79 comments

[ 3.0 ms ] story [ 141 ms ] thread
but its chips don’t make it out of the country because they’re largely used in products built for the Chinese government.

Very wise of them. Right now the UK needs to make a serious push to make ARM our primary choice, and fabbed here too. This is as strategic as nukes.

It's too late for that ARM is no longer a UK company.
The argument of national security goes a long way.
It certainly goes a long way towards redirecting money to favoured state companies and building protectionist barriers, yes. Whether it provides any actual security is another matter.
Are you aware of historical events like the 1982 Siberian pipeline explosion? It was allegedly caused by sabotaged software inserted by a Canadian company on behalf of the CIA.

Modern chips are certainly dense enough and have enough layers that you can hide Trojans in them. Buying e.g. routers with Chinese ICs is a credible threat to the US government.

> Modern chips are certainly dense enough and have enough layers that you can hide Trojans in them. Buying e.g. routers with Chinese ICs is a credible threat to the US government.

Why bother hiding the malware in hardware? Most electronics are made in China these days anyway, and only a very, very small fraction of users ever replace or investigate the firmware on their devices.

Much easier and cheaper just to load a compromised firmware during manufacture and snoop that way.

> …only a very, very small fraction of users ever replace or investigate the firmware on their devices…

Remember that the users we are talking about here are US government agencies. Yes, they do investigate firmware. Yes, they do code reviews of your firmware source code. No, they won't let you just ship them a Chinese router with factory firmware on it.

There's a strange little collaboration between BT, GCHQ and Huawei going on at "Adastral Park" (Martlesham Heath). It's all top secret, but the person who told me about it hinted that it wasn't about removing the surveillance backdoors but making sure that GCHQ had access to them.
The explosion happened because Soviets stole control software and tried to run it on their system.
The software was not stolen, it was purchased from a Canadian company. The explosion happened because of the CIA's sabotage efforts.
It works as well as "think of the children".
You mean ARM Holdings plc, a wholly owned subsidary of SoftBank Group Corp.?
Well, quite. Didn’t the US block the sale of MIPS to a Chinese buyer?
It wouldn't really matter anyways since the Chinese have been making MIPS64 chips for more than a decade.

The current high end chip in China is actually new arch based on the DEC Alpha.

(comment deleted)
Statist nonsense
Even before Softbank bought them, ARM weren't even a fabless operation: their whole business model is IP cores. There's very little strategic value in that; you'd be as well off founding a semiconductor manufacturer to build RISC V processors.
Drumming up national security for the UK? I guess if you're planning to go to war with the EU... Otherwise it's as useless as Trident.
It's a joint venture with municipal government of Shanghai as the article says.

Zhaoxin Semiconductor is Chinese government funded and part of the National Integrated Circuit Industry Development Program.

Ah, the old VIA chips, which VIA built from the tech they bought from Cyrix. If you're interested in the history of Cyrix and some of the older x86 compatible chips, here's a great video on it:

https://www.youtube.com/watch?v=iWGAdoMz1c0

Also these Zhaonix's are just 32-bit right? I didn't seen anything on the site about supporting x86_64.

No, they mention 32/64 on the specs pages.
VIA's CPUs were from their purchase of Centaur Technology, not fro Cyrix, right? (IIRC there is a HNer that worked at Centuar for a while, perhaps they can expand).
My first PC had a Cyrix CPU, a P166+, running at 133MHz. PC vendors building middle- to low-end PCs liked them, because they were cheaper than Intel, and for most stuff people did with their PCs back then, it offered acceptable performance. Unless you wanted to play Quake, that is. ;-)

But the integer performance of these chips was rather impressive.

The other problem was that Cyrix CPUs used more power than Intel's, so they generated more heat and therefore needed a better cooling system. PC vendors often used cheap low-end fans that were well-suited for Pentiums, but caused many, many crashes due to the system overheating. Once I replaced the fan with a bigger one, the crash-rate dropped by about 90%.

My parents bought a Packard Bell PC as the first family computer, back in 1999 or so, that had a Cyrix, 333mhz from what I recall.
The page says "Zhaoxin's home-grown x86 CPU is completely designed from the ground up within China", but other sources suggest that their cores are derived from Centaur microarchitecture. Anyone know what the truth is on this?
According to the official One China Policy the island of Taiwan is a province of China. So yeah, maybe it all depends what your definition of "China" is...
VIA bought Centaur, so you can argue that, tenuously. VIA is a Taiwanese company. Some people insist that being Taiwanese is being Chinese.
Taiwan insist it's Chinese. Its government claims to be the rightful government of the whole of China.
The People's Republic of China and the Republic of China both agree there is a single China which includes Taiwan. They differ over who should govern it.
Not all Taiwanese are pro republic of China. There is a huge split in Taiwan between mainlanders (post 1949) and natives (also descended from the mainland, just pre 1949).
Hans started moving to Taiwan in the 1600s, then you had that big influx as the Chinese Revolution wound down.

There's also Native Taiwanese - https://en.wikipedia.org/wiki/Austronesian_peoples, they even make up about 2% of the population, I think they all want the Hans to pickup and leave their lands.

The native natives are more aligned with the KMT (ok, super aligned). The Hans before 1949 are mostly from Fujian, who are not known for being the most patriotic (and indeed, became dominant in overseas Chinese communities in SE Asia, NYC, and so on...). Post 1949 was the rich Chinese and KMT officials from everywhere escaping the communists, who also own much of the island's economy and want closer ties to China (so they can make more money).
The Republic of China probably agrees partly because waiving their territorial claims could trigger military action through China's Anti-Secession Law.
Even granting the considerable stretch that the VIA acquisition somehow renders part of Austin "within China" (I'm reminded here of the Babylon 5 episode "Day of the Dead"), I have my doubts that they'd throw away all pre-acquisition IP just to be able to make this claim. I guess they might have done so regardless to develop something with a newer workflow (e.g. for better verification), but my understanding is that it's not unusual to ship ~20-year-old HDL cores (e.g. UARTs) that are just synthesized with different gate libraries.
> I have my doubts that they'd throw away all pre-acquisition IP just to be able to make this claim

China doesn't seem to be above claiming that they "domestically developed" technology that they really acquired from foreigners. An example I'm aware if is China's "home grown" 3G cellular protocol called TD-SCDMA. It was really a failed European standards proposal from Siemens, but passed off as an indigenous Chinese development for prestige reasons.

See https://www.samizdata.net/2012/06/here-is-the-rac/

But if you go back far enough, that tech actually came from Cyrix, an American company. I highly doubt they rewrote everything.
Taiwan IS part of China.
>"Zhaoxin's home-grown x86 CPU is completely designed from the ground up within China"

Yes, just not that China, another one

Well, given the territorial claims, yes, kind of that one. It just depends which government you believe as to whether the island of Taiwan is occupied territory, or everything else in China is occupied territory.

And these are governments we're talking about, so, needless to say, both sides killed millions to make this happen. Not because those people that got killed worked for the other side or were doing anything at all (not even protesting most of them), but merely because they wanted to frighten the rest enough to not do anything against them.

One of the nicknames for one of the actions of the Taiwanese government was "the White terror" (death toll: at least 10000, there had been a protest march against them, so naturally ...), the mainland Chinese government had a policy that each little town had to have at least one public execution, and millions died.

https://en.wikipedia.org/wiki/History_of_Taiwan#Republic_of_...

It’s worth noting that while the government in Taiwan still officially claims all of China, the government is currently led by a president (Tsai Ing-wen) whose party does not whole-heartedly endorse that position.
Ya, see the difference between the DPP and KMT. Chiang Kai Shek’s legacy is also very dubious amongst many Taiwanese (especially in the DPP), who don’t align with the communists either (and let’s not get into Japan).
Whatever they claim, in the present time reality the RC (Taiwan) and the PRC are 2 distinct independent countries that are not going to re-unite any soon. "Believing" any of the 2 just doesn't make much sense.
Also, as far s I understand, the division has began as a conflict of 2 political parties - the leftists (CPC) and the rightists (Kuomintang) that has escalated into a civil war. And as the result each party now controls its part of what was once meant to be a single country and claims the right to rule both. What makes me curious is why can't 2 parties just control their parts the way they do in Taiwan and PRC but do it peacefully and without nonsensical claims when the people of one part are more sympathetic to one and the people of the other part are more sympathetic to the other. Why is it so important to have one party to rule them all, including those who doesn't want it?
I believe this is just a bad translation from a popular Chinese propaganda (自主研发).
They say it has a new microarchitecture, pipeline reduced by 5 stages, new branch prediction, new support for multi-chip-modules, and new cache coherency protocol to support that.

It seems as much a ground up design as Zen.

http://www.eefocus.com/mcu-dsp/399906/r0

I surprised most cpu chips don't have government back-doors in them.
Those mentioned are surely just convenience features ;) to believe that a single intelligence agency might exploit them in some way is clearly a conspiracy theory! ;) ;)
Hasn't it been established that most chips have backdoors? I remember reading an article about that HN a while back. That's why china, russia, etc are trying to develop their native cpus. Also, isn't this why sensitive US agencies don't allow chinese made hardware?
Remember export-quality encryption? Yeah...
Do they have meltdown and/or spectre bug?
They have spectre, any CPU with speculative execution and caches of any sort (excluding some capability machines, no ROP exploit machines and PLB enforced SASOS machines), including branch prediction (!) is vulnerable to spectre. (For those of you not in comp arch, that roughly translates to "literally every single non embedded CPU, and then some")

That's what makes the flaw so deep and troublesome.

What about Atom? As I understood it Atom processors did not use speculative execution.
That was only the case until 2013 or so. Modern Atoms are vulnerable.
And the old Atoms were so slow that few people bothered with them.
There were tens of millions of EeePCs sold. At one point Asus EeePCs were outselling all Macbooks.
IIRC, the first generation Xeon Phi had cores very similar to Atoms too.
Those cores were a bastard child of P54C[0] cores with the beginnings of (though not instruction compatible with) AVX512 + 64bit support (original Pentium had neither of these). They were not like Atom cores at all.

The second generation Xeon Phi cores are pretty much based on beefed-up Airmont Atom cores.

0. https://en.wikipedia.org/wiki/P5_%28microarchitecture%29#P54...

Is the Mill architecture solving problems orthogonal to this?
The Mill doesn't do speculative execution, so I guess it should be safe from Spectre? It also doesn't have a normal TLB and a very different idea of memory protection, so perhaps it will be safe from Meltdown too.

At the end of the day, though, Meltdown and Spectre are both consequences of all the caching between a modern CPU and main memory, so the Mill will likely need to at least be careful.

Nah, the Mill does branch prediction, so it will still be in trouble. The Mill also does speculative execution, not OoOE but still speculative via predication.
Is speculative execution actually worth it? Can't compilers do the job and optimize execution sufficiently so we won't need speculative execution?

Anyway, "caches of any sort" sounds frightening: I can remember disabling all the caches on Pentium-3 to make it about as slow as 80386 so some ancient games that can't run on fast CPUs would run ok and be playable.

Speculative execution is a workaround for the fact that CPUs read and process data much more quickly than main memory can provide it. The CPU tries to trace as many code-paths as possible so it can discover which memory addresses are likely to be used in the near future and queue them up so they'll be ready in the local cache when they're needed.

Smarter compilers wouldn't really help, this is all about how the code is running for the next millisecond on this specific workload on this specific CPU, rather than how it's running today or on this cluster or in general.

IA64 would like to reaffirm that this indeed worthwhile.
I thought the issue is that the fallback doesn't clear the cache affected by the speculation but not speculation per se?
Yet another 80386 on steroids with no 64-bit and no SSE or a real modern CPU capable of running all the today OSes and apps efficiently?