Even GPUs, which have arguably far more predictable instruction mixes, struggled massively to get a useful utilization on VLIW architectures – AMD tried two different ones from 2006 to 2011 before they finally gave up on the concept and started using RISC architectures like Nvidia had been using all the time.