FusionML: Prefill, Not Decode - Mechanism and Boundaries of CPU+GPU Co-Execution on Unified-Memory Apple Silicon

• URL: http://arxiv.org/abs/2607.22785v1
• Date: Fri, 24 Jul 2026 11:42:17 GMT
• Title: FusionML: Prefill, Not Decode - Mechanism and Boundaries of CPU+GPU Co-Execution on Unified-Memory Apple Silicon
• Abstract summary: Previous attempts, including our own, failed or produced precision-confounded wins.<n>We identify the cause: MLX's lazy-graph scheduler emphserializes cross-stream workWhenever a CPU-stream operation consumes an unmaterialized GPU result inside one evaluation graph, a row-split matmul runs x1.38 faster with materialized inputs.<n>We characterize the boundaries equally carefully: decode cannot benefit, bound by shared bandwidth co-execution does not add.
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Apple-Silicon SoCs share CPU, GPU, and Neural Engine over one unified memory system, raising the question of whether transformer inference can be accelerated by splitting single operators across units. Prior attempts, including our own, failed or produced precision-confounded wins. We identify the cause: MLX's lazy-graph scheduler \emph{serializes} cross-stream work whenever a CPU-stream operation consumes an unmaterialized GPU result inside one evaluation graph, so a row-split matmul that runs \x{1.38} faster with materialized inputs runs \x{0.66} slower than GPU-only inside a lazy graph; an eager materialization boundary restores concurrency (\x{1.34}). \sys{} implements a per-layer, contention-aware CPU+GPU row split for transformer prefill built on this fix. Evaluated across five chips and three Apple-Silicon generations, community-replicated, the split accelerates Llama-shaped decoder-block prefill by \x{1.15}--\x{1.38}, unchanged at full 32-block depth, and reaches \x{1.18}--\x{1.25} faster time-to-first-token on a real Qwen2.5-7B checkpoint served through stock MLX-LM, with token-identical outputs and unchanged decode throughput. We characterize the boundaries equally carefully: decode cannot benefit, bound by shared bandwidth co-execution does not add; precision-matched training loses \x{0.86}--\x{0.97} on all five chips; ANE dispatch overhead excludes it at layer granularity; and a no-regression runtime gate becomes self-defeating under memory pressure, where probing an alternative mode evicts the active mode's working set. Code, raw results, and generation transcripts are released.

Related papers

• VC-Attention: Value Smoothing and Softmax Casting for Low-bit Attention [47.30761996669105]
  We propose VC-Attention, a training-free low-bit attention framework.<n>It addresses both by smoothing value tokens by a probability Cast. V-Smooth reorders value tokens by lightweight clustering online, so the tokens in a fused hardware block quantize well together.<n>It maps log-domain scores directly to E4M3 probability codes with one fused multiply-add, eliminating the FP32 exponential and the format conversion.
  arXiv  Detail & Related papers  (2026-09-14T16:17:35Z)
• Pre-Compiled Pipeline Shards for Distributed LLM Inference on Intel AI PC Fleets [0.0]
  We show that a handful of AIPCs can serve models beyond the capability of any single one.<n>A four-node deployment of Lunar Lake AI PCs on Intel Tiber Cloud serves a single user at interactive speed.
  arXiv  Detail & Related papers  (2026-08-19T17:33:28Z)
• Ready Cohorts: Bounding GPU Opportunity and Avoiding Host Round Trips in LLM-Agent Control [0.0]
  We formalize the ready-cohort boundary using fixed-share F, exact offline share P*, local upper bound U, and online achieved share A.<n>A separate mechanism study keeps a GPU-computed binary decision on device instead of returning four bytes to the host and redispatching.
  arXiv  Detail & Related papers  (2026-08-12T14:42:15Z)
• Meganeura: Portable GPU Training and Inference through Vulkan and Metal [0.0]
  Meganeura asks whether one compact native compiler can span both phases on consumer GPU.<n>We compare five matched workloads with PyTorch on NVIDIA and AMD discrete GPUs.<n>The results show that general consumer graphics APIs can support a compact shared train-to-deploy stack at useful, sometimes vendor-competitive performance.
  arXiv  Detail & Related papers  (2026-08-03T00:42:55Z)
• DMax: Aggressive Parallel Decoding for dLLMs [77.24184219948337]
  We present DMax, a new paradigm for efficient diffusion language models (dLLMs)<n>It mitigates error accumulation in parallel decoding, enabling aggressive decoding while preserving generation quality.<n>At the core of our approach is On-Policy Uniform Training, a novel training strategy that efficiently unifies masked and uniform dLLMs.
  arXiv  Detail & Related papers  (2026-04-09T14:35:42Z)
• ParaCodex: A Profiling-Guided Autonomous Coding Agent for Reliable Parallel Code Generation and Translation [5.103450268268085]
  ParaCodex is an HPC-engineer workflow that turns a Codex-based agent into an autonomous OpenMP offload system.<n>We evaluate translation from serial CPU kernels to OpenMP offload kernels on HeCBench, Rodinia, and NAS.<n>We achieve geometric-mean speedups of 3x on HeCBench and 5x on Rodinia, and outperforming a zero-shot Codex baseline on all suites.
  arXiv  Detail & Related papers  (2026-01-07T19:04:53Z)
• FuseSampleAgg: Fused Neighbor Sampling and Aggregation for Mini-batch GNNs [51.56484100374058]
  FuseSampleAgg fuses neighbor and sampling mean aggregation into a single pass for GraphSAGE.<n>Operator is deterministic, integrates with standard PyTorchs, and ships with scripts that reproduce all tables and figures from CSV logs.
  arXiv  Detail & Related papers  (2025-11-17T17:57:18Z)
• Nexus:Proactive Intra-GPU Disaggregation of Prefill and Decode in LLM Serving [4.309392302169281]
  Engine-level prefill-decode (PD) disaggregation avoids interference but incurs higher hardware and coordination overhead.<n>PD achieves up to 2.2x higher throughput, 20x lower TTFT, and 2.5x lower TBT than vLLM; outperforms SG by up to 2x; and matches or exceeds disaggregated vLLM.
  arXiv  Detail & Related papers  (2025-07-09T07:27:18Z)
• A Simple Linear Patch Revives Layer-Pruned Large Language Models [58.056251480151104]
  Layer pruning has emerged as a widely used technique for compressing large language models (LLMs)<n>textscLinearPatch fuses two operations into one matrix multiply at the pruning interface.<n>The patch can be further refined with 5K unlabeled samples via memory-efficient offline distillation, pushing the retention to 95.16% within only 30 minutes on a single GPU.
  arXiv  Detail & Related papers  (2025-05-30T15:06:08Z)
• 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)
• Sparse VideoGen2: Accelerate Video Generation with Sparse Attention via Semantic-Aware Permutation [84.00166854547241]
  Diffusion Transformers (DiTs) are essential for video generation but suffer from significant latency due to the quadratic complexity of attention.<n>We propose SVG2, a training-free framework that maximizes identification accuracy and computation minimizes waste.
  arXiv  Detail & Related papers  (2025-05-24T21:30:29Z)
• Hardware-Aware Parallel Prompt Decoding for Memory-Efficient Acceleration of LLM Inference [23.633481089469836]
  Auto-regressive decoding of Large Language Models (LLMs) results in significant overheads in their hardware performance.<n>We propose a novel parallel prompt decoding that requires only $0.0002$% trainable parameters, enabling efficient training on a single A100-40GB GPU in just 16 hours.<n>Our approach demonstrates up to 2.49$times$ speedup and maintains a minimal memory overhead of just $0.0004$%.
  arXiv  Detail & Related papers  (2024-05-28T22:19:30Z)
• Hybrid Models for Learning to Branch [81.93868699246214]
  We propose a new hybrid architecture for efficient branching on CPU machines. The proposed architecture combines the expressive power of GNNs with computationally inexpensive multi-layer perceptrons (MLP) for branching.
  arXiv  Detail & Related papers  (2020-06-26T21:03:45Z)

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.