PANTHEON
Why Pantheon?
Pantheon is a cross-platform (CUDA/ROCm) stress testing tool designed to isolate and hammer specific GPU subsystems. Unlike generic benchmarks, Pantheon targets specific silicon limits to expose hardware degradation, thermal throttling, and architecture bottlenecks.
-
VRAM & Infinity Fabric --- Target Memory/GDDR memory and interconnects with aggressive crosstalk patterns to detect bit flips and signal integrity loss.
Test:
memory_write_agg,memory_tsv_thrasher -
Physical Matrix Cores --- Push hardware Tensor and Matrix cores to their absolute thermal limits using universal FP16/WMMA intrinsics.
Test:
mma_virus,tensor_virus -
VRM Transients (dI/dt) --- Oscillate maximum load at 10Hz to induce high current transients, testing your power supply and voltage regulators.
Test:
pulse_virus,voltage -
MMU & TLB Avalanche --- Force near 100% Translation Lookaside Buffer misses with random page-boundary jumps to choke hardware page-table walkers.
Test:
tlb_avalanche -
ECC & RAS Validation --- Continuously read pristine patterns to expose the latency jitter of active ECC scrubbing and silent data corruption.
Test:
ras_validator -
Hardware Scheduler --- Spam micro-kernels across 64 concurrent streams to force the GPU dispatcher into multiplexing mode and trigger context switch locks.
Test:
scheduler
Quick Start
# Install required dependencies
sudo apt-get update
sudo apt-get install python3-tk python3-pip
sudo apt-get install nvidia-cuda-toolkit (replace with hipcc on ROCm devices)
sudo pip install numpy psutil pandas openpyxl cmake customtkinter pyinstaller uvicorn
# Run the full suite (30 seconds per test)
./pantheon --test all --duration 30 --verify