It's About Time Intel, AMD Dropped X86 Games And Turned To The Real Threat

Opinion This week, Intel and AMD set their decades-old rivalry aside to ensure x86 remains relevant amid growing adoption of competing architectures.

The formation of the x86 advisory group, announced at OCP in San Jose, California, is a long time coming and frankly, should have happened years ago.

The group, which includes folks like Linux-kernel tsar Linus Torvalds and Epic's Tim Sweeney along with the usual cast of hyperscalers, cloud providers, and OEMs, is focused on driving consistency across the x86 ecosystem and identifying new areas of innovation.

Ultimately, the union boils down to this: the last thing anyone needs or wants to deal with is compatibility edge cases because Intel or AMD decided their implementation of the x86 ISA was better than the others'.

By the sounds of it — and we'll have to wait and see how this actually plays out — x86 giants have finally seen the light and realized they aren't just competing with each other anymore. In what feels like the blink of an eye, Arm has established a foothold in markets once thought safe havens for the x86 giants.

Even if the British chip designer's overall share in these markets remains small, its parts aren't just nipping at Intel or AMD's heels, they now pose a very real threat.

This is particularly true in cloud. It is estimated that Amazon's Arm-compatible Graviton CPUs accounted for roughly 20 percent of AWS CPU instances by mid-2022. We can only imagine that share has grown following the general availability of Graviton 3 that year and the launch of Graviton 4 late last year. Seeing that, it's no surprise that over the past year Microsoft and Google have announced and deployed their own Arm-based CPUs. Both have no shortage of internal workloads they can optimize for.

More importantly, because all of those parts share a common ISA, any workload developed for one should enjoy broad compatibility for the other, which is more than can be said about x86 architecture over the years.

As AMD EVP of datacenter solutions Forrest Norrod put it during a press briefing earlier this week, "x86 is the de facto standard. It's a strong ecosystem, but it's one that really Intel and AMD have co-developed in a way, but at arm's length, and you know, that has caused some inefficiencies and some drift in portions of the ISA over time."

"Drift" is putting it mildly. x86 history is filled with stories of one chipmaker racing ahead with new extensions, leaving the other to catch up or risk falling behind or worse incompatibility. AMD64 may be the first example that comes to mind, but it's just one of many.

For the better part of a decade, Intel was the only x86 chipmaker to support 512-bit vector extensions — something that might have given them a leg up if early implementations of the SIMD instructions weren't so hot, power hungry, and generally inconsistent in their execution.

It wasn't until the launch of AMD's Zen 5 cores earlier this year that the chipmaker finally delivered a true AVX-512 extension that didn't require double pumping a 256-bit data path to do it.

These kinds of mismatches in direction are the missteps Intel and AMD now hope to avoid going forward. And to be honest, it's about time. There are plenty of ways for the two companies to differentiate, but mucking with the ISA in hopes of getting a leg up on your competition just isn't the way to do it anymore.

Perhaps if they'd figured this out sooner, the two chipmakers could have agreed on a better spec for a matrix math engine — something Intel has in AMX, but AMD still lacks — and avoided burning die area on NPUs just to make Microsoft happy. Just a thought.

Intel and AMD's commitment to improving consistency across x86 chips isn't limited to the existing specs. Working with industry partners, the advisory group aims to shape the ISA's direction going forward.

That all sounds good on the surface, but we can only imagine the gridlock that could ensue if Intel and AMD can't agree on how best to address customer demand. Oh to be a fly on the wall of those meetings.

• Spectre flaws continue to haunt Intel and AMD as researchers find fresh attack method
• Intel, AMD team with tech titans for x86 ISA overhaul
• AMD downplays risk of growing blast radius, licensing fees from manycore chips
• Arrow Lake splashdown: Intel pins hopes on replacement for Raptors

On the flipside, major changes to any instruction set aren't without risks. Additions like AVX-512 or AMX usually aren't that big of a deal since no one is forcing you to use them. Deprecating features, on the other hand, can be rather contentious.

If you're the first to cut support for something before the market is ready, you could potentially end up driving customers to your competitors. And with Intel already looking to modernize and de-bloat the aging x86 ISA for the modern age, the newly formed advisory group, along with tighter collaboration with AMD, should help to de-risk some of that. The last thing Intel needs is to give customers more excuses to ditch its silicon.

With that said, we can't imagine anyone running modern servers is going to be too upset about dropping 16-bit or 32-bit support from x86 in 2024, as Intel has recently suggested. Better for these decisions to be made, not just with Intel and AMD at the table, but with the broader ecosystem weighing in on the what, when, and how of the whole thing.

What changes the x86 Ecosystem Advisory Group will bring in the near term? Neither chip biz is willing to say just yet, and with development roadmaps for new silicon being what they are, we don't expect to see much in terms of change for at least a year if not two. ®