Question x86 and ARM architectures comparison thread.

[poke01]

Might as well make this thread that can be used to make direct comparisons of x86 and ARM. All ISA tangents in vendor specific threads can be moved here too.

Edit:
If you want to compare use actual facts and evidence. Saying x86 is going to die or ARM, isn’t good enough. Please also refrain from attacking companies as it provides no basis for architectural comparison.

Do not compare market share that is totally irrelevant.

 

[CouncilorIrissa]

This is going to be a civil thread.

 

[igor_kavinski]

@poke01 As one of the very few who rocks both M4 and a decently recent x86 chip, what is your experience like, speaking as a power user of both?

I have the M1 MBA but since I'm probably never going to own another Apple device again, I use it mainly for watching movies. To do anything serious, would mean to have to get a dedicated USB-C storage boot device which I just don't feel like it's going to be worth it for me.

 

[johnsonwax]

Seems cruel to put Intel hopium in the title so soon after earnings.

 

[mikegg]

This post is brutally honest. You may not like it.

x86 on the consumer side is becoming irrelevant. It's so soundly beaten by ARM that it's not funny.

Let's suppose Intel's chips have the same performance and power efficiency of the M4 series. You guys would be calling for AMD's funeral.

Heck, even a 20% power efficiency difference between AMD and Intel and you guys would be setting up a funeral.

M4 is 2-4x more efficient than AMD and Intel's very best chips. And it has much faster ST speeds. Any non AAA gamer should never buy an AMD and Intel computer again. I know some of you are very proud and are holding onto the last straw. You just can't believe that a fanless phone is faster than your water cooled overclocked monster PC in many common tasks.

• AMD AI 395+: 116.8 Points in CB24 ST, 2.62 points/watt. 44.24 watts when running ST
• M4 Pro: 178 points in CB24 ST, 20.2 points/watt, 8.9 watts

So M4 Pro is 52.4% faster in ST while also being 397% more power efficient. Source is Notebookcheck. Anyone who has ever used an Apple Silicon machine knows how much better it's than an AMD one. The real world experience and the numbers agree.

But M4 is only Apple. Yea but Qualcomm's next-gen Oryon cores are looking like they'll bust AMD and Intel's asses and leave them behind for good. Qualcomm Snapdragon 8 Elite already beats LNL soundly. Next gen X Elite will use the core design after 8 Elite. I don't see how Intel and AMD can keep up unless they improve their efficiency by 100-200% in 2026.

On the server side, x86 still leads in raw performance. But ARM just reported that it now owns 50% of the total server CPU market through hyperscaler in-house designs. So x86's dominance in the server is being eroded quick too.

x86's decline correlates with Anandtech's demise. I started out on Anandtech because I was very interested in the fastest CPUs which happened to be in the x86 DIY market. But now that even a phone SoC is faster than in ST than AMD and Intel, I lost all interest in x86 DIY market.

 

[511]

On the server side, x86 still leads in raw performance. But ARM just reported that it now owns 50% of the total server CPU market through hyperscaler in-house designs. So x86's dominance in the server is being eroded quick too.

This stats is bull**** Intel has 55% total CPU market share currently

 

[mikegg]

This stats is bull**** Intel has 55% total CPU market share currently

My original statement regarding this was wrong. It's 50% shipped to top hyperscalers. So probably AWS, Azure, Google, Oracle. They take up lion's share of the market anyway. Doesn't really change my conclusion which is that x86 is quickly being eroded in the server.

Fast forward to today, the adoption of Neoverse has reached new heights: close to 50 percent of the compute shipped to top hyperscalers in 2025 will be Arm-based.

Half of the Compute Shipped to Top Hyperscalers in 2025 will be Arm-based

Close to 50 percent of the compute shipped to top hyperscalers in 2025 will be Arm-based, with AI redefining the cloud computing landscape.

newsroom.arm.com

It's not just Epyc causing Intel's server CPU share to collapse. It's mostly ARM.

 

[MS_AT]

So what is the purpose of this thread? To discuss ISAs, or chips using those ISAs or everything related to ARM/INTEL/AMD?

And small nitpick, x86 is dead. x64 is with us for 20+ years

And the sad thing is there are still people upset that somebody would like to drop support for x86...

 

[511]

And the sad thing is there are still people upset that somebody would like to drop support for x86...

It should be dropped to give Designer more verification time for other stuff Tom Forsyth has said that xtor count is tiny but the pain to maintain is quite big. x86_64V3(Haswell) should be the idle target imo but Intel is shipping CPUs without AVX2.

 

[poke01]

Come guys. only facts and not opinions are to be said here in this thread.

If you want to compare use actual facts and evidence. Saying x86 is going to die or ARM, isn’t good enough. Please also refrain from attacking companies as it provides no basis for architectural comparison.

Also do not compare market share that is totally irrelevant.

 

[Covfefe]

So what is the purpose of this thread? To discuss ISAs, or chips using those ISAs or everything related to ARM/INTEL/AMD?

It gives ARM users a place to dunk on x86 to their hearts delight without polluting the AMD and Intel threads.

 

[poke01]

To discuss ISAs, or chips using those ISAs or everything related to ARM/INTEL/AMD?

All the above

 

[511]

ISA doesn't matter tbh in many cases Only how you architect your stuff what are your design goals and Process also the Software matters.

 

[poke01]

ISA doesn't matter tbh in many cases Only how you architect your stuff what are your design goals and Process also the Software matters.

The microarchitecture is key. Mike Clark did say the Zen uArch could be made on ARM too.

An Interview with Zen Chief Architect Mike Clark

Zen is one of the most important microarchitectures in the history of the x86 ecosystem.

www.computerenhance.com

I don’t see AMD ever doing this on the Windows side, why give up a good software ecosystem that is win32.

 

[511]

In the end everything is going to get converted in uOPS wether ARM/x86_65/RISC every thing from top level view is the exact same.

 

[Covfefe]

On a more serious note. It's clear that ARM CPUs on the market today are more efficient than x86 CPUs, but how much of that is due to innate differences stemming from the ISA, and how much is due to different design requirements? Every ARM core was designed to target smartphone and laptop. Every x86 core was designed to target server, desktop, and laptop.

My question would be, is it possible to design an x86 core that is high performance and sips power? And if not, what's stopping them?

 

[511]

My question would be, is it possible to design an x86 core that is high performance and sips power? And if not, what's stopping them?

Money they design according to their business needs.

 

[Covfefe]

Money they design according to their business needs.

I'm more so looking for technical reasons.

 

[poke01]

My question would be, is it possible to design an x86 core that is high performance and sips power? And if not, what's stopping them?

I would argue AMD already did this with Strix Halo. It’s powerful yet sips power.

 

[511]

I'm more so looking for technical reasons.

look at Atom Cores than it's the most ARMish design i would say

I would argue AMD already did this with Strix Halo. It’s powerful yet sips power.

Strix Halo is good

 

[poke01]

If we are talking below 15 watts, I don’t think there is a x86 core that’s as powerful as any latest ARM core in single thread in this scenario.

The thing is most ARM vendors need to keep their single threaded peak performance below 10 watts so they come up with architectures that enable that while maintaining the performance possible.

As you scale up tho better cores are negated by more cores and more threads. AMD shines here.

 

[DavidC1]

On a more serious note. It's clear that ARM CPUs on the market today are more efficient than x86 CPUs, but how much of that is due to innate differences stemming from the ISA, and how much is due to different design requirements?

~20-25%. Rest are all design and due to massive ARM share, all the top engineers going there.

My question would be, is it possible to design an x86 core that is high performance and sips power? And if not, what's stopping them?

All the smart engineers go to where they get paid and are treated well. In 2000's that was Intel. In 2025 that is ARM vendors. Even ARM vendors have defections such as with Gerald Williams III and Apple.

Of course there's a third reason. x86 vendors are in a reality distortion bubble because they are protected by it. Thus they don't need to be efficient.

They call Lunarlake efficient right? Well, how about an 8-inch x86 Tablet offering 6-8 hours of battery life with a tiny 15WHr battery? That was achieved with Atom Silvermont core in 2014. And it was cheap too. I had such a Tablet. It barely heated up as well.

If they were forced to compete they would innovate to be more efficient or suffer the consequences.

Let's suppose Intel's chips have the same performance and power efficiency of the M4 series. You guys would be calling for AMD's funeral.

Yes, if either x86 vendor had that big of a difference, the other would fade into irrelevancy. Not during the heights of their differences(P4 vs Athlon, Bulldozer vs Sandy Bridge), the differences were anywhere near that large.

That's why I say Core 2's relative position wasn't that impressive, considering what the ARM chips would achieve just a few years later.

The differences in how low ARM can scale while having smashing performance shows how badly executed x86 platform is. This isn't multi-generational difference, it seems almost insurmountable at this point.

 

[poke01]

The differences in how low ARM can scale while having smashing performance shows how badly executed x86 platform is

No ARM vendor tries to scale up as fast as x86 does in the server space. Both Qualcomm and ARM do have aspirations for the server market, let’s see how that turns out.

As for how low x86 can go. I think the PS6 handheld which is rumoured to be semi-custom should be good indicator of low power x86 designs below 15 watts.

 

[MS_AT]

It's kind of funny that people mix ARM with Apple. It's Apple that is both ahead and have usable ecosystem.

I mean the Snapdragon 8 Elite is waste as mobile SOC, aside from throttling android is not a platform that would allow you to easily use the chip for productive tasks. Android should ship with fully working desktop mode out of the box, to justify performance of these chips... But my laptop still beats the utility of my smarthpone even if it's not hard to find smartphones that have better specs than my laptop.

Snapdragon X Elite was another unsuccessful attempt at WoA that does not have enough advantages to outweigh issues it has on Windows, not to mention the Linux support is questionable still.

Dimensity 9xxx chips are mobile only so also waste of silicon really.

The only well supported ARM platform accessible outside of Apple ecosystem, that has any sort of utility is Raspberry Pi 5, which is outdated, having only A76.

While there are SBCs with A720, the support is lacking and buying one of these you are bound to become a beta tester one way or another.

So while within Apple's walled garden stuff looks nicely, outside ARM still has a lot of work to do in consumer space. It's also worth remembering that while hyperscalers can tune the software themselves to work on ARM, they can also tune it to work on RISC-V if they find that ARM license becomes too costly. And ARM wants to rise prices further.

 

[Covfefe]

I would argue AMD already did this with Strix Halo. It’s powerful yet sips power.

look at Atom Cores than it's the most ARMish design i would say

If that's the best x86 can do, then I think the logical conclusion would be that x86 is fundamentally worse than ARM.

If we are talking below 15 watts, I don’t think there is a x86 core that’s as powerful as any latest ARM core in single thread in this scenario.

The thing is most ARM vendors need to keep their single threaded peak performance below 10 watts so they come up with architectures that enable that while maintaining the performance possible.

As you scale up tho better cores are negated by more cores and more threads. AMD shines here.

Yes, I'm primarily looking at single thread efficiency and battery life in light workloads. Though even the best case scenario makes x86 look mediocre in terms of PPA; Strix Point ekes out a win in Cinebench multi, but Zen5 is a larger core with far more cache.

For the record, I do think most of the difference can be traced down to different design requirements, but I'm playing devil's advocate because I haven't heard a good reason why AMD and Intel wouldn't make their CPU cores more efficient if they could.