Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2603.08747

Diagnosing FP4 inference: a layer-wise and block-wise sensitivity analysis of NVFP4 and MXFP4

Musa Cim·Burak Topcu·Mahmut Taylan Kandemir
View original source ↗Read source document ↗Export metadata ↓

Abstract

Quantization addresses the high resource demand for large language models (LLMs) by alleviating memory pressure and bandwidth congestion and providing significantly scaled compute power with a tolerable impact on accuracy. Four-bit floating point (FP4), the lowest-precision format that preserves essential numerical properties such as exponent and sign, has begun to be adopted in cutting-edge architectures, including Blackwell and AMD CDNA, to support LLM quantization and reduce deployment costs. Although aggressive quantization can yield efficiency gains, the quantization sensitivity of within-transformer layers and whether these sensitivities generalize across existing FP4 formats and model scales remain underexplored. To elucidate quantization sensitivity, this study conducts a systematic analysis of two FP4 formats, MXFP4 and NVFP4, across three Qwen2.5 model scales (0.5B, 7B, and 14B), using controlled component-wise and block-wise isolation methodologies. We observe that MLP up- and down-projection layers consistently dominate in terms of sensitivity, while gate and attention projections are moderately and substantially less sensitive to FP4 quantization, respectively. We further find that sensitivity does not universally localize to the final blocks, but early blocks can be highly sensitive, particularly under MXFP4. Our results provide a diagnostic characterization of the inference behavior of FP4 across components, depths, and FP4 formats.

Explore related subjects

cs.ARcs.AI

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Musa Cim, Burak Topcu, Mahmut Taylan Kandemir. 2026-03-05. Diagnosing FP4 inference: a layer-wise and block-wise sensitivity analysis of NVFP4 and MXFP4. https://arxiv.org/abs/2603.08747

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related papers

arXiv2026

A System Architecture for Low Latency Multiprogramming Quantum Computing

As quantum systems scale, multiprogramming quantum computing (MPQC) provides a practical way to improve device utilization and throughput. However, because quantum executables are device-dependent, non-portable across qubit regions, and highly susceptible to noise and crosstalk, current MPQC pipelines rely on expensive online compilation to co-optimize concurrently running programs. This online step dominates runtime and impedes low-latency deployments for practical, real-world workloads in the future, such as repeatedly invoked quantum neural network (QNN) services. We present FLAMENCO, a fidelity-aware multi-version compilation system that enables independent offline compilation and low-latency multiprogramming at runtime. \textbf{At the architecture level}, the system abstracts devices into compute units to reduce the search space of region allocation. \textbf{At compile time}, it generates diverse executable versions for each program---each bound to a distinct qubit region---allowing dynamic region selection at runtime and overcoming non-portability. \textbf{At runtime}, it employs a lightweight orchestrator that uses post-compilation fidelity metrics to avoid conflicts and mitigate crosstalk, supporting conflict-free co-execution without online co-optimization. Evaluations show that FLAMENCO achieves over 5$\times$ runtime speedup in post-scheduling execution while maintaining comparable execution fidelity on common-success workloads. When integrated into existing scheduler-coupled systems, it raises workload-level conflict-free orchestration ratio from 0.183 to 1.000 for HyperQ and from 0.050 to 0.400 for QOS.

cs.AR↗
arXiv2026

HBFSim: Fast and Faithful Simulation of High-Bandwidth Flash Under Real GPU Execution

High-Bandwidth Flash (HBF) places high-capacity NAND beside HBM to relieve the memory-capacity bottleneck of LLM inference, yet its system-level behavior cannot be evaluated before hardware becomes available. Cycle-level GPU simulators are too slow for production-scale models. Trace replay has a further shortcoming: it cannot capture the allocation, migration, and execution changes induced by different HBM-HBF configurations. Our key insight is that HBF need not be evaluated by simulating the GPU: only the program-visible effects of HBF need to be modeled. And only a real LLM workload running on real hardware can answer the arguments about HBF. Hence the modeled service has to be injected into that running program, and the injection must not destroy the GPU concurrency that would hide the original I/O latency. We present HBFSim, an open-source HBF simulator that executes LLM workloads on a real GPU while modeling HBF timing, thermal, and other behaviors online. HBFSim rewrites the PTX of the workload's kernels and routes accesses inside a registered address range into the HBF simulator. It supports asynchronous TMA transfers and capacities beyond physical GPU memory. HBFSim leaves the model's run unaffected across ordinary-memory, TMA, and capacity-mode tests. The delay it injects matches the delay requested to within 0.152%. We also design a coupled thermal module that puts HBF, HBM, and the GPU in one advanced package, which is important for answering how severe the hot throttling problem becomes after HBF runs for a long time. Experiments with Qwen3-30B show how package heating, HBM-HBF allocation, and shared MoE demand jointly constrain the design space of future HBF accelerators.

cs.AR↗
arXiv2026

Bio-inspired Learning and Decision-Making with Probabilistic In-Memory Computing Hardware: Part 2

This report extends our previous work (Part 1), which introduced an energy-based model for learning and decision-making under uncertainty. The model leverages stochastic Langevin dynamics to continuously evolve approximate probability distributions over neuron states and model weights. However, as noted in Part 1 and confirmed through GPU-based implementations, large-scale probabilistic energy-based models of this nature face significant scalability challenges due to excessive execution latency. This latency stems from a fundamental mismatch: massively parallel models with low arithmetic intensity (such as energy-based models) are being executed on processor architectures like GPUs that rely on high-bandwidth memory (HBM) interfaces. The HBM imposes brutally sequential execution constraints on inherently parallelizable models, creating the false impression that such models are unscalable. In reality, it is the GPU architecture itself, with its dependence on HBM interfaces, that is not a scalable processor architecture for this class of AI model. In this report, we demonstrate using a detailed transaction-level model (TLM) of a probabilistic analogue in-memory computing (AIMC) processor that the same energy-based model can execute well over 1000x faster than data-center-grade hardware by eliminating the HBM interface and performing computation directly within on-chip memory.

cs.AR↗