Breakthrough Low-Latency, High-Energy-Efficiency LLM Inference Performance Using NorthPole

For a 3-billion-parameter LLM, a research prototype inference appliance with 16 IBM AIU NorthPole processors delivers a massive 28,356 tokens / second of system throughput and sub-1 ms / token (per-user) latency while consuming merely 672 W for 16 NorthPole cards in a compact 2U form factor. With a focus on low latency and high energy efficiency, when NorthPole (in 12 nm) is compared to a suite of GPUs (in 7/5/4 nm) at various power consumptions, at the lowest GPU latency, NorthPole provides 72.7x better energy metric (tokens/second/W) while providing better latency.

Authors

Publication Details

Published
2024-09-23
DOI
https://doi.org/10.1109/hpec62836.2024.10938418
Citations
3
Primary Topic
Image and Signal Denoising Methods
Type
article
Field-Weighted Citation Impact
0.75
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
article

Breakthrough Low-Latency, High-Energy-Efficiency LLM Inference Performance Using NorthPole

Filipp Akopyan, Guillaume Garreau, Myron Flickner, Takanori Ueda et al.
3 citations
Image and Signal Denoising Methods
0.75
article

Breakthrough Low-Latency, High-Energy-Efficiency LLM Inference Performance Using NorthPole

Filipp Akopyan, Guillaume Garreau, Myron Flickner, Takanori Ueda, Rathinakumar Appuswamy, Michael DeBole, Peter J. Carlson, Brian Taba, W. P. Risk, Megumi Ito, Andrew S. Cassidy, Deepika Bablani, Rameswar Panda, Ignacio Terrizzano, Steven K. Esser, Rafael T. de Sousa, Jeffrey L. McKinstry, Nathaniel J. McClatchey, Dharmendra S. Modha, Yutaka Nakamura, Pallab Datta, Carlos Ortega Otero, Jeffrey A. Kusnitz, Jun Sawada, Jennifer L. Klamo, John V. Arthur, Neil McGlohon, Rajamohan Gandhasri, Jay Sivagnaname, Alexander Andreopoulos, Matthew Stallone, David Cox, Arnon Amir, Tapan K. Nayak, Nelson Mimura Gonzalez, Daniel F. Smith, Kai Schleupen
article en
3 citations

Abstract

For a 3-billion-parameter LLM, a research prototype inference appliance with 16 IBM AIU NorthPole processors delivers a massive 28,356 tokens / second of system throughput and sub-1 ms / token (per-user) latency while consuming merely 672 W for 16 NorthPole cards in a compact 2U form factor. With a focus on low latency and high energy efficiency, when NorthPole (in 12 nm) is compared to a suite of GPUs (in 7/5/4 nm) at various power consumptions, at the lowest GPU latency, NorthPole provides 72.7x better energy metric (tokens/second/W) while providing better latency.

Affordable and clean energy
Openalex Percentile: Top 26%
Image and Signal Denoising Methods
0.75
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.