ss60f@sdcc4.ucsd.edu (Nathaniel Beck) (10/18/89)
I have a 25Mh 386 clone (with 387, 4MB ram). I do computationally intensive analysis on relatively small amounts of data. My setup (DTK) has a 64K cache which can be upgraded to 256K. The cost of the 256K 25ns chips seems high, so I want to know whether I will notice a performance difference with the upgrade. Would really appreciate experience (either practical or benchmark). Thanks in advance Neal Beck Dept. of Pol. Sci. UCSD beck@ucsd.edu
rob@prism.TMC.COM (10/19/89)
I'd doubt that going from a 64K to a 256K cache would be worth it in your case. Increasing cache size beyond a certain point is a classic diminishing-returns practice, at least on single tasking systems like DOS (which I'm assuming is what you use). The cache increase would probably gain you about 5% on CPU bound tasks. If the work you're doing is floating-point intensive, the increase might be a bit less, since such problems tend to be bound more by compute time than memory access time (which is what a cache improves). Finally, since you say you work with fairly small amounts of data, it's very possible that the 64K cache is entirely adequate, so that expanding it would offer only a negligible performance increase.