In 1997, the world was treated to the luxury of not one, but two 200 MHz CPUs, each championed by one of the two major desktop competitors of the day, Wintel and Apple. In 1995, the Windows/Intel juggernaut had introduced the 180 MHz Pentium Pro, alongside their pre-existing full 32 bit operating system, Windows NT 3.5. In 1997, the Pentium Pro was upgraded to a full 200 MHz, complimenting Windows NT’s 1996 upgrade to Windows NT 4.0. Apple was not far behind, in 1997 introducing the Power Macintosh 7300, powered by a 180 MHz or a full 200 MHz PowerPC 604e CPU. The stage was now set – two heavy duty 200 MHz CPUs duking it out for consumer attention and dollars.
This was BIG news at the time. The original Macintosh had debuted in 1984, featuring an 8 MHz Motorola 68000. Over at the rival PC camp, the original IBM PC had come to market in 1981, featuring a 4.77 MHz Intel 8088 (a slower eight bit bus version of the more powerful 16 bit 8086 processor). By late 1993, Macintosh computers had evolved to the Quadra 660AV and Quadra 840AV lines, employing then cutting edge 25 MHz and 40 MHz Motorola 68040s, respectively and running under the capable and well rounded Mac OS 7.x line of operating systems. By now the PC camp had evolved to Windows and Intel, or Wintel, and was serving up to 33 MHz 80486 processors powering DOS and early versions of Windows. Over the course of approximately 10 years, CPU clock rates, the consumer-accepted measure of “speed”, had only increased by a factor of 4 or so.
Between 1993 and 1997 however, processor clock speeds rocketed ahead from 25 MHz all the way up to 200 MHz, almost a full order of magnitude in just 4 years. 200 MHz was a landmark. This was BIG news. By 1997, both the Apple and Wintel camps were fielding 200 MHz monsters, competing head to head for the lucrative personal computing market.
This was truly a clash of the titans. Mated to a full 32 bit operating system such as Windows NT, the Intel Pentium Pro promised unparalleled power and speed. The PowerPC 604e based Power Macintosh 7300 featured the already 32 bit clean Mac OS 7.6 and described itself as incredibly powerful and possessed of nearly limitless capabilities.
Cutting through the marketing hype was difficult. Which was truly faster, the lightning fast 200 MHz Pentium Pro, running under Windows NT 4.0, or the brawny 200 MHz PowerPC 604e, under the able supervision of Mac OS 7.6?
I set out to benchmark these two and discover the answer, using some less than rigorous and perhaps somewhat ad hoc methods. OK, perhaps “benchmark” is an overly ambitious word in this context, since I did not have identical operating system software on each of the two CPUs. Hence, some degree of variability had to be accepted in the results. I attempted to cut through this difference by focusing on CPU intensive tasks that would be less influenced by factors such as OS performance, graphics capabilities and disk I/O. In this case I settled on two test cases: rendering a very large JPEG image, and creating a full set of thumbnails for a large folder of images.
I was helped in this task by the incredible good fortune of having both a Windows and a Mac OS version of two excellent thumbnail and viewer programs: ThumbsPlus and ACDSee. I used these two programs, available on both platforms, as the basic test vehicles.
It is instructive of the computing power of today’s desktops and laptops that we have forgotten how mathematically intensive decoding a JPEG image is. No matter how large the image, it seems to simply snap open. Back in 1997, this was very definitely not the case. A large JPEG could take tens of seconds to render, and because of this, decoding time for a large JPEG was a useful measure of CPU power. As a consequence of this, at the time I had a particular image that I used as my benchmark of how much faster a new computer was than its predecessor. I did this by measuring how much more quickly the chosen JPEG rendered on the new platform vs. the older one. This seemed a good measure of raw CPU horsepower back then, and it still seemed to make sense for the tests at hand for this post.
Let’s look at the two contenders. From the Wintel side, powered by a 200 MHz Intel Pentium Pro with 1 MB of onboard L2 cache, was a custom built PC running Windows NT 4.0 and featuring 512 MB of RAM, a 120 GB IDE (ATA/100) hard drive and an nVidia Riva TNT 64 graphics card with 16 MB of video RAM.
From the Apple side, powered by a 200 MHz PowerPC 604e, was the Power Macintosh 7300/200 with 256K of external L2 cache, and running Mac OS 9.1 in 544 MB of RAM. The 7300 was equipped with an 18 GB SCSI hard drive, a 120 GB IDE (ATA/33) hard drive, and built in video acceleration with the full complement of 4 MB of video RAM.
Astute readers will note the large disparity in disk speeds between the two computers, and wonder if that might skew the results in favor of the computer with the faster disk. Other readers will note the significant deltas in cache layouts and wonder the same. Lets look at each of these factors.
The Power Macintosh 7300/200 had two hard drives, one SCSI and one IDE, both of which were significantly slower than the hard drive on the Pentium Pro PC. To understand whether this made a difference, I tried all of the thumbnailing and decoding tests twice on the Power Macintosh 7300, once from the slower SCSI disk (benchmarked at 8.1 MB/S) and once from the much faster IDE disk (benchmarked at 31 MB/s, or nearly 4X faster than the SCSI disk). As expected, there was no measureable difference in the times turned in between the two disks – the tests were completely CPU bound, meaning that relative disk speeds between the two systems were effectively immaterial. So, no advantage accrued to the Pentium Pro system due to its markedly faster hard drive.
What about the cache layouts? The PowerPC 604e integrates 64 KB of onboard L1 cache, split evenly between instruction and data caches, at 32 KB each. The Pentium Pro has only 8 KB of integrated L1 cache, that is shared by both instructions and data. The 604e’s larger L1 cache would seem to give it a clear advantage. However, offsetting this was the fact that the Power Macintosh 7300/200 had only 256 KB of external L2 cache, while the Pentium Pro featured 1 full MB of L2 cache, integrated right into the chip packaging and accessible by a separate, dedicated backside bus. The net result is an L2 cache that is dramatically faster than the L2 cache of the 7300/200.
So, the Power Macintosh 7300/200 has significantly more L1 cache, but by contrast, the Pentium Pro had significantly more, and significantly faster, L2 cache. Did this balance things out from a performance perspective? We shall see in a moment, but in the meantime, it sounded like a fair fight, and as it turned out, it was.
…or was it? It will come as no surprise to learn that I am not the first person to think of pitting the 200 MHz PowerPC 604e against the 200 MHz Pentium Pro. I unearthed numerous comparison tests, from Byte Magazine, Mathematica, Richard Soberka and others. The general finding was the PowerPC 604e was the standout performer in integer math (Byte reported that it was 81% faster than the Pentium Pro) and the unconditional winner in floating point math (where Byte rated the PowerPC 604e at 26% faster than the Pentium Pro). However, most tests found that the Pentium Pro excelled in memory intensive tasks such as Photoshop filters, as a direct result of its superior L2 cache. Since real world work on a computer is often a combination of memory accesses and math, you could almost guess that the two machines would come in about the same when challenged with actual day to day work, and this seemed to be the case.
One major problem did “taint” these results somewhat, unfortunately. While I had both Macintosh and Windows version of ThumbsPlus and ACDSee, they were not the same version between the two platforms. For ThumbsPlus, I had v3.2 for the Macintosh (a beta that was never formally released, as Cerious Software, the vendor, decided at the last minute not to enter the Macintosh market) and v5.0 for Windows. For ACDSee, I had v1.6 for the Macintosh, and v2.4 for Windows. Direct comparison of results between different versions of the same software on the two platforms may therefore contain more than just a small margin of error, and we will shortly see the impact of this as we look at the results.
OK, enough setup! How did they fare? Which of the two CPUs takes the crown? Here are the results for the Thumbnailing and Large Image Rendering tests. Note that the thumbnailing test involved generating thumbnails for an entire folder of family photos. I used two such folders, one with photos from 2010 (412 images), and one with photos from 2014 (121 images).
Given the vast disparity between the above results, which was very much at variance with the expected results, I concluded that some additional data points were needed, and so I went on a “best of” testing spree, where I tried to find the fastest and best image viewing programs for each platform, just to see what was possible on each CPU when a performance minded programmer put their skills to best use. On the Mac I tried out ThumbsPlus, JPEGView, ExifViewer and QPict. On the PC, I tried out ThumbsPlus, ACDSee, IrfanView and NTj. Here are the results:
What do I conclude from all of this? Well, lets start with the obvious. Clearly, comparing the results between earlier Macintosh versions of a program and later Windows versions of the same program is probably not a reasonable thing to do. As any software product evolves, it has the opportunity to be optimized, and in the CPU-bound world of image decoding in the late 1990s, it is reasonable to expect that this happened, driven by the need for competitive advantage. We can further conclude that the algorithms used by each of the programs, and the specific ways in which they were coded for optimum performance, was probably a greater influencer of results than the CPUs themselves. This can be seen by the equal decode times in the Large Image Rendering test by the best of both programs for each of the Macintosh and the Windows PC.
In theory, the PowerPC 604e should have had the edge in a CPU bound task like JPEG decoding, given its larger L1 cache and its superior floating point performance, but the Pentium Pro PC balanced this out with its dramatically larger and faster L2 cache, which probably allowed all of the JPEG decompression code to run 100% from cache, with no external memory accesses at all. In the end, we can guess that it should have been pretty much of a dead heat, and as the results show, it was.
Therefore, for the purposes of this definitely less than rigorous benchmarking exercise, I declare it to be a tie. Back in 1997, you would have been well served, from an image viewing perspective, by either of these two worthy contenders.