// close to the metal · measured · explained
Fakhir Shaheen
C++ engineer and CS professor, going deep on GPU performance and ML kernels.
Now
currently→CPU path tracer — multiple importance sampling done
next→BVH construction + profiling pass
sgemm 4096³ vs cuBLAS71%▲ +6% this week
MIS render, 512spp41.2s▲ 2.3× faster than naive
$