Should You Upgrade to 500MHz PC?

We compared a 500-MHz Pentium III PC to a 350-MHz Pentium II. Some tests showed only a negligible improvement, while others ran substantially faster.

Dave Wilson and John Hanks, National Instruments, Austin, TX -- Test & Measurement World, 10/15/1999

Each new PC introduction brings with it a faster processor, but how does that faster processor—often accompanied by more memory—affect test applications? We wanted to know, so we ran some speed comparisons between a typical 500-MHz Pentium III (PIII) PC and a typical 350-MHz Pentium II (PII) machine. While some applications ran 80% faster on the 500-MHz system, others sped up by only 13%. So, the value of upgrading will depend on your application.

In our tests, we used a Dell Dimension 500-MHz PIII XPS T500 with 128 Mbytes of memory and a Micron Millennia 350-MHz PII with 64 Mbytes of memory and then with 96 Mbytes of memory. We wanted to see if any performance improvements were due to memory rather than to processor speed. We tested the computers in both data-acquisition and image-processing applications.

For the data-acquisition test, we measured how many times per second a 200 ksample/s data-acquisition card could sample data while the PC performed fast Fourier transforms (FFTs). The FFTs were 1 ksample and 16 ksample in length. We measured the processing time for both FFTs with and without plotting the waveforms and their spectrums on the PC screens. All tests used the same data-acquisition techniques—double buffered, bus-mastered, scatter-gather input. Stated simply, the PC’s processor didn’t have to transfer the data from the data-acquisition board to the computer’s RAM. The card handled that. Therefore, the CPU could spend its time performing the FFT calculations and producing the plots.

Table 1 shows the results of the data-acquisition, FFT, and plotting operations with both computers. In some cases, the 500-MHz PIII performed considerably better. In other cases, the differences were negligible. For the application where the system performed but didn’t plot the 1-kpoint FFTs, the 500-MHz PIII processed nearly 300 more FFTs per second than the 350-MHz PII. Even when the system plotted the 1-kpoint FFT, the 500-MHz computer was considerably faster.

When we increased the FFT size to 16 kpoints, the speeds dropped and so did the difference between the computers. The faster computer processed just three more FFTs per second than the PII machine—only a 13% increase. If you need to perform large, 16-kpoint FFTs, you may wish to upgrade, unless you need to plot the results. Then, replacing a 350-MHz, 64-Mbyte PC may not result in increased productivity. If you perform moderate-sized FFTs, though, an upgrade can pay dividends regardless of whether you plot the results.

We also wanted to know if increasing the memory on the 350-MHz machine from 64 Mbytes to 96 Mbytes would speed up data acquisition. The test results in Table 2 indicate that performance improved negligibly at best.

For the benchmark test, we ran the program with and without image display. The display size was 512x512 pixels. We acquired the images with a high-speed digital camera and a 32-bit digital image-acquisition board with 16 Mbytes of onboard memory.

In the second application, we performed location analysis using pattern matching—a typical vision tool application. Using pattern-matching software is a simple two-step process—learn and search. For the learn step, you create a “golden template” representation of the object. You then search for images that match the template. In our test, we searched for one match of the golden template in each acquired image. By intelligently selecting only a subset of the pixels in the template image, the pattern-matching algorithm quickly finds the approximate region where an object is located.

We used a correlation algorithm to achieve a high-accuracy match between the template and the object. The pattern-matching software uses a 94x52-pixel template. The software searched for matches of the template in 512x512-pixel images. We ran two pattern-matching tests. In the first test, the system searched for templates that weren’t rotated. The objects were only translated in the x direction and in the y direction. Pattern-matching software can quickly find the object you are searching for if it is not rotated. In addition, pattern-matching software can locate objects that are scaled (varying in size by 610%), out of focus, and poorly illuminated. In our application, the objects were the same size as the template, in focus, and properly illuminated.

For the vision applications, the 500-MHz PIII showed performance gains ranging from 61.9% to 80.5% more than the 350-MHz PII machine (Table 3). The gauging application with the 500-MHz system inspected a remarkable 299.3 images/s. In each acquired image, the software inspected 11 pins; that’s 3292 pins/s. The location analysis search times were much shorter on the 500-MHz PIII. The computer found the object in any orientation in 222 ms, a rate of 4.5 images/s. That’s 69.8% faster than the 350-MHz machine.

In all cases, the 500-MHz PIII showed at least a 60% improvement over the times for the 350-MHz PII. Therefore, we conclude that for inspection applications, the productivity gains from the increased speed of the 500-MHz system justify the cost of upgrading. T&MW

Dave Wilson is director of data-acquisition marketing, E-mail: dave.wilson@natinst.com. 

John Hanks is vision product manager at National Instruments, E-mail: john.hanks@natinst.com.