I am aligned this looks like final proof that these 4 counters are silently dying. Can't help much further
==================================================
In terms of buffers, here is my understanding (sharing it so that people can correct or reference it later):
- you have onboard memory to store acquired samples (or generate samples... well, the right term used in datasheets is FIFO). For Analog Input, it tends to be like 2K/4K per ADC. For counters, I see 127 samples per counter
- PC memory buffer, the one that we can size from input_buf_size. The onboard FIFO samples are DMA copied to it
My card goes only to 1Mhz, 1ADC so I have no issue with the FIFO to PC memory copy... while this could be an issue on your side.
On the contrary, I am doing a lot of computations and graphics on the PC so I had to correctly size the PC memory buffer and readout rate to be able to handle callbacks much delayed by CPU load peaks. But again, I am far from your rates