During the 1990s, semiconductor manufacturers such as Intel seemed confident that processor clock speeds would continue to rise ever higher, as they had done for the previous two decades. This seemed sensible at the time – the original 8086 CPU ran at speeds of between 5Mhz to 10Mhz; its successor, the 80286, ran at speeds of between 6Mhz and 25Mhz. The first Pentium CPU had a clock speed of 60Mhz, but quickly ramped up to 300Mhz as the manufacturing process matured. Within just a few years, a typical PC went from a clock speed of 66MHz to more than a gigahertz, so predictions of 10GHz clock speeds sounded entirely reasonable at the time.
But it wasn’t to be – the Netburst architecture of Intel’s Pentium4 CPU turned out to be an evolutionary dead end. No matter how hard Intel tried, they could not create a chip that ran faster than 3.6Ghz which operated reliably. Worse still, those 3.6GHz chips ran at completely unreasonable temperatures without offering any significant performance gains over the 3.0GHz variants due to the number of failed instructions caused by pushing the silicon to such high speeds. A new paradigm was needed if PC performance was to continue improving in line with Moore’s law – the idea that processor performance would continue doubling every 18-24 months.
AMD Steals the Performance Crown
The mainstream PC processor industry was a much more dynamic place during the 1990s than it is today. The contemporary manufacturers that we all know – Intel and AMD – were both well established at that time, but VIA, and Chinese manufacturer Zhaoxin were all serious contenders back then. AMD managed to take the performance crown from Intel for the first time with their K6 chips and did so again with the launch of their Athlon64 and AthlonX2 processors many years later.
It’s that last part that we want to focus on today – the AthlonX2 was the first mainstream dual core processor for personal computers and sent shockwaves through the industry when it was first released. Creative professionals suddenly had double the power for their workloads; you could use your machine to browse the web and play the latest free slots at the same time with no slowdown; soon, even regular games would begin to harness the additional power available in these new CPUs.
Beyond Dual Core
Dual core processors were immediately hugely popular with PC users, so much so that the single-core chips became practically extinct within just two to three years of the first dual core models coming onto the market. It wasn’t difficult for predict which way things would go next, considering that success – when would we get more than two cores on out CPUs? As the demand for CPU optimization grew, Intel released some great chips around this time, such as the Q6600 quad-core processor, while AMD began to struggle a little and shifted its focus to the value end of the market instead.
AMD did have one very clever idea though – to sell their faulty quad core chips as triple core models, with the faulty core disabled. This enabled them to increase their yield significantly, whilst Intel concentrated on keeping the lead in performance which they believed justified the higher price of their own processors. There was quite a delay before things went further than this, but we did eventually get our six and eight-core processors a few more years later and were promised that chips with more and more cores would be on the way very soon. Fast forward to today, and eight cores is still about as high as a typical desktop machine goes – more cores are available if you are willing to pay a huge price premium, but this certainly doesn’t feel like the future we were promised.
Will More Cores Become Standard Soon?
So what happened? The first AMD Ryzen and Epyc chips that featured 16, 24, 32 and even 64 cores were released more than five years ago now. The first generation of these monsters were termed “Threadripper” by AMD and offered amazing performance, and yet the second generation of the chips actually topped out at a lower core count than the first generation. How could that be?
The reality seems to be that when users did purchase chips with more than eight cores, very few of them could find any real use for a machine with so much parallel processing. It’s that old law of diminishing returns showing its ugly head again, unfortunately – once you get past the first four cores, it becomes harder and harder to improve performance by adding more.
Some specialist applications such as computer aided design, artificial intelligence research, code-breaking and the like can take advantage of massively multi-core machines well, but the software has to be specifically designed in this way, and that is a skill which is only fully understood by a limited number of programmers, even today.
It seems likely that we will eventually move forward to 16 core chips becoming the standard for regular PC’s, but the transition is likely to be a slow one. Just as the move from 8GB of RAM to 16GB is taking a long time for the latter to become the standard, the same looks set to be true in the CPU industry.
It’s quite amazing to say it – back in the 1990s we never would have believed it… but it seems that today, we already have computers powerful enough for the majority of people.