Architecture-Aware LLM Inference Optimization on AMD Instinct GPUs: A Comprehensive Benchmark and Deployment Study

• URL: http://arxiv.org/abs/2603.10031v1
• Date: Fri, 27 Feb 2026 13:21:48 GMT
• Title: Architecture-Aware LLM Inference Optimization on AMD Instinct GPUs: A Comprehensive Benchmark and Deployment Study
• Abstract summary: Cross-architecture evaluation of production LLM inference on AMD Instinct MI325X GPUs.<n> benchmarking four models spanning 235B to 1 trillion parameters across three architectural families.
• Score: 0.0
• License: http://creativecommons.org/licenses/by-nc-nd/4.0/
• Abstract: We present a cross-architecture evaluation of production LLM inference on AMD Instinct MI325X GPUs, benchmarking four models spanning 235B to 1 trillion parameters across three architectural families (MoE+MLA, Dense+GQA, MoE+GQA) on an 8-GPU cluster with 2TB aggregate HBM3e using vLLM v0.14.1. Our results demonstrate that architecture-aware optimization is essential: MLA models require block size 1 and cannot use KV cache offloading, while GQA models benefit from both. The AMD AITER runtime is required for competitive MLA inference throughput and must be selectively disabled for architectures with incompatible attention head configurations. A controlled AITER ablation on Llama-3.1-405B (n=5 per condition) reveals a modest 3-5% throughput benefit at high concurrency but 2-16x higher measurement variability, confirming that AITER's large speedups target MoE/MLA kernels specifically. Under text-only workloads, Llama-405B and DeepSeek V3.2 achieve comparable peak throughput (15,944 and 15,343 tok/s) despite an order-of-magnitude difference in active parameters. Under vision workloads, Qwen3-VL-235B reaches 47,873 tok/s, 6.5x higher than Kimi-K2.5 (7,327 tok/s). Active parameter count per token is associated with inference throughput, though confounded by differences in quantization, AITER acceleration, and tensor parallelism. All four models exhibit a common throughput saturation point consistent with a memory-bandwidth bottleneck (~500 concurrent for short sequences, ~100-200 for longer sequences). All models maintain 100% HTTP-level success rates through 1,000 concurrent users, processing 18.9 million tokens across 17,406 requests without failures.

Related papers

• A Modern Multimodal Assistant on a 6 GB 2011 GPU: Stage-Validated, All-GPU CUDA Inference for Fermi [0.0]
  This report keeps the hardware and asks what a model that fits can do.<n>We deploy Mini-V-4.6, a modern multimodal assistant pairing a SigLIP2 encoder and window-attention merger.<n>The system answers an image question end-to-end in 1.7s.
  arXiv  Detail & Related papers  (2026-07-16T04:48:44Z)
• Efficient, VRAM-Constrained xLM Inference on Clients [0.0]
  We present pipelined sharding, a novel, benchmark-profile-guided CPU-GPU hybrid scheduling technique.<n>It achieves efficient, VRAM-constrained inference for both dense and mixture-of-experts (MoE) large language models (LLMs) on client systems.<n>This paper is accepted at the 9th MLSys Conference (Industry Track), 2026.
  arXiv  Detail & Related papers  (2026-04-29T06:35:35Z)
• Density Field State Space Models: 1-Bit Distillation, Efficient Inference, and Knowledge Organization in Mamba-2 [0.0]
  We present Density Field State Space Models (DF-SSM), a framework for compressing SSMs to a 1-bit scaffold with int8 low-rank correction.<n>We achieve a 278 MB model (9.7x smaller than the 2.7 GB FP16 teacher) that runs at 21.4x faster inference on GPU.
  arXiv  Detail & Related papers  (2026-04-28T14:25:10Z)
• Scaling DoRA: High-Rank Adaptation via Factored Norms and Fused Kernels [83.99688944263843]
  Weight-De Low-Rank Adaptation (DoRA) extends LoRA by decoupling weight magnitude from direction.<n>At d_in = 8192 and rank r = 384, a single module's norm requires about 512 MB of transient working memory in bf16.<n>A factored norm decomposes the squared norm into base, cross, and Gram terms computable through O(d_out r + r2) intermediates, eliminating the dense product.
  arXiv  Detail & Related papers  (2026-03-23T17:57:24Z)
• Native LLM and MLLM Inference at Scale on Apple Silicon [0.8122270502556375]
  We present vllm-mlx, a framework for efficient LLM and MLLM inference on Apple Silicon built native on MLX.<n>For text models, we achieve 21% to 87% higher throughput than llama- across models ranging from Qwen3-0.6B to Nemotron-30B.<n>For multimodal models, we introduce content-based caching that eliminates redundant vision encoding by identifying identical images through content hashing, regardless of input format.
  arXiv  Detail & Related papers  (2026-01-27T03:11:02Z)
• Chronicals: A High-Performance Framework for LLM Fine-Tuning with 3.51x Speedup over Unsloth [0.0]
  We present Chronicals, an open-source training framework achieving 3.5x speedup over Unsloth.<n>We provide complete mathematical foundations: online softmax correctness, FlashAttention IO complexity O(N2 d2 M-1), LoRA+ learning rate gradient approximations.
  arXiv  Detail & Related papers  (2026-01-06T00:00:55Z)
• dInfer: An Efficient Inference Framework for Diffusion Language Models [54.80918957287927]
  Diffusion-based large language models (dLLMs) have emerged as a promising alternative to autoregressive (AR) LLMs.<n>We present dInfer, an efficient and framework for dLLM inference.
  arXiv  Detail & Related papers  (2025-10-09T16:19:42Z)
• Puzzle: Distillation-Based NAS for Inference-Optimized LLMs [17.72841008597783]
  Large language models (LLMs) offer remarkable capabilities, yet their high inference costs restrict wider adoption.<n>We present Puzzle, a hardware-aware framework that accelerates the inference of LLMs while preserving their capabilities.<n>We showcase our framework's impact via Llama-3.1-Nemotron-51B-Instruct (Nemotron-51B) and Llama-3.3-Nemotron-49B, two publicly available models.
  arXiv  Detail & Related papers  (2024-11-28T13:45:42Z)
• MARLIN: Mixed-Precision Auto-Regressive Parallel Inference on Large Language Models [58.3342517278868]
  This paper describes the design of Mixed-precision AutoRegressive LINear kernels. It shows that batchsizes up to 16-32 can be supported with close to maximum ($4times$) quantization speedup. MarLIN accomplishes this via a combination of techniques, such as asynchronous memory access, complex task scheduling and pipelining.
  arXiv  Detail & Related papers  (2024-08-21T16:10:41Z)
• vTensor: Flexible Virtual Tensor Management for Efficient LLM Serving [53.972175896814505]
  Large Language Models (LLMs) are widely used across various domains, processing millions of daily requests. Large Language Models (LLMs) are widely used across various domains, processing millions of daily requests.
  arXiv  Detail & Related papers  (2024-07-22T14:37:58Z)
• QMoE: Practical Sub-1-Bit Compression of Trillion-Parameter Models [64.34635279436054]
  Mixture-of-Experts (MoE) architectures offer a general solution to the high inference costs of large language models (LLMs) via sparse routing. We present a solution to this memory problem, in form of a new compression and execution framework called QMoE.
  arXiv  Detail & Related papers  (2023-10-25T17:24:53Z)
• QLoRA: Efficient Finetuning of Quantized LLMs [66.58009990713134]
  We present QLoRA, an efficient finetuning approach that reduces memory usage enough to finetune a 65B parameter model on a single 48GB GPU. QLoRA backpropagates through a frozen, 4-bit quantized pretrained language model into Low Rank Adapters(LoRA) Our best model family, which we name Guanaco, outperforms all previous openly released models on the Vicuna benchmark.
  arXiv  Detail & Related papers  (2023-05-23T17:50:33Z)
• AdaMTL: Adaptive Input-dependent Inference for Efficient Multi-Task Learning [1.4963011898406864]
  We introduce AdaMTL, an adaptive framework that learns task-aware inference policies for multi-task learning models. AdaMTL reduces the computational complexity by 43% while improving the accuracy by 1.32% compared to single-task models. When deployed on Vuzix M4000 smart glasses, AdaMTL reduces the inference latency and the energy consumption by up to 21.8% and 37.5%, respectively.
  arXiv  Detail & Related papers  (2023-04-17T20:17:44Z)

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.