Loongson releases next-generation CPU
ecns.cnI’d love to get my hands on one of these. They’re MIPS machines that are on-par with the Intel core i3 14100. Loongson is also involved with Deepin Linux, so they’re attempting to build out a Chinese-native technology stack. While I generally don’t care for things under control of the CCP due to surveillance concerns, I do love competition for bringing about new tech innovation. We’re now getting back to the 80s/90s as far as diversity with AMD64, MIPS, RISC-V, ARM, OpenPOWER, and then semi-custom variants of RISC-V and ARM. It’s fun.
It is the golden age if you are a Chinese chip designer or operating systems engineer; getting to reinvent/rebuild the whole computing stack.
Finally, the chance to build CCP/IP
That sounds like it would be an absolute blast.
The performance they quote also seems to be in x86 compatibility mode, I wonder if you can bypass that and compile for the native architecture.
It's Loongarch, some sort of MIPS + RISC-V.
Wikipedia [1] has some more details, but not too many on the main page itself.
Also from the top of that page I learned that "Loongson" means "Dragon Core" which to me as a westerner makes it sound very clearly Chinese (because dragons) but also a lot cooler (because ... dragons!). Nice.
The loongarch spec
https://loongson.github.io/LoongArch-Documentation/LoongArch...
Chinese dragon = snake with four legs. It can fly (magically). Western dragon = dino with wings. It can fly.
They're unrelated things. A Chinese dragon is more like a unicorn than a western dragon.
don't both of them breathe fire, though?
In chinese mythology, long are also control for rain.
Associating China with dragons seems like such a lucky PR win to me. How many countries are, outside of that country, associated with any mascot at all, let alone such a badass one?
Also it seems appropriate that the Western concept of a dragon is much less friendly than the Chinese one -- at least, judging from the dances.
The Welsh have their own dragon (on their flag)
This was the reason for Dragon computers btw.
Only one that comes to mind is the american eagle lmao
MIPS asm, but not MIPS machine code. Different bit encoding.
If they can keep up with the AMD and Intel CPUs they list, then that's actually pretty impressive and actually useful. There is a lot of people still daily driving 5 - 10 old processors just fine, it's absolutely plenty for a office desktop or even light development work.
I would like to see actual benchmarks though.
I use a core i7-920 from first generation that I recently changed with a Xeon compatible socket and another i7-3770 from third generation as daily drivers. And I do some gaming with non-intensive games. The computer works perfectly with no major drawbacks.
The third generation is noisy because it is in a small format case and with very deficient ventilation that I can't solve.
Did you try to at least remove the dust/dirt which tends to clog those parts over these timespans? With forced air from a pressured can to blow it out, or the other way around with a vacuum cleaner? (But arrest the fan, or unplug it while doing so, if possible! Otherwise it can spin like crazy and working as a generator, even frying the controller!) Furthermore the cooling paste/glue/mat between the die, or the lid of the CPU and the underside of the cooler tends to degrade/dry out, also because of time. Exchanging that requires disassembly, removing the old stuff carefully with isopropanol, reapplying some thermal paste, nothing special, just Arctic MX4 is good enough, and then reattaching the cooler.
Would work wonders, I guess.
I had changed the thermal paste last year, and I think it had still the original thermal paste (completely dry) and it was improved a lot. Like going from 90 degrees to like about 70 degrees.
But the machine is still noisy. I also got a nvidia 1050ti small format into it, and it completely breaks the small flow path that exist. The flowpath is provided with the cpu fan and a plastic piece to redirect the flow.
Lately is having some very nasty blue screen of death. That when I look online they say might be produced by the RAM failing. But running mem86+, one stick at a time and both stick together, always pass with 0% error. So I am getting out of ideas.
> actually useful
Except it's a custom ISA (fork of MIPS). It could be fine for Chromebook-like use cases (basically web browser machine), but not really for anything more serious ...
I guess you mean on the desktop? If this thing runs linux, it is as serious as everything that can run Linux.
OpenBSD has a history of supporting the Loongson arch. https://www.openbsd.org/loongson.html
Only some Chinese distros support it. For a lot of serious things, you will need to have binary support - commercial software, docker etc.
That's why they have:
LoongBT, faster x86 and ARM binary translation, 213 instructions
I remember the Loongson MIPS laptop made headlines many years ago. Richard Stallman even used one for a while.
It got a lot of attention among my hacker space fellows, but I don't remember why. Was it an open hw design or something?
I believe the BIOS was open or something like that. In general, though, it is hard to have a meaningful consistent position here because you can have most of the software be open and some random peripheral with closed firmware can DMA all over everything it wants :/
That issue can be dealt with using an IOMMU under host control. Those limit where the peripheral can DMA regardless of its firmware.
Isn't this just an instance of "trusting trust?" How do you know the IOMMU hasn't been backdoored? "Open" firmware doesn't mean open RTL. Where is the line drawn?
Whether you want open firmware in the first place is a significantly different question from how you isolate hardware with closed firmware.
But more directly, worrying about one part having a backdoor is a lot better than worrying about twenty parts having a backdoor.
Just an example, DARTs and IOMMUs help close down that line of attack but there are still many proprietary and inscrutable blobs/peripherals/monitors that alter the behavior of modern computers which are almost impossible to avoid in general.
I used to have one as my only laptop for years and I miss it dearly. Built quality was above average, keyboard was great, mate screen (not as rare back then though), easy to open, lack of hardware shenanigans and usable/open source BIOS, no proprietary drivers required, came with Linux preinstalled. Reasonably fast.
It was a MIPS little endian, which is also very good. I can't remember if it was fanless or not.
Ten years ago it was kinda the only x86 alternative available in a laptop form factor, but was otherwise unremarkable except for being quite buggy.
I don't know if this is statement accurate, but from the (archived) manufacturer's website:
"The world's first fully free software. All system source files(BIOS, kernel, drivers etc.) are free software, no close firmware needed." [0]
[0] https://web.archive.org/web/20160703160118/http://zkml.lemot...
Any chance systems based on it will be sold outside China?
You can buy systems based on earlier Loongson CPU's from Ali Express right now if you want to. I can find dev boards, some small raspberry pi-like devices they call Loongson Pi as well as the tower-pc-like LS3A5000-7A2000 system.
They seem pretty expensive for how weak they are, so probably not much point in buying them besides to play around with the architecture.
The USA’s recent sanctions against the PRC has made Loongson forbidden if I remember correctly. Hopefully, this will change.
I was under the impression that the sanctions worked the other way around, i.e. Loongson can't buy a lot of things from American companies.
> Entities on the Entity List are prohibited from purchasing or licensing American technologies, even indirectly. For example, Loongson can't have its CPUs manufactured by American equipment, ruling out most foundries with modern nodes.
According to https://www.techspot.com/news/97817-us-blacklists-china-loon...
The sanctions are here to stay. The USA wants to make China capitulate. There is an economic war against China. When you cannot win by merit, you invent excuses to slap sanctions.
The sanctions also seem to have done a great job accelerating Chinese mainland semiconductor technology. Before the sanctions most mainland companies were happy having their chips fabbed by TSMC
> The sanctions are here to stay. The USA wants to make China capitulate. There is an economic war against China. When you cannot win by merit, you invent excuses to slap sanctions.
There are two types of games: finite games and infinite games.
Finite games have fixed rules, a winner and loser, etc.
Infinite games don't have fixed rules, conventions are just conventions to be tested/broken, and the goal is not to win but to stay in the game for as long as possible.
Most games in the real world are infinite games. "Winning by merit" is a finite game not an infinite one.
Other way around, until last month China officially banned most of the loongson chips from being exported. The ban was reportedly lifted due to interest from Russian companies in using the chips. https://www.tomshardware.com/news/loongon-cpus-bound-for-rus...
Fitting, because https://en.wikipedia.org/wiki/Baikal_CPU didn't really happen.
Why do they use 12nm process node as opposed to SMIC 5nm or 7nm? Too expensive?
Does someone know what reference size to take for those SMC nanometers?
It may sound irrational to ask this kind of question, but marketing killed the meaning of "nm" a long time ago. As reference, through electron microscope the sizes of Intel 14nm seems equivalent to AMD.TSMC 7nm [1]
[1] https://www.techpowerup.com/272489/intel-14-nm-node-compared...
They're aligned to TSMC terminology.
Huawei probably has first dibs on capacity given their contribution to SMIC 7~5nm process.
As far as I know, it's because China does not have access to the latest fab equipment from ASML and have to use an older technology to manufacture their chips.
I think it's down to the technological challenges that involves and they are just not there yet.
So far it appears that Huawei is the only Chinese firm producing chips on 7nm; while western commenters have stated that it is a SMIC process - I don't actually think they know for sure as none of these companies have made it official - in other words the process may be proprietary to Huawei.
However 7nm is in volume production at a level so high that can do non-critical projects such as the Mate 60 phone rather than just the stuff where all imports to China are sanctioned (high-end AI chips, chips for spacefaring, and military).
Likely other sanctioned Chinese chip designers will use the process as it appears that Chinese authorities want internal competition rather than just have one single dominant company (Huawei).
>So far it appears that Huawei is the only Chinese firm producing chips on 7nm
True a good video about it from Asianometry
"China's 7nm Semiconductor Breakthrough"
Some of this is simply because until the latest sanctions, other Chinese chip designers could fab at TSMC at 7nm.
Also Huawei has a quite capable AI offerings:
But are you really willing to risk it all go into a TSMC based work flow just to end up on the US bad guys list.
I think even Chinese companies believed that Huawei was banned and sanctioned because they did something wrong and not just because they were Chinese and really good at what they do.
Now of course they are all sanctioned and Huawei is sitting pretty with years invested in a sanctions proofed supply line.
> I think even Chinese companies believed that Huawei was banned and sanctioned because they did something wrong
I'd be surprised to see anyone seriously thinking that way outside of the western bubble.
I personally can't even recall what exactly they were accused of. Probably something typical along the lines of "tech theft" or clandestine? I do remember all this happening during Trump presidency, who was already attacking other Chinese companies (e.g. forcing Tik Tok to sell shares). And that there was some heavy competition for 5G stuff at the time.
My impression (and definitely not mine alone) was that it was all about clearing the market.
SMIC 7nm doesn’t scale just yet.
It scaled well enough for the Kirin 9000s.
But that's probably why Loongson isn't on it, capacity is probably being spent on Huawei's next chip.
> It scaled well enough for the Kirin 9000s.
For a meager 1.6 million phones sold to date.
Again, 7nm is a fantastic milestone for a forge that does not have access to newer ASML equipment, but context matters.
1.6M chips is orders of magnitude more chips than Loongson has sold AFAIK.