20 lines (16 loc) · 607 Bytes

GPU Benchmarks ⚡

Comprehensive GPU benchmarking and profiling suite.

Benchmarks

Memory Bandwidth — HBM/GDDR read/write/copy throughput
Compute Throughput — FP32/FP16/INT8 FLOPS measurement
Kernel Latency — Launch overhead and execution time
Shared Memory — Bank conflict analysis
Occupancy — Block size vs occupancy profiling

Usage

python benchmark.py --suite all --device 0
python benchmark.py --suite memory --iterations 1000
python benchmark.py --suite compute --dtype fp16 fp32

Output

Results saved as JSON + HTML report with charts.