Mashed Articles

Could the golden age of ever-increasing PC performance be over? That’s the conclusion I’m struggling to shake off following the latest processor product announcements from AMD and Intel. After getting on for 30 years of fun, it looks like the pursuit of processor performance for its own sake has finally drawn to a close.

The chips in question are Intel’s upcoming Sandy Bridge E and Ivy Bridge processors, along with AMD’s new family of Bulldozer-based CPUs. Sandy Bridge E is Intel’s next uber-high performance CPU. At least, that’s where it will sit in Intel’s hierarchy and price lists. But here’s the thing – it only has six cores.

OK, six cores is hardly shabby, and when Sandy Bridge E arrives at the end of 2011, it will bring a few performance enhancements beyond mere core count. But Intel has been flogging six cores for desktop PCs since March 2010. Moreover, if you start digging down into the details, you can only conclude that Intel is no longer interested in pushing the limits of performance.

A key statistic here involves transistor counts. Intel’s existing six-core chip, known as Gulftown and sold under the Core i7 sticker, packs a little under 1.2 billion transistors. We don’t know how many transistors Sandy Bridge E will sport, but given the only major change is the addition of some simple I/O and probably a little extra cache memory, it’s unlikely to increase much.

Meanwhile, Intel’s current quad-core chip clocks in at just under a billion transistors. It’s very nearly as big, in other words – something worth remembering when you consider that Intel’s six-core chips top out at £800, while the quad-core model can be had for just £150.

Things will get even fishier when Intel Ivy Bridge chips roll out in early 2012. In Intel’s ‘tick-tock’ speak, Ivy Bridge is a tick. It’s largely a carry-over architecture with four cores, so it won’t offer dramatic performance increases despite being built on a new production process – in this case fancy 22nm tri-gate transistors.

But get this: Ivy Bridge will pack a total of 1.4 billion of those tri-gate transistors. Ivy Bridge is launching with just four cores, so Sandy Bridge E, Intel’s highest performing processor, could well have fewer transistors than the mainstream four-core Ivy Bridge chip.

Of course, Ivy Bridge’s 22nm transistors will probably mean it’s actually a smaller chip than the 32nm Sandy Bridge E, and some of Ivy Bridge’s 1.4 billion transistors are spent on an integrated graphics core that Sandy Bridge E doesn’t get, but the bottom line is that Intel’s most complex chip won’t be its most expensive chip.

Put another way, I fancy the creation of an eight-core version of Sandy Bridge E would be a trivial job for Intel.  In fact, it’s likely that Intel does have an eight-core variant ready to go into servers. It just won’t be offering it on the desktop.

As for AMD, its new Bulldozer-based processor, the FX CPU, is due out any day now. You can read all about the details in PC Plus issue 314, on sale 18 October, but the headline specification includes a claimed eight cores. The reality, however, is that Bulldozer is really a very hefty quad-core chip with some hardware thrown in, enabling two threads per core. It’s only slightly faster than an existing Intel four-core processor.

Oh, and did I mention that nothing exciting is happening regards clockspeeds from either Intel or AMD? The current limit around 3.5 to 4GHz will remain. Put it all together and plot a graph of absolute PC processor performance over time, and I reckon a fairly dramatic levelling off is becoming increasingly likely. There’s nothing on the road maps for at least the next six months that’s going to significantly outperform the Intel Gulftown six-core chip that’s been knocking around since March 2010, and that means performance will have been static for around two years.

Part of the explanation is a lack of competition. If AMD’s new FX processor only just has the legs on Intel’s four-core chips, Intel is under no pressure to release something dramatically better than its six-core offerings.

But there’s more to it than that. I suspect AMD is capable of pushing Intel much harder. The reason it doesn’t is that for the vast, vast majority of PC users, there’s really no need for more performance, which means there’s not much demand for faster chips. If there was, AMD and Intel would make them. But there isn’t, so they aren’t. I’m risking an (admittedly apocryphal) Bill Gates 640k-moment, but the processing power we have today is enough for the foreseeable future. Sorry, folks. The fun’s over.