Re: Mathematica 3.01 is slower than Mathematica 3.0 on PowerMacs!
- To: mathgroup at smc.vnet.net
- Subject: [mg9775] Re: Mathematica 3.01 is slower than Mathematica 3.0 on PowerMacs!
- From: NOHAMcrose at c2.telstra-mm.net.au (Colin Rose)
- Date: Tue, 25 Nov 1997 00:07:29 -0500
- Organization: Theoretical Research Institute
- Sender: owner-wri-mathgroup at wolfram.com
charles.bouldin at nist.gov (Charles Bouldin) wrote: > http://fampm201.tu-graz.ac.at/karl/timings30.html. If you look at > reference 89 on that page, you will see a report that Mathematica 3.01 > is 15% slower than Mathematica 3.0. This is exactly the same result > that I reported to WRI within days of the release of 3.01. Here are 2 examples of the slowdown, as measured using the standard Karl Unterkofler's benchmark test: v3 v3.0.1 _________ _________ Timing[N[Pi, 13000]][[1]] 2.6 secs 3.65 secs Timing[10001!][[1]] 0.86 secs 1.73 secs [ Timings are on a PowerMac 8600/200/1meg L2 cache under OS8 with extensions ON. Switching extensions off improves performance by about 5%. ] An Oddity: is there really a slowdown ? Or is Timing just wonky ? _________ The above results are the times as measured and reported by Karl Unterkofler's benchmark test. This uses a Block structure which prepares some calculations first. One of these is DSolve. For instance: v3.0.1 In[2]:= Block[{}, DSolve[D[y[x], {x, 2}] + y[x]*E^x == 0, y[x], x]; {Timing[N[Pi, 13000]][[1]], Timing[10001!][[1]]}] Out[2]= { 3.6 Second,1.75 Second} __ Now, if this calculation is repeated (from a new kernel) but without the DSolve[], the Timing results should NOT change. But just look what happens: v3.0.1 In[2]:= Block[{}, {Timing[N[Pi, 13000]][[1]], Timing[10001!][[1]]}] Out[2]= {2.85 Second,1.06667 Second} The timing results have changed dramatically, and they should NOT do so. It is no longer clear whether or not there really is a timing slowdown under v3.0.1. Unfortunately, the Timing function itself appears quite wonky. If the timing fn is not reliable, then it questions the suitability of benchmark tests under Mathematica, both across platforms, and across versions of the same program, which is rather a pity. Cheerio Colin -- Colin Rose tr(I) - Theoretical Research Institute ______________________________________ NOHAMcrose at c2.telstra-mm.net.au http://www.usyd.edu.au/su/tri/