mccalpin@loligo.uucp (John McCalpin) (03/24/89)
In article <1064@bayes.ptolemy.arc.nasa.gov> self@ptolemy.arc.nasa.gov (self) writes: > >(4) Are there any rumours about faster floating-point options > > for the NeXT? The 68882 is not fast enough for serious work. > You can typically double or triple floating point performance on the > 68881/2 by using an inline math library with GCC. > Matthew Self That is a bad sign, since I was basing my estimate of the 68882 performance on the hardware limitations! If GCC only gives one-third to one-half of this, then it is going to be a real dog.... The numbers that bother me are the number of clock cycles required by the 68882 to do FP adds or multiplies. They are in the 40-50 range. At 25MHz, that means an absolute limit of about 0.5 MFLOPS. The DECstation 3100 (which someone else mentioned) and which is in the same price range (with university discounts) gives peak performance of over 4 MFLOPS on some Livermore kernels (32-bit), and can sustain 1.5-2 MFLOPS on real codes.... ---------------------- John D. McCalpin ------------------------ Dept of Oceanography & Supercomputer Computations Research Institute mccalpin@masig1.ocean.fsu.edu mccalpin@nu.cs.fsu.edu --------------------------------------------------------------------