GoKawiilRTX 4090 teardown traces how STG.E writes results back to memory
A technical deep-dive follows the STG.E instruction—the global memory store operation in a simple vector-add CUDA kernel—through an RTX 4090's pipeline, from register read through the load/store unit, coalescer, L1 cache, crossbar, and L2 slice, down to DRAM. The author benchmarks how frequently a single warp can issue store instructions, finding a new STG.E can be pushed through the register file, LSU, and coalescer roughly every 6.1 cycles (about 2.3 nanoseconds at 2.6 GHz).
GoKawiil's interpretation of the reporting above, not reported fact.
This kind of instruction-level tracing exposes the real bottlenecks in GPU memory writes that abstract programming models hide, giving low-level performance engineers concrete cycle counts to reason about kernel throughput. As AI and graphics workloads increasingly hinge on memory bandwidth rather than raw compute, understanding these store pipelines helps developers and hardware designers identify where optimization actually pays off.
- STG.E stores require reading three register rows: two for the address and one for the data.
- A warp can issue a new store roughly every 6.1 cycles (~2.3 ns) at 2.6 GHz on an RTX 4090.
- The store instruction traverses the same hardware path as loads—LSU, coalescer, L1, crossbar, L2, DRAM—but in reverse for writing data.
Programming Massively Parallel Processors (Hwu, Kirk, Hajj) — If tracing an STG.E through L1, the crossbar, L2 slices and DRAM sounds like fun, this book is the canonical deep dive into CUDA's memory hierarchy and how kernels like a vector add actually map onto GPU hardware. It pairs perfectly with the article's instruction-level walkthrough, giving you the model behind coalescing, caches, and write behavior.
See Programming Massively Parallel Processors (Hwu, Kirk, Hajj) on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.Source: blog.doubleword.ai — Fergus Finn, 2026-09-08
Published there as: “What happens when a GPU writes memory”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.