We describe the first parallel implementation of an adaptive
particle-particle, particle-mesh code with smoothed particle hydrodynamics.
Parallelisation of the serial code, ``Hydra'', is achieved by using CRAFT, a
Cray proprietary language which allows rapid implementation of a serial code on
a parallel machine by allowing global addressing of distributed memory.
The collisionless variant of the code has already completed several 16.8
million particle cosmological simulations on a 128 processor Cray T3D whilst
the full hydrodynamic code has completed several 4.2 million particle combined
gas and dark matter runs. The efficiency of the code now allows parameter-space
explorations to be performed routinely using $64^3$ particles of each species.
A complete run including gas cooling, from high redshift to the present epoch
requires approximately 10 hours on 64 processors.
In this paper we present implementation details and results of the
performance and scalability of the CRAFT version of Hydra under varying degrees
of particle clustering.