Large atmospheric models appear often in important applications. Long-range transport of air pollutants is one of the phenomena that can be studied by using such models. These models lead, after some kind of discretization, to very huge systems of ordinary differential equations (up to order of 10**6). The use of modern parallel and/or vector machines is a necessary condition in the efforts to handle successfully big air pollution models. However, this is often far from sufficient. One should also optimize the code in order to be able to exploit fully the great potential power of the modern high-speed computers. The air pollution code developed at the National Environmental Research Institute has been optimized for several different types of high-speed computers (vector machines, parallel computers with shared memory and parallel computers with distributed memory; including in the last group the massively parallel computers).
Results obtained on two computers will be discussed in this paper. The first computer is a vector processor, a CRAY Y-MP C90A computer. The second one is a parallel computer with shared memory, a POWER CHALENGE product of Silicon Graphics.