Skip to content

The b2b_gemm Example Fails Tests on A100 #144

@haruhi55

Description

@haruhi55

When I run the b2b_gemm example on A100, it raises the following errors:

[16, 16, 16, 16], batch = 1, passed.
[16, 32, 16, 32], batch = 1, passed.
[32, 64, 32, 64], batch = 1, passed.
[64, 64, 32, 64], batch = 1, passed.
[256, 128, 64, 64], batch = 1, passed.
[1024, 1024, 128, 128], batch = 1, passed.
[16, 16, 16, 16], batch = 2, passed.
terminate called after throwing an instance of 'thrust::THRUST_200301_800_NS::system::system_error'
  what():  trivial_device_copy D->H failed: cudaErrorIllegalAddress: an illegal memory access was encountered
[1]    972511 IOT instruction (core dumped)  ./fused_gemms

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions