Skip to content

Runs with large numAmps per GPU fail #719

@bvillasen

Description

@bvillasen

When I try to run more than one numAmps=2^31 per GPU, I get the following error:
terminate called after throwing an instance of 'thrust::THRUST_200805_400200_NS::system::system_error'
what(): parallel_for failed: hipErrorInvalidConfiguration: invalid configuration argument

This happens when I try, for example:

  • 32 qubits on one GPU
  • 33 qubits on 2 GPUs
  • 34 qubits on 4 GPUs

I'm running this on a GPU with 192 GB of memory, so memory capacity should not be an issue, as the above runs would require 128 GB per GPU.

Note that I can run the following:

  • 32 qubits on 2 GPUs
  • 33 qubits on 4 GPUs
  • 34 qubits on 8 GPUs

So, it works as long as numAmps<=2^31 per GPU.

I suspect the issue is that somewhere the numAmps per GPU overflows int32_t or uint32_t

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions