Nvidia patents memory trick to squeeze more speed from chip calculations
Nvidia describes a hardware method that helps computer chips read data from shared memory more efficiently. It uses a circular buffer system to handle large math tasks that don't fit neatly into standard memory blocks, allowing the chip to keep working without waiting for data.
In the filing’s own words
Apparatuses, systems, and techniques for shared memory optimization through circular memory buffers are disclosed herein. The system may include an address generation unit to generate addresses for a single matrix operation that processes data spanning non-contiguous buffers in shared memory. The address generation unit may store a first buffer address and a second buffer address and coordinate access to both buffers for the single matrix operation. The system may include boundary detection logic to determine when the single matrix operation requires data from multiple buffers. The system may support matrix operations that process multiple precision formats that require a predetermined amount of data per operation. The predetermined amount may be a non-integer multiple of a buffer storage capacity. The address generation unit can enable optimal memory bandwidth utilization by supporting full-capacity buffer loads while accommodating fractional buffer consumption requirements through circular buffer organization.
What exactly they claim (claim 1)
1. A processing device comprising: an address generation unit to generate addresses for a single matrix operation that processes data spanning non-contiguous buffers in shared memory.
Facts
- Company
- Nvidia (NVIDIA Corporation)
- Official title
- SHARED MEMORY OPTIMIZATION THROUGH HARDWARE-IMPLEMENTED CIRCULAR BUFFER
- Publication number
- US20260277470A1
- Application number
- 19355061
- Filed
- Oct 10, 2025
- Published
- Sep 17, 2026
- Inventors
- 5
- Claims
- 20
Classified under
G06F 3/0656 Physics & computingG06F 3/0604 Physics & computingG06F 3/0673 Physics & computingMore from Nvidia in Chips & hardware
- Nvidia patents method to skip zero data in neural network math Oct 13, 2026
- Nvidia patents hardware to speed up motion blur in ray tracing Oct 13, 2026
- Nvidia patents method to sync data across split processor chips Sep 17, 2026
- Nvidia patents chip math trick to speed up AI calculations Oct 8, 2026
- Nvidia patents data center cooling that uses cryogenic fluids Oct 15, 2026
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.