SPONSORED LINKS
Randy, As you might be aware, I have some strong opinions on the benchmarking of PCs, in part developed as a result of having performed a similar CPU upgrade on my desktop PC (see www.math.hawaii.edu/~ross/cpu.html). Let me preface them by saying that (1) I think its great that PEP is finally doing this (though they probably should have done this a year ago, before many of us just changed to a new system), and (2) despite my skepticism of the price/performance ratio, I think you'll be surprised at how much faster the AMD is at some tasks (I was, on my desktop). Now, for some of my opinions: (1) There are some genuinely useful benchmark resources on the web; I recommend: (a) Intel's very nice site, http://www.intel.com/procs/biz/vol1no1/feature/index.htm (b) the comp.benchmark FAQ (an old version of which is at http://www.inf.ufrgs.br/~sagula/bench.FAQ.html) (c) and my favorite, http://www.sysopt.com/bench.html (the FAQ might be available here as well). (2) Specific benchmark, or even operating system, is not as important as running a wide spectrum of benchmarks, from FPU-specific (like whet/dhrystone) and CPU-specific (ZD's CPUmark; Landmark) to systemic (Winstone; Quake FPS). The user can then decide which benchmark is closest to his/her general use, and adjust accordingly for OS. A program like winbench - which measures the components of a system separately - can be misleading, BTW, since the user is probably not as good at mentally integrating the test components as he thinks he is! (3) Winstone: This absolutely should be run, if possible. I don't know whether it actually compensates for CDROM speed (i.e., by only initiating timing *after* the program is read from the CDROM). I have a suspicion that Winstone97, in trying to rework true parallelism, might have to access the CD while doing other tasks. I liked Winstone95, but am very leery of Winstone 97. Still, if possible it *should* be run. (4) Configuration: W95 benchmarks should be run in a couple of configurations: minimal (performance enhancements like caching disabled) and normal (what you usually use). As for graphics, I suggest 256 color instead of hi-color (since many 701 owners have passive matrix). BTW, I believe that the APM features can be used to crank your DX4-75 down to DX2-50 speeds. You might want to do this too, so that DX2-50 701 owners can make a reasonable comparison. (4) Quake and Doom: I like these as tests of the CPU/memory/Video/bus interaction. Don't bother to do both. (5) FPU: You'll find that the AMD performs brilliantly on *any* FPU-intensive test, so you might as well just use the old standards, whetstone and dhrystone (these are included in the Windows Magazine Wintune benchmark [which any W95 user should have anyway, as a system tuneup app], as well as N. Juffa's nice CompTest benchmark). One lister thought it wouldn't do as well on FPU tasks, but in fact the clock-quadrupling more than makes up for the fact that Intel has the fastest FPU microcode. (6) Statistics: If you have a statistics package on your machine (Excel might suffice), do some linear regressions with both small datasets and large ones. The small ones will test the CPU/FPU/memory interaction, while large ones will include disk and cache management. (7) Be careful, take good notes, write you results up carefully, and publish them as an article for which you get paid. No reason to do all this work for free! Plus, then you can deduct the upgrade cost. As an alternative benchmark, my colleague across the hall suggests installing Windows 98 beta, then clicking 'My Computer'. Good Luck! - David R.