Search arXiv⌕ Search
Home/Archive/arXiv

arXiv · 2610.14244

Understanding the Power Consumption of Modern High Bandwidth Memory: Experimental Analysis and Modeling Using Real HBM2 DRAM Chips

Ataberk Olgun·Spiros Galanopoulos·Andreas Kosmas Kakolyris·İsmail Emir Yüksel·Haocong Luo·F. Nisa Bostancı·Onur Mutlu
View original source ↗Read source document ↗Export metadata ↓

Abstract

We perform the first comprehensive experimental characterization of power consumed by real HBM2 DRAM chips. Our extensive characterization yields 1) the empirical basis for developing new, accurate HBM power models or validating existing power models and 2) key insights into structural and data-pattern-dependent variation of power in HBM2 chips. Out of our 13 new findings and 6 key takeaway messages, we highlight: 1) a read operation consumes significantly higher current than any other operation and 2) the toggled data bus signals in a data transfer burst significantly affect current draw where toggling all HBM2-internal bank group bus signals causes 29.4% higher power consumption than keeping them constant. Leveraging our empirical insights into HBM2 power consumption, we engineer an access pattern that yields 23.6% higher power consumption on average across 36 tested HBM2 chips than the highest-power-consuming JEDEC-specified current measurement loop. Our results have important implications for HBM-based system design and validation. Relying solely on JEDEC- or vendor-specified test loops to bound HBM power could cause system failures and/or thermal throttling in real-world high-power AI/ML workloads. To improve 3D-stacked DRAM power modeling in open source architectural simulation infrastructures, we develop a new HBM power model. Our model predicts the power of 36 real HBM2 chips across diverse workloads (including a real LLM decode workload) with 7.6 -- 8.1% mean absolute percentage error, substantially more accurately than state-of-the-art HBM power models. The power consumption predicted by our model for a power-intensive microbenchmark closely matches (with 2.4% to 5.3% error across HBM3E's specified supply voltage range) the average HBM3E memory power consumed by three tested NVIDIA H200 GPUs. We publicly release our power model at https://github.com/CMU-SAFARI/HBM-Power.

Explore related subjects

cs.ARcs.PF

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Ataberk Olgun, Spiros Galanopoulos, Andreas Kosmas Kakolyris, İsmail Emir Yüksel, Haocong Luo, F. Nisa Bostancı, Onur Mutlu. 2026-10-09. Understanding the Power Consumption of Modern High Bandwidth Memory: Experimental Analysis and Modeling Using Real HBM2 DRAM Chips. https://arxiv.org/abs/2610.14244

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

KEEP EXPLORING

Related papers

arXiv2026

ARES: Adaptive Reasoning-Effort Steering for PPA- and Cost-Aware RTL Optimization with LLM Agents

Large language model (LLM) agents optimize the power, performance, and area (PPA) of register-transfer-level (RTL) designs by iterating over edits, synthesis, and PPA analysis, paying a dollar cost for every LLM call. Prior agents report the quality reached without its normalized cost, attribute that quality to an engineered cross-design memory, and hold the reasoning effort of every call fixed. We propose Ares with three corresponding innovations. (1) We introduce a normalized dollar cost per LLM call reported alongside the figure of merit (FoM), enabling fair comparison across effort levels and optimizers. (2) Using this accounting, we find the construction of the long-term memory matters little. An engineered memory brings no dependable gain over a plain concatenation of the same experience. (3) We instead adapt the per-call reasoning effort by escalating to deeper reasoning only once progress at a lower effort stalls, via a patience counter fit on 21 training designs, allocating reasoning where it pays rather than uniformly across all iterations. On three test designs unseen during training, the effort policy lowers the FoM by 22-28% with Opus 4.6 and by 18-34% with GPT-5.5, where the best fixed effort reaches 20-24% and 19-29%, respectively, at equal normalized cost. Ares closes on average 91% of the gap from an LLM-drafted multiply-accumulate unit to its highly hand-optimized counterpart. It also reaches a 5% lower FoM than state-of-the-art Dr. RTL at only 35% of its cost.

cs.AR↗
arXiv2026

Hard and Soft Real-Time DNN Inference Scheduling on Heterogeneous Multi-Accelerator Systems via Reinforcement Learning

We present a scheduling framework for heterogeneous multi-accelerator systems executing both hard real-time and soft real-time QoS-aware DNN inference workloads. The framework combines an online reinforcement learning (RL) scheduler with a deterministic runtime safety layer. The RL scheduler dynamically generates candidate schedules by jointly selecting sub-job priorities and accelerator assignments, while runtime certification guarantees that all HRT deadlines remain satisfied independently of the learned policy. To enable runtime certification in heterogeneous precedence-constrained systems, we introduce a schedulability-preserving prefix replacement criterion based on machine-wise cuts. The proposed criterion guarantees that runtime-generated schedule prefixes preserve the existence of a feasible continuation without requiring complete finish-time dominance with respect to a feasible reference schedule. Experimental results on heterogeneous DNN accelerator platforms show that the proposed framework preserves HRT schedulability while improving QoS-aware deadline satisfaction, client-level service quality, and accelerator utilization compared to state-of-the-art scheduling approaches.

cs.AR↗
arXiv2026

HDLouvain: High-Throughput Incremental Clustering for Hyperdimensional Data

Modern shotgun proteomics generates tens of thousands of tandem mass spectra per hour, producing hundreds of gigabytes of data that strain conventional clustering pipelines. DBSCAN and hierarchical agglomerative clustering (HAC) offer flexible similarity search, but DBSCAN is unstable across varying data densities, HAC costs $O(n^2)$, and both yield poor cluster quality. Graph-based methods such as Louvain produce high-quality clusters but scale poorly on streaming data. We observe that Louvain's poor scalability can be resolved without sacrificing its quality, and present HDLouvain, which combines Louvain community detection with the bit-parallel similarity computation of hyperdimensional computing (HDC). HDLouvain partitions a bootstrap fraction of the stream into precursor-mass buckets, applies Louvain to each bucket's kNN similarity graph, and condenses each community into a single binary hypervector by bitwise-majority bundling rather than iterative centroid recomputation. New spectra are assigned by a GPU-accelerated Hamming distance lookup against representatives in their mass bucket, or seed a new cluster when no match meets a similarity threshold $τ$, so the partition grows incrementally without full re-clustering. On five PRIDE datasets (0.99M to 21.13M spectra), sweeping the bootstrap fraction $α$ from 0.2 to 0.9 exposes an explicit, monotonic throughput-fidelity trade-off. On PXD003258, speedup ranges from 4.5x at $α=0.3$ to 1.65x at $α=0.7$. Modularity rises from 0.75-0.76 at $α=0.2$ to 0.95-0.97 at $α=0.9$, never more than 8% below full static recomputation. The Louvain stage achieves 0.86-0.93 normalized mutual information (NMI) on LFR benchmark networks with 30k to 2M nodes, whereas DBSCAN and HAC degrade to 0.57.

cs.AR↗