Slovensko nacionalno superračunalniško omrežje

Seminar: An Approximate GEMM Unit for Energy-Efficient Processing

An Approximate GEMM Unit for Energy-Efficient Processing

Energijsko učinkovito računanje z enoto GEMM s približnimi množilniki

by Ratko Pilipović, UL FRI

Link to the event: https://uni-lj-si.zoom.us/j/99732970797 (Wednesday, 8 December 2021, 14:15)

 

AGEMM unit

ENG: Approximate computing has emerged as a popular strategy for energy-efficient circuit design, where the challenge is to achieve the best tradeoff between design efficiency and accuracy. The essential
operation in artificial intelligence algorithms is the general matrix multiplication (GEMM) operation comprised of matrix multiplication and accumulation. An approximate general matrix multiplication (AGEMM) unit employs approximate multipliers to perform matrix–matrix operations on four-by-four matrices given in sixteen-bit signed fixed-point format. The synthesis of the proposed AGEMM unit to the 45 nm Nangate Open Cell Library revealed that it consumed only up to 36% of the area and 25% of the energy required by the exact general matrix multiplication unit. The AGEMM unit is ideally suited to convolutional neural networks, which can adapt to the error induced in the computation. The results on honeybee detection with the YOLOv4-tiny convolutional neural network imply that we can deploy the AGEMM units in convolutional neural networks without noticeable performance degradation. Moreover, the AGEMM unit’s employment can lead to more area- and energy-efficient convolutional neural network processing, let it be in the edge devices or HPC centres.

SLO: Približno računanje se uveljavlja kot innovativen  pristop pri načrtovanju hitrih in energijsko učinkovitih vezij. Enota GEMM za splošno množenje matrik, Nvidia ji pravi Tensor Core, močno pospeši množenje matrik. Z zamenjavo množice natančnih množilnikov s približnimi, dobimo mnogo manjšo in energijsko učinkovitejšo enoto. Uporaba GEMM enote s približnimi množilniki pri razvrščanju čebel s kovolucijsko nevronsko mrežo YOLOv4-tiny, da primerljive rezultate, kot enota GEMM z natančnimi množilniki. Predvsem velikost in energijska poraba enote nakazujeta njeno uporabnost tako v avtonomnih napravah kot tudi v superračunalnikih.


Organizator / Organizer


Več informacij lahko poiščete na spletni strani dogodka, kjer se lahko tudi prijavite.

Dostopnost