Fine‐Grained Memory Profiling of GPGPU Kernels

Memory performance is a crucial bottleneck in many GPGPU applications, making optimizations for hardware and software mandatory. While vendors already use highly efficient caching architectures, engineers usually have to organize their data accordingly order efficiently make of these, requiring deep knowledge the actual hardware. In this paper we present novel technique fine-grained memory profiling that simulates whole pipeline flow finally accumulates values way user retains information about potential region GPU program by showing these separately each allocation. Our simulator turns out outperform state-of-the-art models NVIDIA architectures magnitude 2.4 L1 cache 1.3 L2 cache, terms accuracy. Additionally, find our fine grained useful tool optimizations, which successfully show case ray tracing machine learning applications.