Loongson CPU that performs like 2020 Core i3 makes its way to Chinese mini PCs Loongson's current-generation 3A6000 processor, one of the fastest designed and made in China for consumers, is now available in a line of mini PCs. The computers are made by Morefine, which terms its Loongson-powered small desktop offering the M700S. It has pretty standard specs that fit its ¥2,799 (just under $400) price tag …
I still use a ca 2012 i3 laptop with 8G ram which is quite usable with a variety of linux/unix installs so a 2020 i3 equivalent risc box might actually be an improvement although at USD400 probably not really a bargain.
If the processor design and firmware were open and auditable it might be more attractive but I suspect that even the PRC domestic version would be riddled with surveillance hooks, backdoors and vulnerabilites.
We better watch out for the Sony solid-state pocket radio ... oops, already happened ... and the inexpensive yet reliable Toyota Camry ... woah, already done ... and the textile facories for nearly every item of clothing ... same difference ... and a homegrown supercomputer ... we're Fugaku-ed!?
Competition with Japan worked out well in my view, bringing it out of its axis-of-evil past and into 21ˢᵗ century democracy ... couldn't competition with China do the same?
"Competition with Japan worked out well in my view, bringing it out of its axis-of-evil past and into 21ˢᵗ century democracy"
It may be worth including in your theory the fact that, during the period when all of the products you mentioned were being developed, Japan was already a democracy. China isn't. I don't know whether this changes your plans, but if your theory went country makes tech products > country makes good tech products > country becomes democratic, it can be worth considering that Japan didn't do those things in that order.
They would fit their argument more, sure, but since I think their argument is wrong, it doesn't convince me. You could propose all sorts of correlations and allege that they make democracy. For both examples, you could say that right-wing* dictatorships become democracies and left-wing* ones stay dictatorships, but there are plenty of counterexamples to disprove that. You could say east Asian countries become democracies and others don't, also easily disproven. There are dictatorships that produce tech products to some extent, and doing that does not guarantee that they stop being dictatorships. Using a bad example didn't help, but it was not the only problem with their argument.
* Right/left wing: Too simplistic, because there are usually long arguments about which side a given dictatorship is on by people who want to make sure it's not on their side.
While China is nominally communist, and South Korea was nominally "not communist", is there any significant difference in the economic approach China is taking now and what South Korea did about 60 years ago?
What South Korea did worked for them, up to a point, then they changed their approach. China's economic approach seems to be working for now, but it looks like they will encounter similar problems to what South Korea did, and it remains to be seen how they will deal with them.
Missing here is both the narrative and the protectionism that grew out of 1980s Japan eating our industrial lunch, so as to speak. We forced Japan to deflate and effectively stagnate -- they were allowed to compete but only on our terms.
China is a whole different game. Partly because its a whole lot bigger and partly because its not interested in playing our financial games (unless there's something in it for them).
Remember, what we call "freedom and democracy" is really NewSpeak for "financialization sucks the life out of an economy".
China is deflating right now. Japan's deflation lasted for 25 years.
You might be pinning your anti-west hopes on China, but China is submitted to the same laws of economics as other countries. Specifically, China's flavor of capitalism is not benefiting to its internal market. Neither huge state-owned nor huge private firms profits trickle down to small investors. Their appetite for domestic stocks was speculative and ended up being severely punished by a bearish market, itself sabotaged by the CCP. Both enterprise and consumers are very worried about the economic situation and are saving a large part of the GDP, which ends up invested in overcapacity. As long as these factors don't change, internal demand will be weak, global commercial war is guaranteed and, in case of failure, deflation it will be. Only complementary economies (such as Russia or KSA), characterized by natural resources and weak industrialization, will match Chinese hopes. Even BRICS (Brazil and India especially) will resent Chinese commercial war.
"They will improve."
They will, but the question is how quickly? If China goes ahead with banning Intel and AMD, even a strongly worded "soft ban", that will artificially reduce competition and without the need to compete, reduce the imperative to improve.
Having said that, in the commercial world, even a 4 year old i3 equivalent is more than fast enough for most users who are using local "office" apps, server interfaces and "cloudy" apps where there isn't a need for high CPU throughput.
It'll be interesting to see how heavily they lean into the RISC-V side of things going forward, since MIPS seemed to hit a wall the last time Loongson tried to push their own CPUs.
Given RISC-V's open architecture it may be more workable, but also given how wildly configurable RISC-V can be in implementation it could wind up being an overly complex boondoggle.
Loongson is really a modified MIPS clone, and MIPS is a clean RISC architecture. Linux and other free OS'es have supported it for a while, so arguable it is rather open, and likely to have less vulnerabilities like the complex Intel architecture, with its decades long trail of backward-compatibility. The interesting question for security is does Loongson have a separate closed system processor with "superpowers", like most Intel and AMD ones do. That is where one would hide backdoors, if there are any.
The limited number of reviews of the 3A6000 indicate it's only about as good as Intel's Core i3-10100
And that is faster than the cpu benchmark of any computer here.
c.f. my i7 zbook laptop, which tested today at about half of that i3 single threaded performance.
I basically considered it a fraud that my premium priced laptop "i7" turned out to have only two cores, and the TDP limit means only one can run at full speed for sustained loads, and the i5 single threaded benchmarks were higher anyway.
It won't run AI training very well, and even a lot of inferencing will need some more processing, but how many people do you know doing either of those things on a desktop? I know a couple people who have, but all of them used a dedicated GPU to do it, and those can be connected to this as well. Everyone else I know either doesn't use those tools at all or is happy enough to let someone's server do the heavy lifting. I somehow doubt that will prevent this chip being useful to the general office user. Software availability might be the harder problem, though by no means insurmountable.
> but how many people do you know doing either of those things on a desktop?
We're talking about two emerging trends that promises to become mainstream.
- AI taking everyday consumer computing by storm. Look at Microsoft/Apple and how they're steering the industry in that direction.
- China trying to decouple from Intel/AMD in the next 4/5 years. Decoupling for CPU and not for GPU makes zero sense.
Opposing the current situation as an argument is irrelevant.
"- AI taking everyday consumer computing by storm. Look at Microsoft/Apple and how they're steering the industry in that direction."
As a great philosopher[*] once said "why are we spending money developing artificial "intelligence when the money would be better spent on dealing with natural stupidity?"
(A Grogu meme, as reported to me by my wife just this afternoon :-))
Microsoft and Apple talk about AI a lot, but that doesn't mean people actually want to use it as much as they claim or need to do it locally. Microsoft's AI mostly runs in Azure, but that doesn't seem to bug too many people. A Chinese AI could easily run in Baidu's or Alibaba's cloud, accessed by these boxes, and they'll have a similar experience to people using cloud-supported AI software.
All of that, by the way, is inferencing. People do not do training on commodity small office computers like this. Few people do it at all, and those who do use much more powerful computers so it doesn't take weeks to get a complex model. Do you have a reason to think this is changing, and if so, I'd like to hear your proposed use case for the average user to be training models on their machines. Most users do not find training instructions that include statements like "create a virtual environment and install TensorFlow and Torch" to be easy to follow, and that's by far the easiest item in the list of steps for prebuilt ML software.
This 3A6000-based system would have been a much better one to give to 10,000 Chinese schoolkids than that based on the half-as-performant 3A5000 (IMHO), and will also be a much better target for one of Simon's upcoming Desktop Tourism destinations.
The chip tops-off at 2.5 GHz, needing liquid N₂ cooling beyond that, which puts it at a serious disadvantage relative to (laptop-oriented) x86 that readily turbo to 3.9 GHz (and more for Desktop chips). Also, the perf assessment is based on SPEC CPU 2006 which likely underweights dynamic language workloads (JAVA, Javascript, Python, ...).
I think I'll curb my enthusiasm a bit until a proper (i.e. ElReg) desktop tourist field tests this cute little M700S gizmo ... to see if it is really "Morefine" than homegrown varieties, on the yummy spectrum of up-in-smoke performance!
Yes it probably is more expensive to product at the moment, as China have less cutting edge fabs compared to the likes of Intel, and can't just outsource it to TSMC.
But it goes to show that they are only a few years behind the west now and for most general purpose computer use a 2020 era i3 performance is perfectly adequate. I daily drive a Ryzen 3 PC and rarely find myself wanting a more a powerful system, for what i use it for.
This post has been deleted by its author
Probably, yes, at the moment. Economies of scale and all that....
That's not the (overall) point though.
The point is to be able to service the (not insubstantial) Chinese demand with Chinese made kit of similar performance, which they'll probably be able to do in a few years if the current progress is anything to go by.
Not sure (a) where you're getting your prices from and (b) why you're comparing apples to oranges.
(A) The story claimed i3-10000 series compute, not Celeron.
(B) A quick research shows 16gb of DDR4 and 256gb of SSD are both $50 street; does the Celeron come with 16gb? Add in case, motherboard, the actual CPU, power supply, assembly, packaging, shipping and warranty and I don't see how even a new Celeron will be $130
(C) a perusal of Newegg shows i3-10000 series desktops going for -$440, so this Chinese unit seem right in three ballpark.
They were comparing the model of the computer with the same model, made by the same manufacturer, but with an Intel chip. That's why they were using a 2023 Celeron instead of a 2020 I3 as a comparison. I'm not them, but I can answer some of your questions.
"The story claimed i3-10000 series compute, not Celeron.": I've explained why they did that. From benchmarks, the N100 is a little slower than the I3-10100, though much more power efficient.
"A quick research shows 16gb of DDR4 and 256gb of SSD are both $50 [...] I don't see how even a new Celeron will be $130"
I couldn't find the specific model they were looking for, but those prices do exist. I found one from the same company with an N100, 12 GB of RAM, and 512 GB SSD for $144.06. The RAM is a bit lower, but the larger disk probably offsets that. From a different provider, I can match exactly, $129.98, 16 GB RAM, 256 GB SSD. These are from AliExpress. Add $20 and I can get you the same thing on Amazon (US Amazon so I can keep quoting USD), but it's Amazon, so if I search for twenty minutes instead of twenty seconds, I might be able to match it without the premium. So yes, you can get that spec for that price. I wouldn't want to count too much on that warranty.
"a perusal of Newegg shows i3-10000 series desktops going for -$440"
That may be, but there probably aren't many and they're probably not selling. The processor being old enough means that people can easily get a better box for the same price or a comparable one for cheaper. I can easily double the performance for that price.
China has more university places at an affordable cost, they WILL be matching Intel and Amd within the next 15 years.
The west will make sure their children are in debt for education, China will make sure their children are not.
The west will fail unless we change course.
A lot has happened since. Trump has managed to fracture GOP so badly that he can't count on Republican support the way he did in 2016. And that's not even mentioning all the trials he's a defendant in and that come with a real possibility he'll spend the rest of his days in jail.
> China has more university places at an affordable cost, they WILL be matching Intel and Amd within the next 15 years.
Wishful thinking. I've been hearing that for very long. Same for Japan in the 90s. Didn't happen.
> The west will make sure their children are in debt for education, China will make sure their children are not.
For now, what we observe is a Chinese elite that is fighting tooth and nail to get their kids in US universities. Not to mention Indians and others.
> The west will fail unless we change course.
"Believe me the end of the world is neigh" OK.
"For now, what we observe is a Chinese elite that is fighting tooth and nail to get their kids in US universities. Not to mention Indians and others."
Apart from the "world wide big name" universities taking only the "best of the best", most of the best Chinese students opt for the best of the Chinese universities. Those who don't get a place there start looking at the rest of the worlds universities. And as someone else pointed out, it's go to Uni or do your "service" in the PLA.
42W vs 52W (at 2.5 GHz for Loongson, not clear which speed for I3 that clocks to 4.3 GHz, but was downclocked for the perf comparison).
As it is a new CPU, the designers had the option to eliminate speculative execution in their CPU, and the hence eliminate the vulnerabilities which that feature brings. Did they? Or did they include speculative execution to increase the CPU's effective speed?
Speculative execution is a must have.
I don't like speculative execution because it just feels kludgey. It's only a must-have because nobody's yet (as far as we know) thought of a better way to get equivalent-or-better speed out of modern CPU/RAM combinations.
I'm pretty confident it's in there. Whether it is in there with vulnerabilities, known* or unknown, is a better question.
* Known vulnerabilities, not as in deliberate backdoors, but something they considered not important and left in the design. It happens a lot.
The OG LoongSon had ~100 extra (non-MIPS) instructions intended to make it easier/faster to emulate (via QEMU) an X86 machine. I am curious what is the status of that in these newer implementations, in both the HArdware and Software side...
¿Abandoned? ¿Deprecated? ¿As is? ¿Under active development?
If anyone Knows, please chirp in, thanks in advance.
Well, they are never going to improve their fabs if the path of least resistance is to just buy chips from the West.
Good for them. Not because I want China to rule the world, but because I don't like embargos and blockading. These are not exactly acts of peace either and are only going to cause more hostility. China plays a long game, it won't be a "Pearl Harbor" type thing you can prepare for with military might.
Also, the world will be a better place with more chip manufacturing. I'm sick of this artificial scarcity.
"Not because I want China to rule the world, but because I don't like embargos and blockading. These are not exactly acts of peace "
Product dumping and large government subsidies aren't exactly "fair trade" either, which is why import/export tariffs exist in the first place. It's a difficult balancing act even under the best of circumstances but when you have China producing huge amounts of the worlds goods and US style highly aggressive capitalism, it's starts to become impossible without tit for tat spats in all sorts of trade areas. That's without even attempting to consider the different political ideologies!
EUV lithography is hard. So hard that only one company in the world can currently make the machines, ASML. They combine the products of other companies (like Zeiss for the optics) into the final machine. China would have to make massive advances in a whole range of technologies in order to make one. It doesn’t matter how many highly educated folks they can throw at a problem if they haven’t got the technology and manufacturing infrastructure to build on.
Don’t forget, it was fairly recently that China reached the point where they were able to make sufficient good ball points for pens, removing the need to import them. Even their “7nm” chips that they recently announced were created using imported DUV machines (and are probably too expensive to be viable due to the number of multi-patterning steps needed).
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2025
