Has the desktop reached the point where it can get no better? Ian Evenden looks ahead

© RYANJLANE/E+/GETTY IMAGES

Bought a new PC recently? Did you wonder why it was so similar to the one you’d upgraded from? Perhaps you researched components, and realized they’re just improved versions of the ones you bought 10 years ago. We’ve all been there.

PCs have improved in the last decade or so, with increases in clock speeds, transistor counts, and power consumption, but there’s been no great leap; just hype about AI (and new architecture, thanks to Qualcomm). A PC you build today may look similar to one you’ve built previously, with no advances in compact cases; just more USB-C ports that do slightly different things.

There are some interesting technologies on the horizon, however, such as photonic computing, increased cloud streaming, and software updates that attempt to take advantage of improved hardware. So where is the PC going, is there a way off the plateau we find ourselves on, and will we need to keep upgrading our machines in the future?

Surface devices might not be the ideal thing for gaming, but with cloud streaming, they can be.

© MICROSOFT

THE ELEPHANT IN THE ROOM, if we’re going to argue that nothing significant has happened to PCs in the last few years, is Snapdragon. The arrival of ARM-based chips in the PC, following their successes in cellphones, tablets, Macs, and embedded systems, poses an obvious question: is x86 dead?

Intel has repeatedly said no. AMD would likely be inclined that way, too. Arrow Lake, the Core Ultra Series 2 processors from Intel, are very unlikely to be the last of their line. But the rise of a competitor almost 50 years after the release of the 8086 is a highly significant event. So far, Snapdragon X hasn’t made it out of laptop form, but we only have to look over the fence at Apple’s recent range of desktops to see how an ARM-based design can perform when placed under a decent cooling system. The fact that the chips can also operate fine with nothing more than a fancy heatsink attached (in the passively cooled MacBook Air and iPad Pro) is a bonus that means it’s possible to buy a tablet that’s more powerful than you’ll ever be able to make use of. We look forward to Apple’s attempts to put one of its M chips in an iPhone.

What’s more interesting is that the M4 is built on a 3nm process from TSMC, similar to that used by Arrow Lake’s compute tile. The rest of Arrow Lake, which comes in four parts, with GPU, I/O, and SoC tiles in addition to the one with the processor cores, is built on 5nm and 7nm nodes, but all by the Taiwanese foundry. Intel used its own Intel 4 (7nm) process for AL’s predecessor Meteor Lake, but farmed production of the Core Ultra 2 chips out while it works on its own 2nm node, perhaps ready for Panther Lake in 2025.

CENTRAL PROCESSING

Intel remains committed to x86, having terminated its RISC-Vdevelopment program at the beginning of 2023, and even dumping the x86S project, which would have simplified the processor architecture, slicing away legacy instructions, such as 16-bit addressing, and allowing it to start directly in 64-bit mode, rather than having to step up through legacy modes as it boots. Exactly why the company did this—it seems like an obvious step, as we rarely run 16-bit software on our Core i9 chips, and could probably do it perfectly well in an emulator if needed—is unclear, but Intel has thrown its weight behind traditional x86 with the creation of “...the x86 Ecosystem Advisory Group in collaboration with AMD and other industry leaders,” according to a statement. “This initiative reinforces our dedication to securing a strong future for x86, building on decades of software compatibility,” the spokesperson continued. “While we have pivoted away from the x86S initiative, our focus remains on driving innovation and collaboration within the x86 ecosystem.”

It’s important to remember that x86 is an ISA, or Instruction Set Architecture (if you thought it meant Industry Standard Architecture, then get back to your IBM AT or compatible) that defines how software interacts with the CPU. This means processors we class as ‘x86’ can be wildly different in the way they’re laid out and stamped into silicon wafers, but can execute the same instructions. New instructions are added to the set all the time, as we saw with the Pentium MMX (1997), which began the introduction of SIMD (single instruction, multiple data) commands that improved parallel processing, and was built on in subsequent releases, culminating in AVX-512 that first appeared in 2016, but was removed from 12th-gen Core chips in favor of the Xeon server chips, and is implemented in AMD’s Zen 5.