Scruff AI
41 lines
1.2 KiB
Python
41 lines
1.2 KiB
Python
# Heavy GPU stress to push power above 150W
|
|
import torch
|
|
import time
|
|
|
|
def heavy_stress():
|
|
print("Heavy GPU stress starting...")
|
|
print("Target: Push GPU power above 150W")
|
|
|
|
# Multiple large tensors to maximize memory and compute
|
|
tensors = []
|
|
for i in range(5):
|
|
size = 15000
|
|
a = torch.randn(size, size, device='cuda')
|
|
b = torch.randn(size, size, device='cuda')
|
|
tensors.append((a, b))
|
|
|
|
print(f"Allocated {sum(a.nelement() + b.nelement() for a, b in tensors) * 4 / 1e9:.1f} GB")
|
|
|
|
iterations = 0
|
|
start = time.time()
|
|
|
|
try:
|
|
while time.time() - start < 120: # 2 minutes
|
|
for a, b in tensors:
|
|
c = torch.matmul(a, b)
|
|
torch.cuda.synchronize()
|
|
iterations += 1
|
|
if iterations % 10 == 0:
|
|
print(f"Iterations: {iterations}, Time: {time.time() - start:.1f}s")
|
|
except KeyboardInterrupt:
|
|
pass
|
|
|
|
print(f"Completed {iterations} iterations")
|
|
|
|
if __name__ == '__main__':
|
|
if torch.cuda.is_available():
|
|
print(f"GPU: {torch.cuda.get_device_name(0)}")
|
|
heavy_stress()
|
|
else:
|
|
print("CUDA not available")
|