Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Will x86 be dead in a decade?
- Thread starter Apple Fan 2008
- Start date
- Sort by reaction score
Will it be dead, no. Is it already completely outdated, yes! See that's the beauty of Apple. They can decide a technology is outdated or mandatory and within one year you can't buy a Mac with or without it. On the PC side of things you can still buy motherboards with (pre-USB 1.0/1.1) color-coded single-purpose PS/2 ports, purple for keyboard and turquoise for mouse port.
Land of the Past ™
So we can be certain that x86 PCs will never disappear completely and they will even be a hindrance for the Windows ecosystem to make a proper ARM transition. If Windows on ARM even becomes a thing within a decade, it will be a huge achievement. Better expect no progress whatsoever. Not because x86 will catch up on energy-efficiency, only because of inertia.
Land of the Past ™
The clock which makes tick-tock-tock-tock- ... 😆
"An exclusively 64-bit mode architecture"
www.intel.com
Envisioning a Simplified Intel Architecture for the Future
PC computing rests heavily on Intel Architecture processors with an enormous installed base, and cloud computing is synonymous with Intel Architecture.
I will say not dead, but increasingly irrelevant for consumer computing. I think it might cling on a good while yet in servers, enterprise and gaming sectors, but for ultrabooks/ general use laptops* and tablets/ convertibles on the Windows side I do think Microsoft's upcoming big push will take Arm mainstream on those smaller devices it's better suited to.
* depending on if Intel/AMD can stay price and performance competitive at the low end.
* depending on if Intel/AMD can stay price and performance competitive at the low end.
Apple was reportedly hiring RISC-V experts. But it might not be for mainstream Mac/iOS CPUs. Apple has a lot of little processors that run inside their SoCs that are never exposed to the user but run internally. They might be looking to use RISC-V in that context.
"Increasingly irrelevant"? PC Gaming is still one of the biggest segments on the PC side in terms of revenue generated. Between prebuilt gaming PCs/notebooks, custom built PCs, and the gaming industry itself, there are massive and constant income being generated there.
As is office computers. Still the majority in that space even if non office users don't see them. It's like people around here don't even see 90% of the market...
The typical lifecycle of servers in hyperscale clouds is around 4 years (and longer in on-prem enterprise deployments). Currently x86 has ~90% market share, so new deployments are still overwhelmingly x86-based. Unless some kind of huge unexpected disruption happens, x86 will still dominate the market in 10 years. Some large companies that can afford developing and maintaining their own CPUs and platforms (primarily Amazon and Microsoft) will take some share in some segments, but they will probably not come close to Intel's/AMD's scale.
The PC market is more fickle, but at the moment Apple is the only significant non-x86 player, and they are at around 7.5% global market share (it has actually dropped a bit YoY, probably because of their high prices). In the US their market share is more impressive. Microsoft may try to make another ARM push, but I'm not convinced that they are very serious about it.
The PC market is more fickle, but at the moment Apple is the only significant non-x86 player, and they are at around 7.5% global market share (it has actually dropped a bit YoY, probably because of their high prices). In the US their market share is more impressive. Microsoft may try to make another ARM push, but I'm not convinced that they are very serious about it.
If you pull back a bit, though, what really killed PA-RISC and Alpha was market forces.
DEC and HP were vertically integrated systems companies which owned their chip fabs, ISA, CPU design, board design, system design, OS, and other software. The problem for companies like these in the 1980s-1990s was that silicon R&D costs were going up, prices were going down due to competition from the bottom (most notably x86), and they didn't have high volume products to make the economics of fully in-house CPU design and fab work well. Realistically, every one (not just DEC and HP, there were others) faced a hard choice: get out of the private CPU business, or try to increase volumes by selling their CPUs to others.
HP chose the first path. You mention HP's C-suite buying into the plan, but it was actually the other way around. What we now know as Itanium started as a HP project called PA-WW (WW = Wide Word, an allusion to VLIW). HP's architects were VLIW guys, and they believed (incorrectly) that out-of-order RISC was a dead end which would stop scaling very soon. So, when they started work on a successor to PA-RISC 2.0, their design studies were about making VLIW-like concepts work for general purpose computing.
At the same time, HP's C-suite went looking for someone to partner with on this project, because they knew HP couldn't do it all themselves any more. That's how Intel came into the picture. The Itanium ISA as we know it isn't 100% HP. Intel did take overall ownership, but the base ideas came from HP, and HP architects and circuit designers continued working on the project.
DEC tried the second option: they wanted to serve both internal and external customers with Alpha. The trouble with this plan was that DEC botched executing on it, on multiple levels. Similar stories played out with many other 1990s RISCs. So yes, in one sense Itanium killed them, but they were all doomed anyways - none of them figured out how to survive the transition from easy and relatively cheap fab technology and chip design cycles to expensive.
x86-64 dominate supercomputing, HPC, workstation, server, desktop, laptop, console plus it's starting to encroach on handheld console and SBC which were predominantly ARM. At 5nm it can easily take tablet and at 3nm possibly even mobile. Benefits are maintaining compatibility with world's largest software ecosystem and easier for developers to target one common ISA for software releases. So, the real question is not when x86-64 will die but when ARM will die. Personally, I've been long in ARM SBCs and tablets but ready to ditch them for x64 except for ultra low cost applications since it's hard to compete with $15 Raspberry Pi Zero 2 W. Guess we will revisit this thread in 10 years but my bet is always on the largest software ecosystem is king.
Sure there are layers of complexity, but "market forces" gives it a sense of inevitability when I'm not sure anything went according to anyone's plan. My point is that the outcome, the spectacular failure of Itanium, really didn't do much harm to Intel and, in an era of "embrace, extend, extinguish" may not have been all that terribly far from the plan in the end.
I think the failures were mostly human. HP and DEC were struggling with defining a cohesive strategy to maintain their early success. DEC in particular was all over the map, while HP kept bungling acquisition after acquisition. HP lost their way, and foolishly put their faith in Intel. But again the point here was about Itanium, and HP and Compaq closed down their successful PA-RISC and Alpha product lines to focus on Itanium and Itanium let them down.
By the mid-90's, before the conception of Itanium, both HP and DEC were fab-ing their chips outside. PA-RISC was being manfuactured by IBM and Intel, and Alpha by IBM and Samsung.
Alpha ran Windows NT (which was written in large part by former DEC engineers). That lack of vertical integration left them open to the pain of Microsoft later reversing course and dropping support.
Both DEC and HP ran third party software just fine.
Meanwhile the more truly vertically integrated approach, IBM and POWER, has survived all that turmoil. The horizontal Itanium effort among Intel, HP and Compaq/DEC failed.
As I said, started by HP, coopted by Intel. I'm sure it got Intel's ear that HP was making the case that not only was the new architecture being positioned to kill RISC, but x86 as well.
Who else used the Alpha? DEC sold the StrongARM to other manufacturers, including Apple, but I'm not aware of them selling the Alpha as a component.
$15 if you can find it. It’s kind of ridiculous that they are so hard to find. You can find the 400 at least though.
Indeed, they are currently executing a very aggressive plan to catch up to TSMC. There are risks, but so far they seem to be on track. Their "Meteor Lake" CPU later this year will be a test case. It's their first CPU manufactured using EUV lithography, and their first tile-based mainstream CPU.
At the moment ARM can really only compete with x86 in two segments: low-power ultra-portables, and server CPUs with high compute-density that are useful for HPC and some cloud workloads. For the former, Intel is betting on their "Lunar Lake" 3nm CPU, which has already taped out and will probably appear in late '24. On the high-density server side they have "Sierra Forest", which is a Xeon with a large number (up to 144) of E-cores. AMD is also active in this segment with their Bergamo EPYCs. If this all works out as planned, there won't be many reasons left to switch to ARM.
Last edited:
At their current process node, maybe, but I'm not sure why it can't continue to improve as processes improve. The difference is that x86 is probably about as fully optimized as it can be, so there aren't a lot of architectural improvements hiding in there anymore. Arm (or more specifically Apple as an implementer of the Arm ISA) probably still has room for architectural optimizations on top of process improvements.
When both architectures are fully optimized, I'd expect Arm to end up with a lasting advantage. It's a simpler, cleaner architecture with less unnecessary logic. x86s would be part of a solution, but I think Intel needs to go further than that to really be close to parity.
Beyond the Arm/x86 advantage though, I think Apple has an advantage in controlling the SoC and the software together. They can add and deprecate custom coprocessors at will in a way that WinTel would struggle to-- at least until WinTel changes how the approach legacy support.
I'm not sure why that would be the case. Apple has really targeted a narrow power window, but there's no reason to think they or someone else couldn't take the very high performance architecture Apple has shown ARM to be and allow it to consume more power to gain more performance.
It raises the question though why they were running in place for years and now suddenly think they have a plan to get out ahead of the curve again. Were they holding back because they didn't sense competition before? Does fear of extinction focus the mind in a way laurel resting doesn't?
The other option is that nothing's changed and they're just saying what they need to and are hoping for a miracle.
I would think targetting the power profile they are doing now is deliberate precisely because they need to scale up. A single core consuming 20W at peak power cannot be packed too many together into a single die without causing a meltdown. Maybe Apple thinks that the 5W power of their P-Cores are still too much.
That's the use case that's not covered. Low power portables prefer efficiency over performance and server CPUs with high compute-density can tile lots of efficient cores to get more compute power within the overall thermal budget for multi-threaded applications. There's no reason to think they couldn't trade off efficiency for more single threaded performance if they choose to go for that. I'm not sure Apple will be the company to build that, but a self imposed Apple limitation isn't an Arm limitation.
Time will tell. The spot that Intel is in now is unique in their history though. They've always had 3 main advantages over the competition: process technology, architectural savvy, and high volumes to fund it all. When the Itanium failed, they lost their architectural advantage over AMD for a few cycles but they kept their volume advantage while turning up the pressure on process and managed to hold off AMD.
At the moment, they're lagging TSMC on process, the world is coming to realize that they're arguably lagging Arm on architecture, and Apple sells about as many iPhones in a year as all makers sell PCs. Apple's operating income is about 50x that of Intel. Qualcomm is 7x. They're behind, or at least challenged, on all three key pillars.
This is new territory for Intel. I'm not sure they still have the talent on staff to turn it around, let alone a cultural ability to execute. That x86s proposal gives me hope that they're starting to be willing to ask the hard questions internally though.
"Faster" is kind of a non-simple thing. x86 designs rely on higher clock speeds because they need them for better performance. The GeekBench numbers show x86 outperforming Apple processors, but they are running the processor cores at "turbo", ranging up to 6 GHz on more recent models. M1s running at around 3.2 GHz were beating the existing Intel processors which were running at 5.0, a speed that was not really sustainable.
If you try to run it fast, you can outpace the capabilities of your memory system, which starves the cores and ends up just burning clock cycles to negligible gain while memory struggles to keep up. Apple seems to be striving to find the RAM/core performance speed balance line, above which you will just be wasting energy. In addition, ARM architecture has an abundance of internal registers, allowing code to get more work done while waiting on memory.
In other words, the broader system has to get faster in order to get a benefit out of a faster processor. There are definitely limits to how much you can get out of winding out on gravel.
HP was working on PA-WW in the early 90s iirc.
I can't rattle off a list of names, mostly because it seems to be getting hard to Google anything about this. I suspect DEC's lack of success in this area is why it hasn't left a large footprint, either on the modern web or in your memory. However, I was able to find a Byte magazine article on archive.org, written in 1992 by a DEC engineer as an introduction to Alpha, and it has this at the end:
"Fully optimizing" isn't a one-dimensional process. You optimize designs with different goals. Intel has historically optimized for maximum performance and treated mobile as more of an afterthought (which is why they missed the boat in the smartphone market 15 years ago), while the M-series was derived from a mobile device CPU that is tailored for Apple's very specific needs (and makes some compromises elsewhere). "Lunar Lake" is Intel's first attempt in a long time to design a mobile-first CPU, and (if they can pull it off on schedule) will also narrow the gap to TSMC's cutting edge manufacturing process.
True, this potentially allows them to move faster. But (coming back to the thread topic) it also means that their CPUs will remain a single-vendor market niche.
You can't just scale the power and clock frequency if the chip wasn't designed for it.
The new CEO started this turnaround plan a couple of years ago, and they are pumping enormous amounts of money into it.
They are still unrivaled in terms of volumes when it comes to computers. You can't just mingle the mobile device market with the market for servers and PCs. As mentioned earlier, x86 still has somewhere around 90% market share in both segments. The M-series has much, much smaller volumes, which is probably the reason why development since the M1 hasn't been very fast.
An underestimated advantage in favor of x86 is that Intel (and to a lesser extent AMD) is nurturing a big ecosystem of standardized platform components and software to enable them, which makes it possible for a large number of OEMs to enter the market. There is nothing comparable for ARM.
Are they though? There is still no ARM CPU that can keep up with their (and AMD's) "big iron" CPUs in terms of raw performance, even though some of the candidates run on a currently still superior TSMC manufacturing process.