Accurate Models of NVIDIA Tensor Cores Article Swipe

MAKE IMPACT

Khattak, Faizan A. , Mikaitis, Mantas · YOU? · · 2025 · Open Access · · DOI: https://doi.org/10.48550/arxiv.2512.07004

Matrix multiplication is a fundamental operation in for both training of neural networks and inference. To accelerate matrix multiplication, Graphical Processing Units (GPUs) provide it implemented in hardware. Due to the increased throughput over the software-based matrix multiplication, the multipliers are increasingly used outside of AI, to accelerate various applications in scientific computing. However, matrix multipliers targeted at AI are at present not compliant with IEEE 754 floating-point arithmetic behaviour, with different vendors offering different numerical features. This leads to non-reproducible results across different generations of GPU architectures, at the matrix multiply-accumulate instruction level. To study numerical characteristics of matrix multipliers-such as rounding behaviour, accumulator width, normalization points, extra carry bits, and others-test vectors are typically constructed. Yet, these vectors may or may not distinguish between different hardware models, and due to limited hardware availability, their reliability across many different platforms remains largely untested. We present software models for emulating the inner product behavior of low- and mixed-precision matrix multipliers in the V100, A100, H100 and B200 data center GPUs in most supported input formats of interest to mixed-precision algorithm developers: 8-, 16-, and 19-bit floating point.

Type

preprint

Landing Page

https://doi.org/10.48550/arxiv.2512.07004

OA Status

green

OpenAlex ID

https://openalex.org/W7110955317

Raw OpenAlex JSON

OpenAlex ID

https://openalex.org/W7110955317

Canonical identifier for this work in OpenAlex

DOI

https://doi.org/10.48550/arxiv.2512.07004

Digital Object Identifier

Title

Accurate Models of NVIDIA Tensor Cores

Work title

Type

preprint

OpenAlex work type

Publication year

2025

Year of publication

Publication date

2025-12-07

Full publication date if available

Authors

Khattak, Faizan A., Mikaitis, Mantas

List of authors in order

Landing page

https://doi.org/10.48550/arxiv.2512.07004

Publisher landing page

Open access

Yes

Whether a free full text is available

OA status

green

Open access status per OpenAlex

OA URL

https://doi.org/10.48550/arxiv.2512.07004

Direct OA link when available

Concepts

Matrix multiplication, Computer science, Rounding, Computational science, Parallel computing, Software, Matrix (chemical analysis), Algorithm, Normalization (sociology), Floating point, Computation, Throughput, Computer engineering, Round-off error, Arithmetic, Matrix decomposition, Matrix representation, Computer hardware, Double-precision floating-point format, Single-precision floating-point format, Theoretical computer science, Outer product, Numerical linear algebra, Adder, Digital signal processing, Sparse matrix, Product (mathematics), Artificial neural network, State-transition matrix, Signal processing, Dot product, Carry (investment)

Top concepts (fields/topics) attached by OpenAlex

Cited by

0

Total citation count in OpenAlex

Full payload