liveBook · Manning

SRAM

As explained in GPU Programming with Triton

SRAM is a transistor-based memory technology that retains state while powered and provides low-latency access without periodic refresh.

On NVIDIA GPUs from Volta onward, shared memory and the L1 cache occupy the same physical on-chip SRAM on each SM and have the same access latency, despite serving different programming roles. At kernel launch, the available SRAM can be partitioned between them. Earlier Maxwell- and Pascal-generation GPUs used separate SRAM pools. section 1.4

Figure 1.8 The memory hierarchy organized by speed and size. Registers top the hierarchy as the fastest and smallest storage; global memory sits at the bottom as the largest and slowest. Shared memory and the L1 cache appear on adjacent rows to reflect their distinct programming roles; on NVIDIA GPUs from the Volta architecture onwards, they share the same physical on-chip SRAM on each SM and have the same access latency, so the figure separates them by function, not by speed.
CH01 F07 Fartale

The memory-hierarchy diagram shows shared memory and L1 cache as distinct roles sharing SRAM on Volta-generation and later GPUs. section 1.4

How GPU Programming with Triton covers it

How deep it goes
Explains SRAM sharing and launch-time partitioning, but does not detail SRAM circuitry.
Examples
Uses a memory-hierarchy diagram to show SRAM shared by the L1 cache and shared memory on later NVIDIA GPUs.
What you can do after
Understand how NVIDIA GPUs use SRAM for shared memory and L1 cache, and how that arrangement differs by architecture generation.
How it connects
Connects SRAM to shared memory, the L1 cache, and the GPU memory hierarchy.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of GPU Programming with Triton. The section links point to the parts of the book it draws on.