It's an ARM-designed CPU core, made in Taiwan. Very little difference from say MediaTek's upcoming Dimensity SoC just announced earlier because it is made on TSMC N3P instead of N2.
And everybody forgets the true power of Apple: scale. Can that supposed miracle chip be readily found in million of phones within 6 months? Or will Xiaomi keep selling 90% of their phones with 2 year old chips, reserving these chips to their 2000$ foldables?
This post serves Xiaomi because it implies competitiveness when the reality on the ground will remain unremarkable, and it helps fearmongers because they can latch on to Xiaomi's attempt to appear competitive.
Their power of scale applies to less and less regions. Xiaomi, in particular, delivers that to China where Apple previously slapped, now is becoming a niche player. It's the same as with German car makers. The trend now reversed with Chinese manufacturers starring to dominate on foreign markets and it's also something that was unfathomable just a few years back.
They have proven countless of times by now that they are not to be ignored.
To make conversation productive, please define "national security" in this case.
For example, if China can't install its hardware in Pentagon, why is it a threat to national security.
On the other hand, if you say, Intel will be in trouble because of these chips, then I would say, so what? Wasn't this a dream of MBAs, educated by the top US schools, to outsource everything in business to cheaper service providers (e.g. manufacturing to China, software development to India and etc,.)
Over the next decade, China plans to invest up to RMB 1 trillion (US$161 billion) into its domestic semiconductor industry as it seeks to become a global power in the sector. In recent years, Beijing has repeatedly stated that developing its own integrated circuit (IC) industry is of integral national strategic concern. - https://www.china-briefing.com/news/chips-all-in-investing-i...
OH, BUT we have fable! We have AI! of course we can answer these questions? can't we? which is true: America's AI is a super powerful defense against our enemies or America's AI is too stupid to defend against such a threat, and only sactions and import controls will suffice.
Just seems like you're inventing things to be scared of.
I love this, really, competition on a global scale is good for us, bad for greedy US corporations.
For example, I consider Meta as bigger threat to society, than Xiaomi, DeepSeek, z.ai and other top Chinese companies (haven't used Chinese social media, so can't comment about its content)
This is still an ARM C1-Ultra. Same one used in Mediatek dimensity 9500, which also got more than 4000 points in GB6 lab test. But actual results inside phone with smartphone cooling and wattage requirement became close to 3300.
But perhaps the most important thing, Xiaomi is now able to make chip similar to Mediatek. They are the third largest Smartphone manufacturer by shipment volume. And they are still growing.
Missing the most important metric: processing power per watt.
I have some server CPUs laying around that can also beat Apple CPUs. But putting them in a compact, densely packed, sealed box (aka a phone) would result in a fireball and almost no battery life.
This is nothing. China is about to start doing their own 5nm. When they roll out their own silicon at scale, I don't think there's catching up, nobody can match their manufacturing capabilities. Not even close.
Xiaomi has not been sanctioned like Huawei. So it is made by TSMC on N3P.
The Huawei chips, made by SMIC, are so far behind they're not competitive at all with even Xiaomi's first generation chip.
Is raw perf considered Apple's moat? I think it's their software. I would buy a MacOS laptop even if the performance is ghastly ... in fact, I did: the MacBook Neo.
Yes it is. I've bought a MacBook air despite despising MacOS (with the hope that Asahi support will come eventually) because the processor currently has no competition.
This smells somewhat legitimate and somewhat BS. The architecture itself looks impressive — TSMC 3nm, a 4+ GHz prime core, tons of cache, and a very wide execution engine. And the claimed Geekbench numbers — around *3,945 single-core and 15,221 multi-core* — are especially impressive, if not a little suspicious.
But it’s sorta like, let’s throw everything against the wall and get the biggest number we can. Do they show sustained performance over ten minutes? What’s the power draw when it’s doing 15K? What happens when you put it in an actual phone?
This was done in a cold room. It seems more like a proof of concept for what they could do than something that could actually ship in a phone and sustain those numbers.
It will be interesting to find out the details. From the block diagram, the core only issues 10 instructions per cycle, so not especially wide. Maybe the performance will be worse once in a real phone in real thermal conditions, but if that's true, it could make for a very interesting laptop or desktop chip. Time will tell.
The C1-ultra has a decode of 10 instructions but can issue up to 23 uop per cycle, that’s a good bit wider than any other CPU. In particular, I believe it’s the first CPU to have a 4th load pipeline.
> The biggest cores on the the Xring O3 are the C1-Ultra. C1-Ultra really powerful cores. They support SME2 (Scalable Matrix Extension 2) for matrix/AI acceleration, SVE2 for data parallelism (SIMD).
This is where the magic is happening (I believe). They have these ultra cores with additional extensions which they are pinning the benchmarks to.
Realistically, that sort of thing isn't useful as non-homogeneous architectures really suck for an application or OS to deal with.
> Realistically, that sort of thing isn't useful as non-homogeneous architectures really suck for an application or OS to deal with.
I agree that if each core doesn't "support" the full ISA it's a huge pain (the one generation of Intel AVX-512-but-not parts were incredible), but that hasn't been common in years. Usually (and in this case, according to my reading) the small cores get a horrible slow/decoder-emulated implementation of every instruction, so that if you schedule on the "wrong" core you don't trap but just run really slowly instead.
Otherwise, heterogeneous architectures are the present and future of mobile and consumer desktops and have been for years. Desktop OSes are getting better and mobile OSes are just fine at scheduling across different core families (the higher level of abstraction everyone complains about in mobile apps actually helps a lot here).
All 10 cores support the same extensions. It’s silly to implement an SoC any other way, and it’s been a decade since someone has last made that mistake.
> But it’s sorta like, let’s throw everything against the wall and get the biggest number we can.
From what I recall reading, the impressive performance of Apple mobile CPUs is partly attributable to throwing more silicon area at the problem while not being as cost sensitive (no need to share margin with SoC vendor)
It depends on the number of channels. The Xring O3 appears to have 4×24-bit channels, so 113.8 GB/s. The iPhone 17 Pro's memory bandwidth is ~76.8 GB/s. Seems fine?
If I recall correctly, Strix Halo gets about 450G/s. The M series processors get between 400 and 800G/s, and an actual NVIDIA card is like 5.5T/s. 113G/s is pretty slow for inference.
Ignoring the native vs java runtime interface application difference between Android and iOS, which the top poster was was referencing.
Your Linux phone statement is a fallacy.
Ask FuriLabs how many of their phones they have sold.[0] High confident it is more than 10.
FuriLabs moved their batman, battery monitoring application, from Python to C and it really improved performance. Native applications really do keep resource usage low and is beneficial.
I think they might be saying that there's still no supported way to build a native-code app on the Android OS. It could be construed as a plea to make that happen in order to unlock greater performance, not necessarily a desire to have a DIY Linux-based OS option.
I believe the comparison is with iOS not Linux. Android is Linux after all. It’s just that most programs don’t get to run a that level of abstraction, and instead run on top of a jvm. At least in iOS the code does seem to be able to be more efficient due to to higher usage of natively compiled code, which is what I think the op was complaining about.
I'm not sure that it matters. You can't put a Qualcomm chip in an iPhone and you can't put Apple Silicon in any other device. So there's no really straightforward way to compare them other than specs on paper, which we all know are theoretical until they get integrated.
I can feel the scores of messages being sent to the federal government from lobbyists saying what a threat Chinese hardware is to national security.
It's an ARM-designed CPU core, made in Taiwan. Very little difference from say MediaTek's upcoming Dimensity SoC just announced earlier because it is made on TSMC N3P instead of N2.
I don't think they have any angle with that.
I'm not one to cry "foreign bots!" but this thread seems astroturfed after realizing this.
Xiaomi doesn't do business in the US anyway.
I wonder if the Trump T2 phone will get this chip.
Man I forgot about that nonsense. I can’t imagine any self respecting person picking one up.
3 replies →
same playbook than Huawei
I mean, yeah?
It's generally a bad idea to use information-handling hardware from your main geopolitical rival.
And everybody forgets the true power of Apple: scale. Can that supposed miracle chip be readily found in million of phones within 6 months? Or will Xiaomi keep selling 90% of their phones with 2 year old chips, reserving these chips to their 2000$ foldables?
This post serves Xiaomi because it implies competitiveness when the reality on the ground will remain unremarkable, and it helps fearmongers because they can latch on to Xiaomi's attempt to appear competitive.
Their power of scale applies to less and less regions. Xiaomi, in particular, delivers that to China where Apple previously slapped, now is becoming a niche player. It's the same as with German car makers. The trend now reversed with Chinese manufacturers starring to dominate on foreign markets and it's also something that was unfathomable just a few years back.
They have proven countless of times by now that they are not to be ignored.
When I buy my phones I do not give a flying fuck about "scale". All I care is what do I get for my pesos from various offerings.
1 reply →
> I can feel the scores of messages being sent to the federal government from lobbyists saying what a threat Chinese hardware is to national security.
The motivation for such messages may be greedy, but it doesn't mean the facts aren't wrong:
1. As a practical matter, no one can audit systems so thoroughly to determine a system built by an adversary won't act adversarially.
2. Free market incentives work counter to national security all the time, in many ways (e.g. causing domestic capabilities to wither and die).
Can we audit American corporations?
15 replies →
To make conversation productive, please define "national security" in this case.
For example, if China can't install its hardware in Pentagon, why is it a threat to national security.
On the other hand, if you say, Intel will be in trouble because of these chips, then I would say, so what? Wasn't this a dream of MBAs, educated by the top US schools, to outsource everything in business to cheaper service providers (e.g. manufacturing to China, software development to India and etc,.)
Where is the threat and what is the threat?
1 reply →
3. China was not stupid enough to rely solely on the free market to develop strategic industries. A few deliberately old sources:
Back in 2014, the central government set up the China National Integrated Circuit Industry Investment Fund. - https://www.scmp.com/tech/enterprises/article/2145422/how-ch...
Over the next decade, China plans to invest up to RMB 1 trillion (US$161 billion) into its domestic semiconductor industry as it seeks to become a global power in the sector. In recent years, Beijing has repeatedly stated that developing its own integrated circuit (IC) industry is of integral national strategic concern. - https://www.china-briefing.com/news/chips-all-in-investing-i...
https://en.wikipedia.org/wiki/China_Integrated_Circuit_Indus...
1 reply →
[dead]
OH, BUT we have fable! We have AI! of course we can answer these questions? can't we? which is true: America's AI is a super powerful defense against our enemies or America's AI is too stupid to defend against such a threat, and only sactions and import controls will suffice.
Just seems like you're inventing things to be scared of.
1 reply →
I love this, really, competition on a global scale is good for us, bad for greedy US corporations.
For example, I consider Meta as bigger threat to society, than Xiaomi, DeepSeek, z.ai and other top Chinese companies (haven't used Chinese social media, so can't comment about its content)
This is still an ARM C1-Ultra. Same one used in Mediatek dimensity 9500, which also got more than 4000 points in GB6 lab test. But actual results inside phone with smartphone cooling and wattage requirement became close to 3300.
But perhaps the most important thing, Xiaomi is now able to make chip similar to Mediatek. They are the third largest Smartphone manufacturer by shipment volume. And they are still growing.
This is bad news for Mediatek and Qualcomm.
> This is bad news for Mediatek and Qualcomm.
And good news for everyone else.
Missing the most important metric: processing power per watt.
I have some server CPUs laying around that can also beat Apple CPUs. But putting them in a compact, densely packed, sealed box (aka a phone) would result in a fireball and almost no battery life.
https://xcancel.com/lemire/status/2091894299289874926
https://addons.mozilla.org/en-US/firefox/addon/xcancel/
And on iOS?
(Also I'd recommend https://libredirect.manerakai.com/)
3 replies →
M5 (iPad) AnTuTu: 3.5 million
XRing O3 AnTuTu: 5.5 million
---
M5 iPad Geekbench Single-Core: 3,556
M5 Max Geekbench Single-Core: 4,300
XRing O3 Geekbench Single-Core: 3,945
---
M5 iPad Geekbench Multi-Core: 15,285
M5 Max Geekbench Multi-Core: 29,200
XRing O3 Geekbench Multi-Core: 15,221
I dont think Geekbench is a good cross-platform benchmark, given what it measures
This is nothing. China is about to start doing their own 5nm. When they roll out their own silicon at scale, I don't think there's catching up, nobody can match their manufacturing capabilities. Not even close.
I'm curious as to the performance per watt. There's no such thing as a free lunch.
no free lunch for a given process node I guess. It's using TSMC’s 3nm N3P.
Also note that CPU cluster here is about ~30.5 sq mm. That's about 50% larger than the A19 Pro (~20 sq mm).
No free lunch.
44mb of cache really highlights the bottleneck of how raw compute is cheap, but the real limit is memory latency.
"matches" could've also been written as "this year's product falls short of the competitor's last year one".
And multithreaded is 10 cores vs 6 cores.
I appreciate Apple being kept on their feet, but they're not dethroned yet.
Yeah but they can't use TSMC to make it so they'll have to make due with older or more expensive lithography processes.
That probably explains why there's so few of these CPUs around: the yields of the domestic Chinese lithography processes (non-EUV) are awful.
Xiaomi has not been sanctioned like Huawei. So it is made by TSMC on N3P. The Huawei chips, made by SMIC, are so far behind they're not competitive at all with even Xiaomi's first generation chip.
Xiaomi can and does use TSMC.
Yay free markets China makes a fast chip Yay national security
2020's US geopolitical doctrine:
1) Pick a historically extremely innovative and disciplined culture which also happens to have massive human capital
2) Please arbitrary artificial ill-thought-out economic constraints on them
3) wait a while
4) get a bunch of magnificent spanking in return (AI, chips, etc...)
Nice. Let's see if they perform well in the real world.
RIP Mediatek, RIP Qualcomm
"Arm-designed CPU core matches Apple-designed CPU core from a year ago"
Is raw perf considered Apple's moat? I think it's their software. I would buy a MacOS laptop even if the performance is ghastly ... in fact, I did: the MacBook Neo.
Yes it is. I've bought a MacBook air despite despising MacOS (with the hope that Asahi support will come eventually) because the processor currently has no competition.
The general workload performance (browsing, text editing) is comparable for neo and m* MacBooks afaict.
Ghastly was macOS on an Intel MacBook Air.
This smells somewhat legitimate and somewhat BS. The architecture itself looks impressive — TSMC 3nm, a 4+ GHz prime core, tons of cache, and a very wide execution engine. And the claimed Geekbench numbers — around *3,945 single-core and 15,221 multi-core* — are especially impressive, if not a little suspicious.
But it’s sorta like, let’s throw everything against the wall and get the biggest number we can. Do they show sustained performance over ten minutes? What’s the power draw when it’s doing 15K? What happens when you put it in an actual phone?
This was done in a cold room. It seems more like a proof of concept for what they could do than something that could actually ship in a phone and sustain those numbers.
It will be interesting to find out the details. From the block diagram, the core only issues 10 instructions per cycle, so not especially wide. Maybe the performance will be worse once in a real phone in real thermal conditions, but if that's true, it could make for a very interesting laptop or desktop chip. Time will tell.
The C1-ultra has a decode of 10 instructions but can issue up to 23 uop per cycle, that’s a good bit wider than any other CPU. In particular, I believe it’s the first CPU to have a 4th load pipeline.
Were I to guess
> The biggest cores on the the Xring O3 are the C1-Ultra. C1-Ultra really powerful cores. They support SME2 (Scalable Matrix Extension 2) for matrix/AI acceleration, SVE2 for data parallelism (SIMD).
This is where the magic is happening (I believe). They have these ultra cores with additional extensions which they are pinning the benchmarks to.
Realistically, that sort of thing isn't useful as non-homogeneous architectures really suck for an application or OS to deal with.
> Realistically, that sort of thing isn't useful as non-homogeneous architectures really suck for an application or OS to deal with.
I agree that if each core doesn't "support" the full ISA it's a huge pain (the one generation of Intel AVX-512-but-not parts were incredible), but that hasn't been common in years. Usually (and in this case, according to my reading) the small cores get a horrible slow/decoder-emulated implementation of every instruction, so that if you schedule on the "wrong" core you don't trap but just run really slowly instead.
Otherwise, heterogeneous architectures are the present and future of mobile and consumer desktops and have been for years. Desktop OSes are getting better and mobile OSes are just fine at scheduling across different core families (the higher level of abstraction everyone complains about in mobile apps actually helps a lot here).
All 10 cores support the same extensions. It’s silly to implement an SoC any other way, and it’s been a decade since someone has last made that mistake.
It is just the same as MediaTek Dimensity 9500. They managed 4000+ point in lab, but real world it was much closer to 3300.
It is still not a bad results though. And we are about to see MediaTek Dimensity 9600 based on ARM C2-Ultra.
Doesn’t matter. Details don’t matter. Look at the top comment thread: just politics and no discussion of the tech.
> But it’s sorta like, let’s throw everything against the wall and get the biggest number we can.
From what I recall reading, the impressive performance of Apple mobile CPUs is partly attributable to throwing more silicon area at the problem while not being as cost sensitive (no need to share margin with SoC vendor)
LPDDR6... that's not enough for speedy inference
It depends on the number of channels. The Xring O3 appears to have 4×24-bit channels, so 113.8 GB/s. The iPhone 17 Pro's memory bandwidth is ~76.8 GB/s. Seems fine?
If I recall correctly, Strix Halo gets about 450G/s. The M series processors get between 400 and 800G/s, and an actual NVIDIA card is like 5.5T/s. 113G/s is pretty slow for inference.
1 reply →
Oh thank heavens, computing hardware that the AI companies will not buy all of the supply of.
Not every computing platform is expected to do fast inference.
Maybe,but memory bandwidth is much better than on the DGX Spark at least.
I hate this RAMpocalypse so much
1 reply →
Insane if true. Does Xiaomi sell laptops?
Yeah, but they're just normal Intel stuff.
I have Mi Notebook Pro. Not that bad actually.
But they have stopped making premium laptop since then.
What! I was told China is behind and full of dangerous malware! Everyone knows corrupt little communists can’t make a good chip! /s
Sad thing is that it'll still be running the damn Android and so JVM not C or Swift on the machine.
What's the point of the complaint? Is there anything better that is commercially viable?
Your Linux based dream phone is not going to sell more than 10 units.
Ignoring the native vs java runtime interface application difference between Android and iOS, which the top poster was was referencing.
Your Linux phone statement is a fallacy.
Ask FuriLabs how many of their phones they have sold.[0] High confident it is more than 10.
FuriLabs moved their batman, battery monitoring application, from Python to C and it really improved performance. Native applications really do keep resource usage low and is beneficial.
[0] https://furilabs.com/
I think they might be saying that there's still no supported way to build a native-code app on the Android OS. It could be construed as a plea to make that happen in order to unlock greater performance, not necessarily a desire to have a DIY Linux-based OS option.
I believe the comparison is with iOS not Linux. Android is Linux after all. It’s just that most programs don’t get to run a that level of abstraction, and instead run on top of a jvm. At least in iOS the code does seem to be able to be more efficient due to to higher usage of natively compiled code, which is what I think the op was complaining about.
Apple's *Last Year* CPUs. We're literally about to get a new chip in 2-3 weeks from them.
The real concern is supply chains. The US can design but not manufacture, whereas it appears China can do both.
In this case, a Chinese company is using a Taiwanese manufacturer (TSMC), so they're not doing both.
> whereas it appears China can do both
Not these. They seem to be using ARM-designed cores that others already use and TSMC fabs in Taiwan.
Still better than what Qualcomm is offering?
I'm not sure that it matters. You can't put a Qualcomm chip in an iPhone and you can't put Apple Silicon in any other device. So there's no really straightforward way to compare them other than specs on paper, which we all know are theoretical until they get integrated.