Re: Request for MMA benchmarks & platform comparisons
- To: mathgroup at smc.vnet.net
- Subject: [mg2488] Re: Request for MMA benchmarks & platform comparisons
- From: roussel at physics.mcgill.ca (Harold Roussel)
- Date: Wed, 15 Nov 1995 01:56:55 -0500
- Organization: McGill University Computing Centre
Barnet Wagman <wagman at unbc.edu> wrote: >I'd appreciate information and comments on how MMA performs on different >platforms. I'm looking into getting a new system, and need to decide whether For performance data you could always look at Karl Unterkoffler MMA's benchmark. It is posted regularly in this newsgroup. >to stick with what I'm currently using (a PowerMac) or switch to something PPC's are really fast processors, but the MacOS is certainly not the best OS for large MMA jobs (send flames to /dev/null). >else. In particular, I'd like would appreciate >(1) Benchmarks. I gather that benchmarks were posted recently, but my >newserver has been acting up and I didn't get them. Could some one re-post >any benchmarks that are available? I'm particularly interested in comparisons >of MMA performance on the PowerMac (under System 7) vs Windows vs Linix vs >low end workstations. I'll send you via e-mail the latest copy I have. >(2) A comparison of the X-windows front-end to the Mac front-end. I've heard that the X-window front-end was now in the Notebook style (so it should look and work in the same way as the Mac). >(3) Any comments on reliability and robustness of MMA on different platforms. >(I get a fair number of system crashes and 'out of memory' occurences running >MMA on the Mac, notwithstanding the virtual memory). I've been running MMA on DOS, Win3.1, WinNT, OS/2 (all Intel systems), as well as a bunch of Unix machines including NeXTStep (680x0) and DEC (mips and alpha). As far as stability goes, DOS and win3.1 are quite similar to the Mac, not very good. OS/2, NT, and the unices are all quite stable, but I would pick NT as my first choice (in nearly two months of intensive usage, not a single crash on my machine). So, if I were to pickup a platform-OS combo, it would probably be WinNT running on a P6-200. Without doubt the fastest setup for integer-based calculations in MMA (unless you are ready to spend 40000$ bucks for a 300mhz Alpha 21164 system from DEC). For floating point, IBM's power2 systems are incredibly fast if you don't mind using the AIX OS (there's no data available for the P6 yet but I really don't see how it could match the power2 FPU). // // Harold Roussel email: roussel at physics.mcgill.ca // Ph.D. student phone: (514) 398-6506 // High Energy Physics // McGill University, Montrial, Quibec // // Using OS/2 Warp and NT 3.51. // // "...sous cette peau de mital je dipose devant toi tout ce qui me // reste; muscles, os, sang, et le coeur qui le met en mouvement." // Lancelot au Roi Arthur, dans Excalibur.