Uploaded August 2022 | Updated September 2026, 1 week ago

---

Here we use Sushi Roll (my custom, low-noise kernel) to benchmark memory access times from core to each node on a 96-core, 8-node system.