/Patent Tracker
NvidiaGranted patentChips & hardwareInterest 4/10

Nvidia patents memory layout for mixed-precision neural network data

Patent drawing
Patent drawing from the official filing.

Nvidia describes a method for organizing computer memory to hold different types of numerical data side by side. The system uses a specific index to track where higher-precision and lower-precision numbers are stored within a single block of memory.

Why it mattersThis technique helps processors manage data more efficiently, which is a standard optimization for running complex calculations like those in neural networks.

In the filing’s own words

Apparatuses, systems, and techniques to format memory to store values of mixed precision. In at least one embodiment, processor(s) perform operations to format memory to store values of mixed precision in a block of contiguous memory by, in part, using an indication of a number of higher-precision values to be stored in memory.

What exactly they claim (claim 1)

1. One or more processors, comprising: circuitry to; store one or more first precision numbers and one or more second precision numbers corresponding to one or more neural network parameters within a contiguous range of storage; and generate an index indicating locations, within the contiguous range of storage, of the one or more first precision numbers and the one or more second precision numbers.

Facts

Company
Nvidia (NVIDIA Corporation)
Official title
Data storage techniques
Patent number
US12762056B1
Application number
18923573
Filed
Oct 22, 2024
Granted
Oct 13, 2026
Inventors
2
Claims
21

Classified under

G06F 12/08 Physics & computingG06F 17/16 Physics & computing
Read the full filing (official PDF) ↗ Open in the Patent Tracker

More from Nvidia in Chips & hardware

Summaries are simplified from the filing’s own abstract and first claim. A patent filing is not a product announcement.
Patent data and drawings: U.S. Patent and Trademark Office (USPTO). Nebula is not affiliated with or endorsed by the USPTO.