I encountered an error while using dynamic parallelism:

launch failed because launch would exceed
cudaLimitDevRuntimePendingLaunchCount

To resolve this issue, I increased cudaLimitDevRuntimePendingLaunchCount to 180*180*10, which is a rather large value.

According to the official documentation, increasing this value can have potential risks, such as:

1. Increased resource consumption.
2. Decreased performance, and memory pressure.

However, my understanding is that increasing the value should be beneficial to fully utilize the GPU’s capabilities.

Can anyone tell me what the maximum allowable value is? And what are the specific risks associated with increasing this limit?

But can someone tell me what is the potential risks when I set it to a large value?