SPICE sets up a matrix equation representing the Newton-Raphson iterations at each time point and solves it. The critical parameters of performance are the floating point unit and the cache organization. The cache organization is critical because the temporal access pattern for matrix entries is non-local resulting in each entry being read into and knocked out of the cache several times for a large matrix. Figure 2 shows the relative performance for various architectures. 21064_182 generally gave the best performance. Sparc20_100 performs better than i586_120 probably due to a faster L-2 cache. R4400_60 ended up being the slowest of the lot, consistently performing worse than Sparc10_50, which has smaller clock rate and slower L-2 cache. These two observations seem to indicate a better floating point unit for the Sparc machines.