m1rcd00@fed.FRB.GOV (Bob Drzyzgula) (06/09/88)
The following is a summary of the whole two responses to my posting of a few weeks ago, concerning NAG vs. IMSL for Suns. In the end, we have decided to go with NAG, for three reasons: 1. The source of our primary statistical programming environment, the Speakeasy language, (Speakeasy Computing Corporation, Chicago), is more closely allied with NAG. 2. There was some sparse matrix stuff in NAG that wasn't in IMSL. 3. The responses that we did get indicated that we would not be making a mistake either way. Price was not the controlling factor, as it turned out. Analysis of the *final* offers from both vendors showed that NAG was only less expensive than IMSL if a perpetual license was purchased from NAG and you looked at costs out beyond 10 years. The decision, therefore, was user-driven. Thanks for your help. >From: uunet!mnetor!utzoo!utstat!utstat.toronto.edu!ruth > >We have the NAG Mark 11 package on our Sun 3's. We have NAG instead of >IMSL for exactly the reason you mentioned: price. So far, the only >complaints have been that IMSL ... does certain things either >better than NAG or which NAG does not do at all. On the whole, I >think our department would have been happier with IMSL, had it been >available for a reasonable cost, but this sort of thing is very >dependent on the users. >... >Our feeling is that the routines are written in portable fortran, and that >if there are problems compiling them, then the compiler, rather than the >code, is broken. >... >My two complaints ... you have to know that on Sun 3's, all occurrences >of type real must be changed to type double precision. Also, some of the >matrix inversion routines ... apparently don't make use of floating point >acceleration in hardware ... >From: uunet!e5274b!dquah (Danny Quah) > > I guess I would go with IMSL. For one, I've used it longer. >Also, is NAG originally from the UK? I guess I also have a bias against >the technology of that poor woe-beridden land :-) Bob Drzyzgula, Federal Reserve Board, Washington, DC 20551; uunet!fed!rcd