Fifteen years of GPU and HPC performance engineering — CUDA, OpenMP, MPI, SYCL. PhD, IIT Bombay. I've spent my career on one problem: taking code that is correct but slow, finding out exactly why, and fixing it.
I do this work through Siddhivinayak Advanced Computing Labs (SVACL), and I teach it in public every week.
[ Book a 30-minute call → ] write to mandar@svacl.com
[ See the GPU Performance Audit → ] https://svacl.com/gpu-performance-audit/
PhD, IIT Bombay · B.E. Computer Science, Walchand College of Engineering, Sangli · Ex-Nvidia, Ex-Intel, Ex-CDAC, Ex-IITB · 16+ years in production HPC
Domains:
The domains change. The bottlenecks don't. It is almost always memory, it is almost never where the team expected, and it is almost always fixable.
Taught and mentored at
Also Mentored teams in following hackathons
The fastest route is a 30-minute call. Free. Bring a profiler trace, a repo, or just a description of what's slow. You'll leave with an honest read on whether there's headroom worth chasing — and if there isn't, I'll say so.
Write to : mandar@svacl.com