# How feasible is a preimage attack or speedups to brute force a substring of MD5?

Beyond the birthday problem analysis how feasible is a preimage attack or speed up brute force on a substring of MD5. For instance let's say from the 128 bits I decide to strip 24 bits, the first four and last 20 of a generated MD5.

How can you attack or fast generate a preimage that will collide on the same substring.

ps.It is known MD5 is broken, some hardware can compute billions of MD5 hashes per second. The intent of the question is theoretical understanding.

Right now, the cheapest known algorithms for (partial) preimage search are generic algorithms, costing $${\sim}2^b/t$$ to compute a preimage for any of $$t$$ target $$b$$-bit strings in time $${\sim}2^b/t^3$$ if parallelized $$t^2$$ ways—here $$b = 128 - 24 - 4 - 20 = 80$$.