Under the premise of fully custom hardware, we explore more efficient choices of prime numbers in STARKs. In particular, we show that designing trace tables over specific primes that have previously not been considered leads to smaller traces for zkVMs (50% reduction in some settings) and faster provers. To further accelerate the proof generation, we also introduce a new approach of evaluating quotient polynomials for STARK tables in a massively parallel way. This technique is highly generic and can be applied to a wide range of trace tables.