Question about binomialOptions example.

Just tried to play with the binominal Options example in SDK. In the source code, it has following settings:
NUM_STEPS 2048
CACHE_STEP 512
CACHE_DELTA 32
The GPU is x20 faster than CPU run.

My question is, if NUM_STEPS=64, what value is the best for CACHE_STEP and CACHE_DELTA to get a maximum acceleration?