Here I want to concentrate on the computer's performance with CAD software. When Alienware asked what kind of benchmarks I'd use, I replied I'd have to make up something.
A goal of benchmarking is to reflect real-world usage: will this piece of hardware (or software) make me more efficient. Some benchmarks try to be very real; I don't try that at all, because (1) computers are no longer slow, and (2) my computer spends most of its time waiting for me to respond.
As some readers may recall from my days as technical editor at 'CADalyst' magazine (1985-1991), I concentrated on benchmarking raw power and precision. Back then, hardware was so slow it was easy to time it; I wrote several AutoLISP routines to help automate the timing process.
(A problem occurred when AutoCAD switched from DOS to Windows: the LISP timing routines were no longer accurate. Each time the routine ran, it produced a different timing. I determined the problem was due to Windows doing multitasking, unlike DOS; AutoCAD no longer had 100% attention of the CPU. Fortunately, the transition to Windows coincided with me no longer doing hardware reviews.)
Fourteen years later, and I need to think up a benchmark. I knew I would use the 2.3MB OilModule.dwg file, because it is (1) dog slow to display with its rendered solid models; and (2) a sample drawing that anyone can access.
I opened it in AutoCAD 2005 to determine what I could time. Regens were pretty quick, but switching layout tabs was slow. So, here's the benchmark:
(Before doing timings, restart the computer to flush all caches.)
1. Time to launch AutoCAD 2005
2. Time to open OilModule.dwg
3. Time to switch between the five layouts in three modes:
a. Software accelerated
b. Hardware accelerated
c. Cached layouts
I wrote a script to automate the layout switching:
-layout s piping
-layout s "back & left side"
-layout s "front & right side"
-layout s isometric
-layout s model
And I employed AutoCAD's Options command to make life difficult: I turned on all render options, turned off adaptive degradation, turned on layout regen with switching tabs, and maximized dynamic tesselations. The first test was timed with software acceleration of graphics, and the forced regeneration of layouts. (Timings are rounded to the nearest second.)
1. Launch = 12 secs
2. Open = 13 secs
3. Switch layouts
a. Software = 59 secs
b. Hardware = 1:40 min:sec
c. Cached = 45 secs
"Not that fast," I thought to myself. Maybe this computer is a dud, as happened to Alienware last year with one of their models. I was surprised that (1) hardware acceleration is _much_ slower than software; and (2) caching layout views was not appreciably faster than forced regenerations. The nVidia graphics board is capable of all kinds of optimizations, but I didn't have AutoCAD optimization settings.
For comparison, I timed my desktop computer under the same conditions: it's a 2.4HGz Pentium IV (1.42x slower) with 1GB RAM and an ATI 8500LE graphics board.
1. Launch = 28 secs (2.3x slower)
2. Open = 13 secs (same -- hmmm)
3. Switch layouts
a. Software = 1:26 Min:sec (1.45x slower)
b. Hardware = crashed repeatedly (hmmm)
c. Cached = 1:02 min:secs (1.28x slower)
Except for the AutoCAD launch speed, the timings were in line with the relative speeds of each computer's CPU. I am guessing the Pentium CPU is more powerful than the nVidia GPU.
So, what is the point to ponying up US$3,647 for a notebook computer? You'd buy it if you needed a mobile workstation. It's a computer that's as powerful as the equivalent desktop, but can be hauled around to client sites. After work, it doubles as an entertainment center, whether for watching DVDs or making them.