Honestly, I tried to fight the urge to reminisce today, but August 18th marks the day the youngest member of the Pentium Pro family turns 14 years old. Resistance is futile. The Pentium Pro 200/1M was the swan song for the original, MMX-less P6 core. The chip valiantly forged ahead, even as its Pentium II and Xeon successors washed over the computer landscape, and it happens to be my favorite CPU of all time.
While the P6 introduced many architectural enhancements that persist in CPUs today, my fascination with the Pentium Pro processor is not founded on technical merit alone. When something commits itself to your brain's favorites table, something more than just a list of numbers and features is typically required. For instance, I have a friend whose favorite CPU award goes to be the Applebred Duron, which wasn't chosen based on raw performance figures, I'm guessing. Indeed, if you ask an audience about their favorite chip, you'll receive an eclectic and exhaustive list in return, with accompanying stories to boot. The story of my own personal attachment to the Pentium Pro largely boils down where I was and what I was doing at that point in my life.
Even though the first commercial Pentium Pros wandered off the production line at the tail end of 1995, it would take almost four years beyond that for one to land in my lap for the first time. As a junior in high school, I landed my teenage dream job working for a local computer merchant. The vendor purchased used corporate equipment as it came off lease and then resold it through a then-novel website called Ebay. It was also around this time that the first few batches of corporate workstations and servers sporting Pentium Pro processors were up for retirement.
After wrapping up a couple months of tedious work wiping hard drives and cataloging a huge backlog of IBM 486 PS/2 boxes, a new batch of computers was backed up to the loading dock; hidden among them lurked a Dell Optiplex GX Pro workstation sporting the Pentium Pro 180MHz/256K processor. This workstation had the misfortune of a bum motherboard, which meant it was my duty to part out any usable or resalable components. As I dismantled it, the heatsink fell by the wayside, and I knew it was love.
Up to this point, my personal frankenputers had been based on Socket 3 and Socket 7 chips. The Socket 8 CPU staring up at me was a behemoth in comparison. Removing the chip from its socket only dug the hook in deeper. The physical heft of the the thing, the gold heat spreader, and the unique pin layout on the bottom made an immediate impression. It felt as though I was holding something powerful, capable, and professional in my hand—a dangerous combination of attributes for an impressionable teenager.
Other working systems, replete with Pentium Pro chips, eventually made their way into the warehouse. Because literally everything was for sale, I had to devise a novel ploy in order to selfishly utilize one as my work system. This plan involved scrounging a 2GB hard drive, loading it with my OS and files, and then moving the drive between computers as they were sold out from under me. The scenario was far from ideal, but it allowed me to get plenty of hands-on time with my new hero. Our logistics manager eventually adopted a similar strategy.
In 2001, after the Pentium 4 had been on the market for several months, older Pentium Pro-based systems weren't exactly flying off the shelves. It was at this time that I bit the bullet and finally purchased a one to call my own. While doing my thing at the warehouse one day, I happened upon a Compaq Proliant 850R server that was missing its hard drives but otherwise complete. 50 dollars changed hands, and a bona-fide server was mine to play with.
The Proliant was special in two ways. It was not only my first Pentium Pro-powered computer, but also my first dual-processor system. Initially, it housed a single 200MHz/256K processor and 64MB of ECC EDO RAM. Over the next few months, I managed to rummage up a second, matching CPU with an accompanying VRM module, a couple 4.3GB SCSI hard drives, and an impressive 512MB of RAM, which was the most the system could handle.
The video card in the server was an oddity. It pulled double duty as a SCSI controller, and as video performance is not a top priority for servers, the display output was limited to 256 colors with a maximum resolution of 1024x768. I wasn't about to give up the Riva TNT2 in my home desktop just for the novelty of having dual processors, so the Proliant was relegated to *gasp* server duty.
About a year after I had my server tricked out and happily pulling double duty as a file and DHCP server, I bumped into the final piece of the puzzle. I was in college at the time, but I was still working in the computer warehouse during the summer. A faulty Proliant 6500 came through the door one day, complete with four Pentium Pro 200MHz/1M processors. I had played with Pentium Pros featuring 256K and 512K of L2 cache throughout my illustrious career, but the 1M versions had skated under my radar. As the heatsink was removed from the first 1M chip, revealing a black aluminum heat spreader, I was in love all over again. Naturally, a pair of them came home with me that night.
The 1M version of the Pentium Pro differed from the others because the die was packaged using an organic substrate material rather than a ceramic one. It was also the only version of the lot to get a black aluminum heat spreader. There is a very sensible reason for this cosmetic makeover, but it requires a little history lesson.
Prior to the Pentium Pro, Intel CPUs utilized an L2 cache that was located off package, typically soldered or snapped onto the motherboard. For instance, the Pentium P5 and P54C chips only had a paltry 16K of on-die L1 cache. When the CPU needed something that was not stored in the L1 cache, it had to reach out across the motherboard's backside-bus and query the L2 cache for the desired instruction. In this scenario, the cache was physically distant from the CPU and ran at a fraction of the its core clock speed, which introduced some serious latency into the computation.
Even more important than where the L2 cache was located is why the cache was put there. The short answer: cache is expensive. In the mid-1990s, Intel was fabricating chips using silicon wafers 150 mm and 200 mm in diameter—not the larger 300-mm wafers commonly used today. Because the process technology of the time wasn't as advanced as it is now, the resulting CPUs were comparable in size to many of the chips in production today even though they had much lower transistor counts. Fewer dies were etched onto each wafer, making mistakes even more costly. Adding an integrated L2 cache would have dramatically increased the overall die size of each chip, increased the likelihood of errors within it, and reduced the number of chips that could fit onto an already cramped wafer. That wasn't exactly exactly the best-tasting recipe for success.
Back then, fabricating cache modules separately was simply a function of operational prudence. By detaching the cache from the core, you wouldn't have to waste a perfectly good core because of an error in its cache, or vice-versa. Absolute performance took a back-seat to yield considerations until production lines with smaller lithographic processes and larger wafers came online.
The Pentium Pro was the first Intel processor that contained a L2 cache on the same package as the CPU die. It is important to note that the cache and the CPU were still physically separate dies, but they were both mounted inside the same package. The close proximity of the CPU to its L2 cache eliminated the long back-side bus ride and enabled Intel to run everything at same clock speed as the host CPU, providing a significant performance boost for cached data.
The 256K and 512K models consisted of two chips, one CPU core and one cache module, which could fit comfortably under the original gold heat spreader. In order to shoehorn 1M of cache, Intel had to add a second cache die, bringing the total up to three separate dies on a single package. The resulting row of silicon exceeded the boundaries of the gold heat spreader, necessitating a new design. Thus, the black aluminum heat spreader was born.
I spent a lot of time with my pair of Pentium Pro 200/1M processors. The Proliant 850R required a little initial tweaking to work with the 1M chips, but once up and running, it stayed in service for many years. This was the server l used to experiment with Apache, MySQL, and PHP for the first time. Before my ISP wised up and blocked port 80, I used the machine to host several small websites for myself and some friends. The rig could easily handle more requests than my upstream bandwidth was capable of at the time. Many memorable nights were spent writing code and benchmarking the system against newer machines to see how well the old girl stacked up. I had great fun with it and felt like I was accomplishing something important in the process, which is why the Pentium Pro 200/1M remains my favorite processor. Happy Birthday!
Dare I ask what your favorite CPU is and why?
|In the lab: WASD's Code keyboard with Cherry MX clear switches||23|
|GeForce 344.48 driver enables DSR on Kepler, Fermi GPUs||49|
|ARM intros two new CCN 'uncore' products for data center SoCs||10|
|G.Skill's Phoenix Blade PCIe SSD boasts 2000MB/s transfer rates||22|
|First Win10 Tech Preview update adds Action Center||18|
|Reports: Broadwell-E slips to 2016, but Skylake-S sampling already||29|
|Cooler Master's Mizar mouse reviewed||10|
|Cooler Master's Nepton 240M liquid cooler reviewed||32|
|AMD cuts A-series desktop processor prices||64|