Max size of CUDA arguments

I was trying to figure out what the maximum size of arguments you can pass to a CUDA kernel is, and I found this thread from 2010:

I was wondering, has the 256 byte limit increased since then? I’d love to be able to pass more in. Thanks!

Also, I’m on a Titan X (Pascal) if that matters.

From the CUDA Progamming Guide (

As I recall, the 256-byte limit applied to GPUs with compute capability 1.x, which used different memory for kernel argument passing and which is no longer supported by CUDA.

Awesome, thanks!