10-10-2026, 12:26 AM
Not sure I agree there. I get why that seems right, but on A6000 the bottleneck felt elsewhere. Worth testing the simpler explanation before adopting the more interesting one.
measure twice, benchmark three times