No, it's 5 years old or so.
> Can you link to the relevant documentation?
https://developer.nvidia.com/blog/cuda-dynamic-parallelism-a...
> and doesn’t overall affect perf by much
Oh, but it very easily does. Slow I/O to the device, no parallelization of I/O and compute, etc - are all extremely easy pitfalls to walk into.
No, it's 5 years old or so.
> Can you link to the relevant documentation?
https://developer.nvidia.com/blog/cuda-dynamic-parallelism-a...
> and doesn’t overall affect perf by much
Oh, but it very easily does. Slow I/O to the device, no parallelization of I/O and compute, etc - are all extremely easy pitfalls to walk into.