09-16-2026, 01:22 PM
I tested this tonight. On mini PC with PyTorch, I saw basically the same behavior. The surprise was that memory pressure showed up before throughput collapsed.
throughput matters, but tail latency matters too
