Activation-aware Weight Quantization (AWQ), a hardware-friendly approach for LLM low-bit weight-only quantization. AWQ protects only 1% salient weights and achieves excellent quantization performance for instruction-tuned LMs and, for the first time, multi-modal LMs. We also implement TinyChat, an efficient and flexible inference framework tailored for 4-bit on-device LLM/VLMs.
  arXiv  Detail & Related papers  (2023-06-01T17:59:10Z)
• Edge-MoE: Memory-Efficient Multi-Task Vision Transformer Architecture with Task-level Sparsity via Mixture-of-Experts [60.1586169973792]
  M$3$ViT is the latest multi-task ViT model that introduces mixture-of-experts (MoE) MoE achieves better accuracy and over 80% reduction computation but leaves challenges for efficient deployment on FPGA. Our work, dubbed Edge-MoE, solves the challenges to introduce the first end-to-end FPGA accelerator for multi-task ViT with a collection of architectural innovations.
  arXiv  Detail & Related papers  (2023-05-30T02:24:03Z)
• Hydra: A System for Large Multi-Model Deep Learning [3.571623412954477]
  We present'model spilling', a technique aimed at models such as Transformers and CNNs to move groups of layers between DRAM and GPU memory. We then present a set of novel techniques leveraging spilling to raise efficiency for multi-model training workloads. Experiments with real benchmark workloads show that HYDRA is over 7x faster than regular model parallelism and over 50% faster than state-of-the-art industrial tools for pipeline parallelism.
  arXiv  Detail & Related papers  (2021-10-16T18:13:57Z)
• Multi-model Machine Learning Inference Serving with GPU Spatial Partitioning [7.05946599544139]
  High throughput machine learning (ML) inference servers are critical for online service applications. These servers must provide a bounded latency for each request to support consistent service-level objective (SLO) This paper proposes a new ML inference scheduling framework for multi-model ML inference servers.
  arXiv  Detail & Related papers  (2021-09-01T04:46:46Z)

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.