Searcharxiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.23503

HBFSim: A Simulator for Exploring HBF in Memory Hierarchies for LLMs

Jinheng Li·Guosheng Wang·Hongchao Du·Shangyu Wu·Qiao Li·Chun Jason Xue
View original source ↗Read source document ↗Export metadata ↓

Abstract

High Bandwidth Flash (HBF) has recently emerged as a promising technology that provides terabyte-scale accelerator-local capacity at bandwidths approaching HBM, with a potential cost-per-bit advantage over HBM. However, its NAND foundation brings long latency, out-of-place updates, and finite write endurance directly into the accelerator memory hierarchy, making its effects on performance and lifetime difficult to predict. To study this emerging design space, we present HBFSim, an LLM-aware event-driven simulator for heterogeneous HBM+HBF memory systems that preserves application-level memory semantics while modeling flash-internal behavior. HBFSim models HBM, HBF with host-managed flash translation, and conventional capacity tiers in one simulation driven by a calibrated serving frontend. Using HBFSim, we study decoding of Llama-3.1-70B and Qwen3-235B-A22B under three workloads that differ in their input and output lengths. HBF serves data that does not fit in HBM 11.0--19.0$\times$ faster than host DRAM; HBM is better used to hold the running sequences' KV cache than as a cache for HBF; and at a published peak load of Llama-3.1-70B, HBF lasts 2.0 to 6.9 years, in proportion to the stacks that share the KV writes. The artifact is available at https://github.com/FlashLabAE/HBFSim and will track public HBF device data as it becomes available.

Explore related subjects

cs.ARcs.ET

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Jinheng Li, Guosheng Wang, Hongchao Du, Shangyu Wu, Qiao Li, Chun Jason Xue. 2026-10-16. HBFSim: A Simulator for Exploring HBF in Memory Hierarchies for LLMs. https://doi.org/10.1145/3848591.3856870

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

KEEP EXPLORING

Related papers

arXiv2026

Is Your NPU Ready for LLMs? Dissecting the Hidden Efficiency Bottlenecks in Mobile LLM Inference

Deploying Large Language Models (LLMs) on mobile devices enhances privacy and reduces latency, but is severely constrained by hardware inefficiencies. We present the first comprehensive, cross-layer measurement study of mobile LLM inference, spanning five mainstream frameworks (e.g., llama.cpp and GENIE) and three hardware backends (CPU, GPU, and NPU). To enable this analysis, we develop PowerBench, a fine-grained profiling tool that measures LLM throughput and provides the first backend-specific energy attribution for mobile LLM inference, going beyond traditional device-level measurements. We further conduct hardware microbenchmarks to characterize the memory-system performance and compute capabilities of mobile CPUs, GPUs, and NPUs. Our study yields three critical insights: (1) Framework-induced performance gaps are substantially amplified on NPUs, reaching up to $10\times$ due to differences in offloading strategies, hardware-underutilizing operators, and activation quantization strategies. (2) We identify a distinct ``phase split,'' in which NPUs excel at compute-bound prefilling, while CPUs outperform all other backends in memory-bound decoding. This split is driven by NPUs' preference for large, fixed-shape workloads, which conflicts with the small-kernel, dynamic nature of decoding. (3) Backend-specific profiling uncovers substantial scheduling headroom missed by prior work. Suboptimal thread configurations, uncoordinated NPU sleep latencies, and CPU polling intervals result in unnecessary energy consumption. For example, tuning CPU frequency alone reduces energy consumption by up to $39.7\%$. Leveraging these findings, we estimate that an energy-oriented configuration can reduce energy consumption by up to $53.2\%$ on the NPU backend across three datasets, highlighting the substantial potential for further efficiency improvements in mobile LLM deployment.

cs.AR↗
arXiv2026

A LoRa-Compatible Receiver with Programmable Noise-Power Trade-Off and Integrated Wake-Up Mode

Programmable receivers allow an optimal performance in the variable conditions of the wireless channel. An extensive analysis of their noise-power trade-off enables the development of a main receiver with embedded wake-up circuitry. This makes the seamless integration of wake-up receivers (WuRXs) in RF receivers possible, avoiding power, area, complexity, and cost overheads in the system. In this paper, we propose the first LoRa-compatible receiver with an integrated wake-up mode and a programmable noise-power trade-off. In main-receiver mode, it consumes 4.57 mW for a sensitivity of -138 dBm, but it can save up to 62.6% of power consumption at the cost of an 8.3-dB sensitivity degradation. In wake-up mode, sensitivity down to -126 dBm is achieved for a 32-ms latency and 1.5 mW of power consumption. The power consumption remains relatively high due to a sub-optimal design of the the phase-locked loop (PLL). In the end, the wake-up mode consumes 3X less than the lowest-noise configuration in main-receiver mode.

cs.AR↗
arXiv2026

BS-Spec: BFP-Based Bit-Sharing for Self-Speculative LLM Decoding with LUT-Driven Mixed-Precision Acceleration

As large language models (LLMs) are increasingly deployed across diverse inference services, serving workloads span a wide range of batch sizes, context lengths, and performance requirements. Speculative decoding improves inference throughput, but conventional approaches require a separate draft model and KV cache, while existing self-speculative methods suffer from workload-insensitive draft configurations and inefficient mixed-precision execution. In this paper, we present BS-Spec, a block floating point (BFP)-based bit-sharing framework that jointly optimizes draft construction, workload-aware precision selection, and hardware execution. BS-Spec derives multiple draft representations directly from a single BFP target model by partially loading the most significant bits of weights and the KV cache while preserving shared exponents, eliminating separate draft-model storage. Offline calibration selects the weight and KV-cache precision pair that maximizes estimated speedup for each serving regime, and a precision lookup table applies the selected configuration at runtime. To support this adaptive execution, we design a mixed-precision BFP accelerator with a lookup table (LUT)-based variable-precision datapath for both draft and target computation. Experimental results show that BS-Spec achieves up to $1.9\times$ speedup over conventional autoregressive decoding and improves speedup over the evaluated bit-sharing self-speculative decoding baselines by $67.2\%$ on average, demonstrating the benefits of jointly co-designing bit sharing, precision adaptation, and specialized hardware for diverse LLM serving workloads.

cs.AR↗