China X86 Chips Hitting The Market

Last year, fabless chip maker Zhaoxin announced they were readying a multicore x86-compatible CPU. According to media reports, the chips are showing up on Chinese marketplaces like Taobao shipping around March.

The company is a joint venture between the Shanghai Municipal Government and VIA Technologies, a familiar name in the PC business. It makes even more sense if you remember that VIA bought Centaur who had built simple x86 chips and used the simplicity to add more cache that more complex Intel and AMD chips. These fell out of the hobby market, but they’ve still been pushing forward providing simple designs that are inexpensive and consume low power.

Unlike some older Centaur chips, the new Zhaoxin chip is superscalar and capable of out of order execution. It also supports modern instruction sets like SSE4.2 and AVX. With eight cores, the CPU comes in a few speed grades.

Our Chinese isn’t good enough to watch the video below, but the closed caption is in English and makes sense, so we are assuming it is accurate.

According to the video, the chip is directly mounted on a motherboard and the 16 nm, 2.7 GHz processor consumes 70 W of power and does not have an L3 cache, but does have an 8 MB secondary cache. The CPU integrates the traditional north and south bridge chips.

According to the video, the company is targeting the do it yourself market and the motherboard and chip specs will be a little different in the production model. China is pushing hard to develop independence from foreign technology. This seems like a major step.

 

95 thoughts on “China X86 Chips Hitting The Market

  1. I don’t know about the product, but I was kinda mesmerized by the language and its associated accent. It didn’t sound like Mandarin to me, but what do I know. Not much about the nuances of Mandarin :p

  2. Interesting, always like to see more competition in the x86 market.

    If they can knock out boxes priced less than consoles and capable of playing games like fortnite they’ll be onto a winner. Also if it can replace the guts of any beige box made since 2000 for cheap and run with middling AMD and Intel systems that will move product also.

      1. Sure, let’s set a date, 5 years after everyone begins to write such well structured, well commented code, that it’s trivial to port every last application, device driver and firmware.

        1. Agree with Feinfinger – the x86 architechure is abysmal.

          The PC industry needs to quit trying to make things compatible with older software. In fact, just try running XP on a modern PC – it won’t. So why are they still carrying this millstone, and still pretending it is all backwards compatible?
          What they need to do is push for a sensible architecture and use VMs to allow older software to run.

          Of course, then getting the VM to correctly access random hardware becomes the issue. ISA cards anyone? I have a few that sport unique functionality that never got redesigned for PCI, let alone PCIe.

          1. >Of course, then getting the VM to correctly access random hardware becomes the issue. ISA cards anyone? I have a few that sport unique functionality that never got redesigned for PCI, let alone PCIe.

            Most of the modern day PC have a TPM connector which uses a LPC bus. LPC is a reduced pin count bus for legacy ISA peripherals usually a chip that has floppy, PS/2, Parallel port, serial ports etc. The 4-bit serial bus protocol is well document and simple enough for to make a small CPLD to ISA adapter.

            The software side is more messy. Probably need an x86 emulator that give you direct access to I/O and memory without complain.

          2. “The PC industry needs to quit trying to make things compatible with older software.”

            Backwards compatibility is one of the best things PC architecture has going for it, why would you want to take that away?

          3. “Backwards compatibility is one of the best things PC architecture has going for it, why would you want to take that away?”

            It has *already* gone away. The whole backwards compatability thing is an illusion, which is what I was trying to say. XP is not the only software that will no longer run on a PC with this supposed “backwards compatible” achitecture.

            As others have noted (and I did too), the only alternative is now a VM. Fine for some spreadsheet or database, but no VM can emulate the hardware of an esoteric ISA bus card! There are USB/ISA adapters, but I don’t know how true to ISA they really are (and they are too expensive to buy and try – cheaper to buy some old ‘retro’ PC off ebay).

            Interesting point by tekkieneet about the TPM connector, I’ll have to look into that. However, even if that is a way forward with ISA (or PCI), it still doesn’t address the fact that so many modern “compatible” PCs are simply not.

          4. The architecture has been very sensible. Even though it was hell to program, it turned out to be very scalable. Which is why Intel won the CPU race, and x86 is still around. You can’t deny the facts.

            I really, really enjoyed asm on Motorola’s 680×0 processors. And who knows where we would have been now if Motorola had been able to make the architecture scalable. But they couldn’t, and 680×0 is d-e-d: ded.

            Although to be honest, some guys (Apollo) have made a 68080 core for FPGA. Motorola never got further than the 68060. The 68080 is 680×0 compatible, but can run much faster than the 68060. It’s about 4x as fast as a 68060 at 50MHz. http://www.apollo-core.com/index.htm?page=performance

      2. There are plenty of not-x86 solutions to chose from if you so wish. X86 is in the market position it’s in because it appears that people want backward compatibility. The market has spoken, and it doesn’t appear to agree with you.

        1. I doubt backwards compatibility alone is the reason for market dominance–not in a purely technical sense. Early adoption by Microsoft and IBM are bigger factors, along with the development of mostly compatible clones. The unseen hand of market forces is a fallacy–status quo is very deliberate and planned: the agreements between dominant firms cemented the platform’s position in a top-down manner. For a platform switch to be justified on a mass scale, corporate leaders would require a significant financial incentive. The hardware and knowledge investment in the platform already works against such incentive. Compatibility with current dominant software is more important than backwards compatibility generally–the x86 platform will remain dominant on workstations as long as it is the sole target for professional applications.

          1. “Compatibility with current dominant software is more important than backwards compatibility generally”

            And this is where it is, and what it has always been…. compatability with dominant software.

            Users are told “it’s backwards compatible” but the qualifier ‘with current dominant software’ is always left out, and over time, ‘backwards compatibility’ is no longer what the words imply.

      3. X86 chips are at the good enough point on basically everything except battery life(Which can always be meaningfully improved).

        The compilers already exist, the chips exist, it all works really well… If we’re gonna replace it, the next thing had better be really, really good.

        I’d love to see object oriented CPUs with native object operations, or CPUs with most of the hardware acceleration implemented as FPGAs, but that seems unlikely.

        RISCV and ARM have a long way to go. There’s boatloads of legacy software. Ditching X86 would cause a lot of trouble.

        The tech industry seems to really like to break backwards compatibility these days. Everything is constantly being rewritten and reinvented, rarely is anything designed to be solved once and only minorly maintained after.

  3. One of these won’t be making it into my main systems. The Chinese have already been caught doctoring SupreMicro MoBos to steal information. You couldn’t take the spyware out of the CPU. No telling what they will put in it.

      1. LOL! At this point I don’t trust anything done by big business/government in the computer/software market. But I do have a slightly higher level of trust for AMD and a little less for Intel, than others. They have used foundries outside of China. And when I stop to think about it I start to wonder what all those little ARMs are up too… A network monitor is next on my list of things to write….

        1. What is going be your hardware? Intel ethernet hardware, perhaps? Or another US manufacturer. Considering where the tentacles have been found before, it really wouldn’t surprise many people to find interventions there too. To be sure, you’d need some way of assessing the actual data on the lines, not the interpreted data ethernet chips present you.

      2. Exactly. Was gonna say—China is only following our lead. We invented all this spookyware. I’m sure their version will be horrific in some brand-new ways, but they’re adopting our model, ramping up the super-surveillance and black mirror-style dystopia, and guess what? It’ll be coming right back to us once they have a system that works. It’s not to paranoid to consider their tech surveillance state both a direct export and also a testing ground for our own.

      3. There are some options that are more secure and are tested over the years, but most of the recent products are spyware aswell, still not the argument to shill for China stuff. One wrong cant fix another wrong.

        This is something I’d like to see more – open source hardware that is near to the competition, at this point the price is not even a problem.

    1. Might make sense to partner one of these with an intel and an AMD chip, then set them up so the spyware on each will fight the spyware on the others, maybe the outputted result between all of them could perhaps be considered trustworthy.

      As another point off this topic: so is the idea that this thing could be a directly swappabale replacement for commn intel and AMD chips, able to run the same linux or windows operating systems and the same precompiled x86 and x86-64 binary files?

    2. I don’t think we’ve seen any real evidence of actually doctored SuperMicro boards – no teardown of an ‘infected’ board, no DEFCON talks, just smoke and no light from some non-tech journalists – at this point it’s still just conspiracy level stuff

      1. This is my thoughts as well.
        There have been IPMI vulnerabilities since the dawn of IPMI, and that is why it is essentially universally agreed upon to firewall and secure it away from the internet.

        The nail in the coffin for the Supermicro articles is that the reported issues being all supply chain tampering and hardware implants would be more expensive and less likely to be successful on an ongoing basis. They are also all documented as hardware changes, which mean physical evidence, that nobody has been able to produce.

        I find it extremely unlikely that massive companies all found issues with hardware because of physical evidence, and not a single report or photo has surfaced of an actual compromise.
        If this was done as supply chain tampering it would be everywhere.

        1. The story was based on (an overwhelming number of) anonymous sources, and it’s not clear that the sources actually said what was reported.

          We know two people interviewed by the Bloomberg team, and both of them said “yeah, I guess it’s possible, but there’d be better ways” which got reported as “oh yeah, it definitely happened.” The reporters seemed to be after that particular story rather than after the truth.

          Lacking on-the-record sources and physical evidence, what’s left? Poor journalism.

          Which isn’t saying it couldn’t have happened. Just that there’s no proof.

          1. The telling thing was that organisations which never confirm or deny anything came out and denied it, which is almost unprecedented. No other news organisation came out with their own angles, which is also quite unusual.
            Apparently Bloomberg reporters get a bonus if their stories cause movements in the marketplace, which is a recipe for bad journalism.
            I am certain the story was bogus, and the reporters put 2 and 2 together to make 11. And probably got several $ out of it.

    3. You’ll have a hard time finding recent CPU that don’t have ME or PSP. I have the old Q6600, AMD FM1 APU and some PII, PIII tha didn’t have either. My FM1 APU is still more stable under win10, so I do most things on it other than gamin or any CPU intensive video encoding.

    4. Also fyi, the supermicro story seemingly turned out to be a case of sensationalized journalism which was never able to be verified by any 3rd party professionals or researchers. That of course doesn’t mean it is absolutely false, but it also doesn’t mean it is absolutely true either.

    5. It is being made with the 16nm TSMC process and the KX-7000 series will probably use the 7nm TSMC process.

      So it is as trustworthy as Intel using the 7nm TSMC process for their next-gen DG2 (because they do not have that ability in-house!)
      And AMD using TSMC fabs
      And Apple using TSMC fabs
      And …

      It is far far easy, and cheaper, to add software spyware than hardware. And the software spyware can be designed to fully vanish if it unable to phone home for a period of time, you can not really make hardware based spyware vanish. So the IME can be good (or spyware) and the PSP can be good (or spyware) with a single FISA court order. I do not know, but I would guess that China probably just clone what Intel and AMD are currently doing and add hooks in the hardware that can only be fully accessed by a separate management chip.

      https://libreboot.org/faq.html#intel
      https://libreboot.org/faq.html#amd

      And there will probably be a new section in the FAQ at libreboot on the Zhaoxin chips when/if they use a private management core.

    6. What’s the difference between a diagnostic feature and spyware?

      The person using it.

      This ultimately what the problem is with any complex computing system. So unless you’re willing to build a computer from discrete logic chips, like Ben Eater has done, you have to have trust at some level that the computing system isn’t going to be used against you.

    7. Thanks all for the comments. I threw this out here as much for my education as well as to be an alert. Thanks for the corrections on the SuperMicro MoBos. I ran it down and did find it had been debunked. Sorry. I hate the taste of shoe leather. :-(

      But SuperMicro demonstrates why I trust AMD and Intel a *little* more. Oversight. I happen to know one of the many individuals who sits on the benches testing CPUs after they come off the production line at Intel. In short it means they *are* testing their product for anomalies. I trust/like Intel less for many reasons dating back to the beginning of the ’80s. But the most recent of which are the continuing issues being documented here and elsewhere. The “management” tools they put on the MoBos, that you *can’t* turn off also make me crazy and I work hard to make sure they are not actually attached to the network. If I had a say in the matter I wouldn’t be involved with Intel products.

      And honestly there is a difference between lack of fore-sight and actual intent. The Chinese government’s exploits are well documented. You can start by reading the last couple of DHS cyber alerts. And I’m sure that what’s known is just the tip of the iceberg. The majority of incidents reported by my firewalls and defense measures come from China, followed by AWS, Russia and surrounding territories. Thanks Amazon for making such a handy attack platform. :-/ One can black list the Chinese networks and see a great strain released from their defenses. I do that on some servers. A Chinese made CPU would have less than 0 oversight. Being a communist country gives them the means and they already have the motive.

      Yes at some point one has to accept blindly the hardware they use and hope for the best. I’m not sure I’d call that trust. That’s still not a reason to open a door for the cyber-thugs. I’ve come to the conclusion over the last few years the only hope is Open Source. Every major computing related company out there has demonstrated some degree of ill intent, or at least a desire to do computing in a way I don’t want to. ;-) If you want to see something fascinating. Deprive your Android smartphone/table of iNet access for a week or two. Then connect and try to use it. It will go comatose blasting data, like someone who hasn’t eaten in a long time at an all-you-can-eat buffet.

      I don’t use iPads/iPhones so I don’t know about them. I can also say that the Android experience happens with factory only apps. Length of deprivation may need to be increased. This is one of the reasons I’m going to work on that traffic logger.

      I thought about this as I shopped for a replacement SBC for the C.H.I.P. How can I trust these other SBCs? Well… if the schematics are provided I can look at them. I can then compare them to the received product. Something like a “spy chip” should stick out fairly well. With Open Source software I can spend as much effort as I want to vet the software.

      My next workstation buy is probably going to be a home brew “Pi” cluster of varying kinds with microcontrollers and FPGAs thrown in for good measure. I’m certainly going to be looking at that RISC platform…

      And yes… I’m probably ready to be fitted for a tin-foil hat. ;-)

  4. china wants independence from NSA backdoors typically introduced into USA chip manufacturing.

    furthermore they want to be completely independent of US companies.

    This isn’t for making money this is so they can not care about us existing… They’re striving to be the nation that doesn’t need anyone and the one that every nation needs…

    1. I’d buy all that if they weren’t walking away with everyone’s IP instead of doing it themselves. It’s almost like they don’t have any confidence in their ability to succeed fairly.

      1. You mean how we steal IP from every one else on the planet? Disney steals local folklore and turns it into their copyright. Drug companies routinely steal ideas and compounds from indigenous cultures with no ccmpensation. What you say is almost perfect projection. It is almost as if we had no confidence in our own culture (hint, we don’t).

          1. We depend on stolen chinese IP technology for our food, medicine, entertainment, and weapons, so we need them a lot more than they need us. They need our pop music and our x86 instruction set, both of which can be readily cloned so they really don’t need us at all.

        1. Jeez Louise! You can’t “steal” folklore. What gets copyrighted is a particular *expression* of it.
          “Drug companies routinely steal ideas and compounds from indigenous cultures with no ccmpensation.” Also known as “spending money to do research and development”.
          I guess we should compensate the Indians for cigarettes?

        2. Typical virtue signalling brainless guy. Not even the engineering field is free from them.

          If YOU have no confidence in our culture, why don’t you leave hackaday and return to your cave and your stones?

    2. America is no longer playing Obama nice on IP theft. The Raytheon engineer in the news today is a perfect example. Enjoy prison pal. I would say the whole world is now on notice with China theft. The old days are gone. Why does this site promote IP theft if it comes from China? Somehow it is ok if it can be found in a market in Shenzhen for cheap.

  5. Can it run Linux with 100% open code, absolutely no binary blobs, and boot using Open Firmware? If yes then it looks very promising, particularly if the dual gigabit nics are genuine and have full throughput. Assuming it has the usual ridiculously good pricing that only China seems to be able to offer.

    1. If this thing has VIA ancestry I’d be fairly pessimistic about the iGPU situation. The VIA ‘Chrome’ GPUs typically combined all the punch of an Intel integrated with the FOSS-friendly attitude of Nvidia.

      If you just want dual GigE and enough firepower to do some real packet munging in software it probably doesn’t matter; but unless (in a surprise move) VIA embraced openness in this area during its time in the wilderness it’s not going to be pretty.

      1. That’s a bit unfair, because until the intel HD Graphics, year by year, via chrome was at least double the speed of Intel onboard (northbridge integrated) graphics, they were that bad. Intel GMA series also used to cheat the benchmarks, dropping frames like crazy so they looked comparable to everyone elses lowest end onboard, but they weren’t.

  6. I don’t think China has a monopoly on putting backdoors into hardware. I’m surprised that someone doesn’t confront Intel and AMD about things like the god instruction. Or that sneaky hidden cpu running minix that seems to hard to turn off or nic cards that turn systems that are off back on if you send the right command. Or just that turning off a computer or a phone really never does. Someone please invent “THE OFF SWITCH” again and please don’t use that stupid O with an I symbol in the middle.

  7. The 8 core ZHAOXIN KaiXian KX-U6780A@2.7GHz has a passmark score of 5760 (single thread 877) so it has between a quarter to a third the number crunching ability of the current top 10 most common CPU’s in use today.

    It would be a close match in terms of number crunching performance to an 11 year old 4 core, two threads per core, i7-960 @ 3.20GHz but at less power (70W vs 120W – explained by the 16nm process vs a 45 nm process) which had a passmark score of 5791 (single thread 1390).

    So unless the price of this new chip is undercutting a the current price/performance by a large margin it is probably not worth it (as a desktop), but maybe China will have more new fabs coming online with a process smaller then 28nm that can decimate the price. Zhaoxin (means a million cores in English) is a fabless manufacturer, so it would be relatively painless for them to swap the production of the current 16nm engineering models made at TMSC to mainland China. If there was a 16nm process inside China, but currently all public information says that there is nothing smaller than 28nm (yet), but public information can be wrong.

    I watched the video and at 4 minutes and 4 seconds the cinebench score for the chips were displayed, and to be honest comparing one processor to a model that is two generations earlier that is clocked slower, has 50% less cache, half the cores and made on a 28nm process by the same manufacturer is not really a very useful comparison. Lets compare our current product to a much older product, wow it is better – I was not expecting that at all.

    cinebench (Release 20) score 228 for the 4C/4T 1.83 GHz ZHAOXIN KaiXain ZX-C+C4580 (28nm process)
    cinebench (Release 20) score 845 for the 8C/T8 2.70 GHz Zhaoxin KaiXain KX-U6780A 70W (16nm process)

      1. Thank you, very interesting.

        I found that Taobao are pre-ordering the ITX-motherboard with BGA processor for ¥4300.00 (~€560/~$620/~£470). So not exactly standing out in terms of price/performance.

        Regardless Zhaoxin are defiantly a company to watch, they are planning to go to a 7nm process in 2021. It will be interesting to see information on the production run of the “ZHAOXIN KaiXian KX-U6780A @ 2.7GHz” CPU’s and the final cost. And the company that make their BIOS (Byosoft) should definitively be watched very closely

    1. No, it doesn’t make sense, other than making the chip smaller and cheaper. Modern CPUs tend to be bottlenecked by the RAM (DDR4 limits you to about 8G/s peak, and there’s a large overhead for a cache miss that requires RAM access (this is why SMT was invented, to allow the CPU to get on with other work when the active thread stalls due to a cache miss. This gives about 30% more throughput). L1 caches are split between data and instructions in order for the CPU to be able to fetch and execute at the same time). L2 caches tend to be comparatively small because the larger you make a cache the slower it gets, so the L3 is large and slow compared to the L2, but still fast compared to SDRAM access.
      Modern CPUs like the Ryzen series run 8M of level 2, then 32M or 64M of level 3. These chips have neither SMT nor L3 to help reduce cache misses, and the results show in benchmarks. Adding in L3 and SMT, but cutting the core count would have probably produced a chip of the same kind of size, but faster.

  8. These chips are DANGEROUS.

    Anyone with 2 brain cells knows what happened to the FTDI usb chips AND Scada infrastructure frameworks.

    And they will cut corners, procurement and bean counters alike will chomp on this hook FAST.

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.