Error: DriverError(CUDA_ERROR_NOT_FOUND, "named symbol not found") when loading cast_f32_bf16

[jefftantan]

trafficstars

Win11 Error: DriverError(CUDA_ERROR_NOT_FOUND, "named symbol not found") when loading cast_f32_bf16

[yinqiwen]

it's only available when __CUDA_ARCH__ >= 800 cast kernel

[dommyrock]

@yinqiwen Yep , correct. @jefftantan was probably using 2000 series nvidia chips or older (like i was when i got the same error on my rtx 2070 😞 )

The 16-bit __nv_bfloat16 floating-point version of atomicAdd() is only supported by devices of compute capability 8.x and higher.

You can also find more details in the related issue above.

Hope it helps :)