Cupy does’t used automatically shared memory

by

I recently start using Cupy in Python and I notice something strange. When I run

a = cupy.random.rand(int(30e9/8))

by using one “RTX 3090” then the dedicated 24 GB memory gets full and the remaining part goes automatically at shared memory without any problem. But when I run it using one “RTX 4090 trinity” then I get the error

File cupy\cuda\memory.pyx:1387 in cupy.cuda.memory.SingleDeviceMemoryPool._try_malloc

OutOfMemoryError: Out of memory allocating 30,000,000,000 bytes (allocated so far: 0 bytes, limit set to: 32,212,254,720 bytes).

despite I have the necessary shared memory. Does anyone may knows why this happening ?

Thanks,
Vassilis

I would like to understand the problem in order to determine if I can fix it or not

Leave a Comment