welcome.exe

SYSTEMS THAT RUN
CLOSE TO THE METAL

hi! im atharva, going through life one thread at a time.

See my work
chip_viewer.exe
C:\Projects

TinyGPU

A custom SIMD vector accelerator wired into a RISC-V core — running on real FPGA silicon, not simulation.

15.61x speed up compared to CVA6

Verilog RTL Computer Architecture Vivado Instruction and data parallelism
View repo

CUDA GEMM Optimizations

Hand-written CUDA GEMM kernels, iteratively optimized against cuBLAS using NSight Compute profiling.

Naive → optimized, benchmarked in GFLOPs across matrix sizes

CUDA C++ NSight Compute
View repo
blog.exe
readme.txt

I believe hardware is the future for accessible, intelligent systems and I want to work on improving the current capabilities chips that handle AI workloads.

  • FocusRTL Synthesis
  • NowGPU Design and Implementation
  • ToolsVivado · Nsight Compute · Vitis
contacts.exe

Atharva A. — Comp Arch ·