About: DGEMM Using Tensor Cores, and Its Accurate and Reproducible Versions

Facets (new session)
Description
Metadata
Settings
- owl:sameAs
- Inference Rule:

About: DGEMM Using Tensor Cores, and Its Accurate and Reproducible Versions Goto Sponge NotDistinct Permalink

An Entity of Type : schema:ScholarlyArticle, within Data Space : wasabi.inria.fr associated with source document(s)

Attributes	Values
type	Academic Article research paper schema:ScholarlyArticle
isDefinedBy	Covid-on-the-Web dataset
has title	DGEMM Using Tensor Cores, and Its Accurate and Reproducible Versions
Creator	Imamura, Toshiyuki Mukunoki, Daichi Ogita, Takeshi Ozaki, Katsuhisa
Source	PMC
abstract	This paper proposes a method for implementing dense matrix multiplication on FP64 (DGEMM) and FP32 (SGEMM) using Tensor Cores on NVIDIA’s graphics processing units (GPUs). Tensor Cores are special processing units that perform [Formula: see text] matrix multiplications on FP16 inputs with FP32 precision, and return the result on FP32. The proposed method adopts the Ozaki scheme, an accurate matrix multiplication algorithm based on error-free transformation for matrix multiplication. The proposed method has three prominent advantages: first, it can be built upon the cublasGemmEx routine using Tensor Core operations; second, it can achieve higher accuracy than standard DGEMM, including the correctly-rounded result; third, it ensures bit-level reproducibility even for different numbers of cores and threads. The achievable performance of the method depends on the absolute-value range of each element of the input matrices. For example, when the matrices were initialized with random numbers over a dynamic range of 1E+9, our DGEMM-equivalent implementation achieved up to approximately 980 GFlops of FP64 operation on the Titan RTX GPU (with 130 TFlops on Tensor Cores), although cublasDgemm can achieve only 539 GFlops on FP64 floating-point units. Our results reveal the possibility of utilizing hardware with limited FP32/FP64 resources and fast low-precision processing units (such as AI-oriented processors) for general-purpose workloads.
has issue date	2020-05-22(xsd:dateTime)
bibo:doi	10.1007/978-3-030-50743-5_12
has license	no-cc
sha1sum (hex)	cb96496812e3cb5d524c88e9481d3fcba7cd8cad
schema:url	https://doi.org/10.1007/978-3-030-50743-5_12
resource representing a document's title	DGEMM Using Tensor Cores, and Its Accurate and Reproducible Versions
has PubMed Central identifier	PMC7295351
schema:publication	High Performance Computing
resource representing a document's body	covid:cb96496812e3cb5d524c88e9481d3fcba7cd8cad#body_text
is schema:about of	named entity 'hardware' named entity 'matrix multiplication algorithm' named entity 'processors' named entity 'FP64' named entity 'GPUs' named entity 'precision' named entity 'prominent' named entity 'NVIDIA' named entity 'Accurate' named entity 'REPRODUCIBILITY' named entity 'OUR' named entity 'INCLUDING' named entity 'DENSE' named entity 'METHOD' named entity 'SCHEME' named entity 'RANDOM' named entity 'RANGE' named entity 'IMPLEMENTATION' covid:arg/cb96496812e3cb5d524c88e9481d3fcba7cd8cad named entity 'processing' named entity 'operations' named entity 'TFlops' named entity 'processing' named entity 'GPU' named entity 'standard' named entity 'Cores' named entity 'Core' named entity 'Cores' named entity 'accurate' named entity 'GFlops' named entity 'reveal' named entity 'limited' named entity 'DGEMM' named entity 'DGEMM' named entity 'error-free' named entity 'SGEMM' named entity 'dynamic range' named entity 'GPUs' named entity 'FP64' named entity 'TFlops' named entity 'bit-level' named entity 'matrix multiplication' named entity 'Tensor Cores' named entity 'general-purpose' named entity 'matrix multiplication algorithm' named entity 'FP64' named entity 'Tensor Core' named entity 'GFlops' named entity 'FP32' named entity 'DGEMM' named entity 'Tensor' named entity 'FP16' named entity 'inner-product' named entity 'Algorithm' named entity 'FP16' named entity 'FP64' named entity 'FP32' named entity 'GPU' named entity 'matrix multiplication' named entity 'algorithm' named entity 'single-precision' named entity 'GEMM' named entity 'FP64' named entity 'FP16' named entity 'FP16' named entity 'Tesla V100' named entity 'FP64' named entity 'Tesla V100' named entity 'GPU' named entity 'GEMM' named entity 'FP32' named entity 'vectorization' named entity 'GEMM' named entity 'FP64' named entity 'GFlops' named entity 'DGEMM' named entity 'Tensor Cores' named entity 'HPC' named entity 'computational complexity' named entity 'DGEMM'

Faceted Search & Find service v1.13.91 as of Mar 24 2020

Alternative Linked Data Documents: Sponger | ODE Content Formats:

RDF

ODATA

Microdata

About

OpenLink Virtuoso version 07.20.3229 as of Jul 10 2020, on Linux (x86_64-pc-linux-gnu), Single-Server Edition (94 GB total memory)
Data on this page belongs to its respective rights holders.
Virtuoso Faceted Browser Copyright © 2009-2024 OpenLink Software