Huawei Unveils High-Performance ARM-Based CPU | Hacker News Reader

Huawei Unveils High-Performance ARM-Based CPU

huawei.com

233 pointskarimtr132 comments

Huawei is also a Gold member of the RISC-V Foundation:

https://riscv.org/membership/1745/huawei/

As another potential threat to ARM, here is an ARM licensee that designs their own chips and is interested in RISC-V. It's not a problem today, but if interest in the ISA continues to grow it should be relatively easy for them to switch.

Switch???

There seems to be some widespread confusion what RISC-V is, currently. It is not remotely competitive with any ARM processors, and for those used in smartphones, there is no indication it will be for at least another 10 years. It is solely used in applications where the choice of processor is mostly meaningless and cutting out ARM is 20 cents saved on the BoM. Remember Huawei makes a long long list of products besides smartphones such as networking gear where you need the occasional small to medium sized processor.

> It is not remotely competitive with any ARM processors

It is absolutely competitive with cores at the same gate count. That goes beyond stuff like RPi level cores.

> and for those used in smartphones, there is no indication it will be for at least another 10 years.

I mean, shitting out a new high perf uarch isn't the easiest thing in the world, but you're literally commenting on an article from Huawei claiming that they had. And now Moore's law is more or less mined out, which means it's no longer a Sisyphean task to chase the perf dragon.

> It is absolutely competitive with cores at the same gate count. That goes beyond stuff like RPi level cores.

He may also have been thinking about things like toolchains and debugging tools. There are several professional IDEs geared towards ARM embedded development

There are tool makers and IDE makers that support RISC-V already. Look at the list of companies, it includes quite a few companies like that.

Of coure ARM is way ahead but RISC-V is catching up and people are trying to innovate.

> There are several professional IDEs geared towards ARM embedded development

Having been an embedded developer for around a decade, yeah they exist but are absolutely terrible. I seriously don't understand how Keil, Metrowerks, and IAR still make sales. Like, they're somehow worse than the FPGA toolchains I had to deal with.

If RISC-V doesn't get those parasitic business models attached to it, it'll be better off.

What IDEs/toolchains do you tend to actually use/see used in industry?

I've used all of the above and shipped code with them; when coming into a project it's more important to get things done than bitch about the gains other tools get you.

When starting a design from scratch, my own proclivity generally goes towards gcc/clang +some commandline build (make, cmake,.etc), openocd+gdb, and an eclipse project built up that's ready to go for those who like GUI editors. That free stack tends to have less bugs than the above tooling and way more examples for when you want to do something weird.

One very interesting development is MIPS open-sourcing all(?) of its IP.

From what I understand, MIPS cores are already quite competitive with ARM, with good support for FP etc., and the MIPS ecosystem is quite good.

IDK, the MIPS announcement was worrying to me. IIRC it's only open for 'MIPS Open Registered Companies', ie. it's open in that they're open to the idea of providing source to the normal customers they have, sort of like Microsoft has ways of giving companies, governments, and academia the Windows source.

The dude who bought MIPS is a sales guy who I don't really think understands open source.

It'd be awesome to be wrong about this though.

Maybe you should actually inform yourself before posting a definite statment like this.

That no RISC-V processor is competitve with any ARM processor is just straight up nonsense. RISC-V ISA wins in most metrics against ARM amd chips that implement it get higher performance on the same gate count.

Your next claim that 'its all about that 20c licencing cost' is further nonsense that simple is not reflected by most companies in the RISC-V space. It probebly true for some, but WD for example has said that just based on cost they would not switch.

Next lets talk about high end cores. Esperanto technologies is gone licence a 7nm RISC-V Core that competes directly with A-72 and A-76. That not as advanced as some of the cores you can licence but its very competitive in terms of soft-ip.

That all said, mobile application processors will be a few years away but the primary issue is not the core performance itself.

I've heard hardware guys complain about saving pennies. When you're manufacturing millions of units then saving 20 cents on each of them absolutely does matter. Obviously saving 20 cents now doesn't offset all the other issues with RISC-V but saying it's meaningless to cut out 20 cents is not correct.

I don't think that's what stefan_ wanted to say. I think it's rather that RISC-V isn't effective enough (yet) to be suitable for scenarios where the main thing you optimize for is something other than cost (like heat dissipation, power use, performance, etc). For scenarios where these things don't matter, but cost is very important, RISC-V is the ideal choice.

The problem is that most of these companies will still use commrecial soft cores. Meaning that you still pay for the RISC-V chip most of the time. So it will probably be cheaper, but your not saving 100% most of the time.

There are some companies where is is a huge issue. However for others its all about avoiding up front cost. For others its about flexibility. For others its about narrow innovation.

I'll bet that in <= 6 years we will see popular mobile computing devices with RISC-V applications processors.

But you're right that those will probably be the last to switch. Most SoCs have plenty of ARM cores that could be swapped with RISC-V cores without significant (or even any) impact to vendors.

What do arm licencees actually get?

Technologically speaking, slapping a risc-v decoder on an arm core is trivial no?

So how much of this is Huawei ip, how much arm? Is it not possible that arm will become the leading risc v vendor?

Did you see the list of partners in the announcement? Its one thing to say that linux runs on your core, its another to have a supported release by every major distro along with other OS's people want to run working (windows/etc). It might be said that is the harder piece of work because its not just engineering but politics and business cases.

So beyond that there is an incredible amount of engineering work required to get lots of application stacks to perform well on an architecture. Its one thing if a core is similar performance/watt given some small benchmark suite. Its another to get specJBB or other random large application suite running well. There are literally tens of thousands of pieces of code here or there that need to be ferreted out and fixed, from adding crytp/vector instruction support, to lock tuning. There is also the experience of finding edge cases in the system architecture that don't perform well in a given situation and need to be modified in a future design.

That is something you get with a mature architecture and something the arm ecosystem has spent the last 7+ years working on with organizations like linaro and in major projects like the linux kernel, gcc, clang, openjdk, android, etc. Missing a 20% performance optimization somewhere in a 100 million line code stack can be the difference between performing like a 7 year old core from your competitors or being seen as equal.

Now all that said, risc-v has plenty of opportunity, investment, etc. Nothing I've said above is unique to arm/x86/power/etc its just a matter of putting the effort in. That work hasn't yet been done for risc-v, so its a natural fit for small limited microcontroller/embedded applications where the software stack is more controllable, rather than general purpose high performance applications (phones, servers, etc).

A good sign that the RISC-V hype is out of control here is that the top comment on a news story about something good for ARM claims that it is actually bad for ARM and good for RISC-V.

No, I see this comment as very relevant.

If you think about what happened recently, and Huawei's position as the beacon of China's hihg-tech industry; the fact that they are in Arm and Risc-V suggests the scope of their investment. That is important to indicate the direction they want to go.

If you also correlate that with their investment in K8s, you'll see a clear picture how a complete strategy is being laid out. And Huawei understands Open Source is going to be the future for catching up.

Now piecing everything together:

- If done right, an investment in Arm for short-/middle-term, and long term bet on Risc-V + cluster manager like K8s, will provide a meaningful long-term alternative for cluster computing.

- Then again, Huawei has similar investment in the edge computing space for various kind of low-power phone chips.

So basically their investment in ARM has cross-over advantage to RISC-V (and likewise?)?

What do you mean by crossover?

Presumably that CPU design expertise is transferable between architectures, especially if they are similar.

I only know very little on the surface of chip design in general. My impression is yes, expertise are transferrable. Although the key here is cultivating market and hone the operational prowse, those also require time to do.

When the starting point is RISC-V coming from nothing, a company that has strong chip-design capabilities and interest in both is a step closer to parity.

News can be good for multiple parties.

It's good news for Arm and a potential threat. Editorializing to "actually bad for Arm" is not an accurate reading of the comment.

...a news story about something good for ARM claims that it is actually bad for ARM and good for RISC-V.

Not really. While I was reading the announcement from Huawei it seemed like it was almost as much an announcement from ARM. It also spoke to Huawei being a large supporter of open source and collaborative development. It mentioned several organizations they work with, or are a part of. In light of their open nature, their interest and involvement in RISC-V was notably absent, once again hinting at the announcements importance to ARM.

They make it sound like this is one of the highest performing ARM processors out there, and maybe it is. But at that point you have to ask what exactly does ARM bring to the table for Huawei? I contend that at this point it is mostly familiarity and mind-share.

Today's CPUs from AMD and Intel are going after the volume cloud market with a range of CPUs with a particular configuration of functional units, cores, PCI-e lanes, memory channels, etc. They both push an eco system of motherboards, compilers, operating systems, libraries, and application tuning.

ARM brings (the platform, not just the reference chips) a healthy eco system and allows Huwai to make their own decisions on what the market needs and doesn't automatically give most of the profit to AMD and Intel. Keep in mind that for the first time in several decades Intel and AMD chips doubled in price, power usage, and performance in the last generation.

When buying whitebox servers most of the profit is going to AMD or Intel, not supermicro or similar.

Part of the problem with competing with the Amazon, Google, and Microsoft clouds is that they more assuredly are getting better CPU prices than you will. If you can make your own CPU

So given Linux's maturity and increasing activity with arm on the server side the barrier to entry to competing with Intel and AMD is lowering. This combined with doubling in CPU prices makes AMD and Intel much more of a target than they were 2 years ago.

Huawei is in the game, for sure. Also, comments that dismiss Chinese achievements appear uninformed of the level of effort at technology supremacy under way in China.

BUT there are a few "buts:"

As the article says "the new CPU is designed to boost the development of computing in big data, distributed storage, and ARM-native application scenarios."

This isn't anything like an Apple A12, which has world-beating compute power per watt.

The Huawei chip should be compared to Intel server CPUs and ARM and other chips with other ISAs designed to use in data centers.

This is an important step technologically for Huawei, which will probably use enough of these chips to learn for a next generation. This is probably not a chip that will disrupt anyone else's chip business.

One of the best comments on this. Its natural for Huawei to do this to make them, and China market in general, less dependent on foreign chips. Its also natural considering the areas in which Huawei does business - they effectively are a end-to-end supplier of telco platforms from server, beacons, mobiles.

I think is very good for the world to have additional heavyweight competitors in the CPU space - especially given that its ARM based so no reasons(other than political) that other brands would not want to consider as a supplier.

From the only number I see, the specint score, it is 40% faster than Qualcomm Centriq socket vs socket, and even 6% core vs core. Which is a really great if true. Wonder what the power draw on that thing though.

To clarify, this is not for mobile devices, for those who want to compare it to the A12x Bionic.

Furthermore it's a 64 core design, so aimed at massively parallelized workloads and high bandwidth applications, which is why they bang on about it's impressive set of interfaces. Not surprising for a company that is mainly a telecoms operation and only incidentally makes phones.

I love to the TDP # when 64 cores are running 2.6G.

Also like to know the L1, L2, L3 #.

It is hard for any system company to sell CPU. Apple is only for their own use. I believe the Samsung's AP is for their own use also.

> only incidentally makes phones

They're the number 2 phone maker in the world...

...except that A12x Bionic is itself comparable to desktop CPUs

Hardly. It is closer to Intel U series processors which are much slower than their desktop counterparts both in peak performance (desktops clock higher) and in sustained performance (desktops dissipate heat better). It is additionally slower in mult-core workloads where desktops have access to 30+ cores (none of them "little") and over 20x as much RAM (without even considering multi-socket systems with up to 3 orders of magnitude more RAM).

Most people don't need the extra power a desktop offers, but the performance advantage over mobile systems is enormous.

Pretty sure they were comparing it to an i3/i5, not 30+ core servers. I don’t think anyone is making that comparison.

Considering both intel an amd are selling "desktop" single socket machines with >16 cores, I'm not sure where you got the idea he was talking about servers. TR is now at 32 cores, 64 threads so definitely >30 cores and classified as a desktop.

It might be more useful to argue "workstation" but I think his point remains, people calling their 5W 8 core part "desktop" class are referring to the "laptop without a battery" class of devices that exist in the low end space and don't take advantage of the capabilities of the desktop form factor.

It absolutely is comparable to desktop processors. The latest A12 bionic is more powerful than the average desktop CPU in use across the world.

That's not saying it's better than the BEST desktop CPUs, but then, 99.99% of people don't have a 16 core Threadripper or whatever.

The point is, Apple's mobile chips have reached a point where they are absolutely desktop class processors and most normal users wouldn't notice a thing if you replaced their laptop or Desktop CPU with them. This is because most normal users use their home computer for web browsing, photos, music, and Netflix.

For example, the A12 bionic is absolutely more powerful than my 2015 Macbook Pro and my PC which I built in 2014. You might not like this reality, but unless you're a "power user", this processor is desktop class. End of story.

Ok, so we are moving the goalposts from what you can currently buy as a desktop machine, to what you bought as a desktop machine 5 years ago, which likely wasn't even top of the line.

Then because you can run a web browser on it that makes it desktop class?

Please that argument doesn't hold an ounce of anything. For that matter the processor in the original iphone was desktop class because it was capable of surfing the web better than the Pentium II's common in desktop machines in 2000.

Which is why tons of people just bought ipads and stopped upgrading their desktops. I never made an argument that you couldn't replace a desktop with a phone/tablet/whatever, that wasn't what this was about. It was about claiming that apple's products were performance competitive with (current) desktop class machines. Which for some use cases they probably are, for others like heavy gaming, engineering work, or basically most of what people actually have desktop machines for these days (as opposed to laptops which is probably a better comparison) its not really a comparison outside of some great marketing.

Of course this is the usual apple crap, I've been hearing it since the PPC in the mac was a "supercomputer".

When you start to scale performance, the core starts to matter a lot less. A standard 95w desktop processor has the TDP to push clocks much higher. It has the TDP to add loads of extra cache and reduce those misses. It has extra memory channels and the memory can operate at higher speeds. It has much more RAM reducing HDD read penalties.

When you start scaling out cores, interconnect starts to eat a huge amount of the power budget. All the extra PCIe and other IO eat even more.

When you start to ramp up into the 3+GHz clockspeed range, things start to get really hard and you have to make IPC tradeoffs.

The idea that a 5-10w A12 beats the performance of a 95w x86 processor is talk from a place of extreme ignorance.

Add all the features even a workstation laptop chip has and the A12 will start to struggle just like every other architecture does. There's no magic in chip design. There aren't really big competitive edges to be found anymore as all the low-hanging fruit has already been picked a decade ago.

When the word Desktop is used, most would assume it means vast majority of Desktop, not Prosumer Desktop, Gamer Desktop ( Despite a huge amount of sales are coming into it now ) or Workstation Desktop. And I can assure you most don't have have Quad Core, or even Quad Thread, let alone 32 Core. Most don't have more than 8GB Memory.

And my guess is the OP was using the word Desktop as being PC, Laptop + Desktop. I am just guessing but He should have been more clearer in that.

That's just moving the goalposts and is a non-argument.

A 13" ipad pro starts at $999 and goes up over $2,000. Any desktop in a similar price range will stomp the ipad processor without much issue.

Moving Goal Post?

No one buys $2000 Desktop anymore. Majority don't. And you had to pick the most expensive 0.1% Market Desktop compared to A12X. The A12X here is a specific comparison, while the words "Desktop" is an generalisation. When I say my X Bike is comparable to Road Cars, Road Cars here means everyday cars, not your Lamborghini or Ferrari 455. Which we ALL KNEW would be faster than this Bike anyway.

But you had to nitpick on the word "Desktop". And then decide to use Similar price range as an argument which was never the intention nor what we started to compare it with.

iPhone XS is faster than iMac Pro on Speedometer 2.0 https://twitter.com/dhh/status/1043277162676072449

What makes you say it's closer to Intel U?

I would hesitate to draw conclusions about the general ability of the processor from a browser interactivity benchmark running on two differing OS's, with two likely differing browser builds, etc. Apple obviously spends a lot of time optimizing graphical interactivity/safari performance on the iphone and its shown for years. It might be interesting to try the same version of firefox on both platforms... or just a broader set of benchmarks.

Its really too bad that no one has managed to get a stock linux running on an iphone....

The point I am making in this thread is that they are comparable, I never recommended drawing general conclusions.

The slowest iMac Pro has a max clockspeed of 4.2GHz. The fastest has a max clockspeed of 4.5GHz. The max clockspeed of the A12 is 2.49Ghz. What magic allows Apple to get an almost 200% IPC boost? If that kind of performance improvement is actually possible in the general case, then Apple should exit the consumer market and start selling their chips to everyone. They'd make way more money than they do now.

If something in chip design seems too good to be true, it is.

This is what a lot of people including myself are actually expecting, yes. The most interesting part is that the test is quite real world, not some microbenchmark. It runs to-do list app implemented with modern popular frameworks like Angular.

JS doesn't make a good general benchmark and speedometer is a bad benchmark in particular for that kind of test.

The test itself is basically just writing a tiny todoMVC in a dozen different frameworks. Any one of those sub-tests would fit in only a couple MB of cache (for example, the React todo source is around 550 lines in total).

An i7-8650U has a max clockspeed of 4.2GHz, so for short bursts of single-core processing where everything fits in cache, performance can definitely come close to that Xeon (there's a reason why people buy the Xeon instead of the U-series chip though)

JS uses only 32-bit integers and 64-bit floats and executes on a single core (JS has typed arrays, but they certainly aren't going to be used in those kinds of frameworks).

The power of modern x86 architectures is completely bypassed in that kind of test. A huge amount of the size and TDP an x86 core (I'd guess the majority given the width) is tied up in vector units, but those go entirely unused in JS code (and the JS SIMD proposal was completely killed making that unlikely). What are the other 7 to 17 Xeon cores doing during that benchmark? 30-50% of the TDP in these chips goes entirely to the interconnects because the hardest part of scaling chip design stopped being the actual core a long time ago.

https://ark.intel.com/products/124968/Intel-Core-i7-8650U-Pr...

https://browserbench.org/Speedometer2.0/

I think it would be better to say it is comparable to laptop cpus, not desktop

How so, if iPhone XS is faster than iMac Pro on Speedometer 2.0? https://twitter.com/dhh/status/1043277162676072449

Running a Javascript benchmark and not even specifying which browser was used is not a great way to compare CPUs. It's also entirely single threaded.

Geekbench for example puts the A12 about 15% above the 8650U in single threaded performance and 9% below it in multithreaded performance. I would say that is pretty close to U-series performance.

https://browser.geekbench.com/processors/2064

https://browser.geekbench.com/ios_devices/55

You are shifting a point of conversation. The point is, they are comparable and should be compared.

Single-threaded is an important comparable point.

But if you're talking about desktop vs laptop performance, single thread performance is very similar and the only large difference is multithreading where desktop chips are something like 2-4x faster. If somebody is saying that it's more similar to laptop chips than desktop chips, the only reasonable interpretation of that is talking about multithreading performance.

Sure it's 'comparable' to the iMac Pro if the comparison you want to make is that it is 1/4 as fast.

This ARM server seems like a cloud compute type device to me.

How are Huawei, Apple and Samsung beating Qualcomm so easily?

Did Qualcomm just beat Intel in the mobile market and then just become complacent with patents?

One reason may be that they are better sponsored by the gov. Some kind of tech dumping.

That could only apply to Huawei if I understand what you mean.

This is one thing I can't tell if I'm being "tin-foil hatty" about. What is embedded in that processor?

That argument could easily be reversed: Why wouldn't the Chinese be worried about bugs embedded in foreign tech used for sensitive tasks like military applications? In both cases it's more a matter of trust than anything else.

Since it's more about trust, having more actors on the market is a good things, as it gives more options for the rest of us.

It’s also possible Huawei is getting hassled by the US because it is not susceptible to US government attempts to embed such things on behalf of the US, as some (not all) other companies might be. Like you I can’t tell how crazy this idea is. I’m open to your tinfoil hat theory too, but both can be true.

For Huawei, that's pretty much a given. Doesn't explain Apple and Samsung though. For Qualcomm, I would be surprised if their bread and butter ARM CPUs weren't mobile CPUs. Why stick with reference designs if they could tweak them or come up with higher performing ones? The Apple iPhone CPUs from two years ago are still trading blows with the latest Snapdragons. That's crazy.

IIRC, Apple just spends more; they have more transistors. This is a luxury that is harder for Qualcomm to afford, as their customers wouldn't put up with it.

> IIRC, Apple just spends more; they have more transistors.

Also, Apple has a full ARM license and full control over the OS and the software that runs on it. Which means it is very well possible that they have extended the instruction set to accelerate real-world JS performance.

That was a rumor floating around - that WebKit took advantage of new A12 instructions. It wasn’t true at the time. They may have released the patch since then.

https://twitter.com/saambarati/status/1049202132522479616

It's not just about WebKit. The remainder of the OS, including the graphic stack, is most likely to be optimized as hell for the A12 CPU.

There is a _LOT_ to be said for gaining a few %5 here and there improvements. Apple can both optimize their software for a very limited set of micro-architectures (a luxury that doesn't exist on android which runs on everything from tiny in order cores to larger more general purpose OoO cores) as well as optimize the cores themselves for an exact instruction sequences they determine to be critical for _their_ OS/applications/power envelope. This means their cache sizes/latency/TLBs/ROB buffers/frequency/power profiles/etc are decided not based on what generates a good specCPU score but what generates a good iOS/safari experience.

Its a complete vertical integration and no one should be surprised that a general purpose kernel (linux) running a AOT/JIT'ed environment on top of an os abstraction layer (android) on top of processors frequently designed to exist in applications that aren't phones gets beaten on a phone centric benchmark. Call me when someone gets a big hadoop benchmark or dpdk results on an apple core.

The new A12 processor comes very close in the same benchmarks to the best desktop processors, not mobile specific or phone centric benchmarks. I don't know why you persist in your non factual and biased comments.

"What is quite astonishing, is just how close Apple’s A11 and A12 are to current desktop CPUs. I haven’t had the opportunity to run things in a more comparable manner, but taking our server editor, Johan De Gelas’ recent figures from earlier this summer, we see that the A12 outperforms a moderately-clocked Skylake CPU in single-threaded performance. Of course there’s compiler considerations and various frequency concerns to take into account, but still we’re now talking about very small margins until Apple’s mobile SoCs outperform the fastest desktop CPUs in terms of ST performance. It will be interesting to get more accurate figures on this topic later on in the coming months."

https://www.anandtech.com/show/13392/the-iphone-xs-xs-max-re...

"we see that the A12 outperforms a moderately-clocked Skylake CPU in single-threaded performance"

moderately clocked skylake != "best desktop processors"

There is a good 2x perf difference between a 2 something Ghz skylake and a 4.5Ghz skylake, particularly on L2 cache bound single thread benchmarks (which is what specint2006 is these days, its more a cache benchmark except for libquantum which is "broken" and more accurately reflects clockrate*cores). A moderately clocked skylake is a laptop.

But you say, IPC matters!!! Sure it does, but a microarch with the same IPC and the ability to clock 2x faster is generally a 2x faster CPU.

Nor is apple shipping a phone with 16+ cores...

People have been saying their phones were as fast as their desktops for years, and its not anymore true today than it was 5 years ago. Its questionable whether it will really ever happen as there is this heat/power dissipation problem in phones that desktops don't have. Not to mention desktops don't have nearly the physical constraints on RAM/storage/IO capacity you find in a phone/laptop.

There are hardware-level optimizations in A12 for Objective-C, at least.

Further, Apple chip design probably doesn't need to respond to market pressure the same way Qualcomm does.

If Snapdragons get too expensive, phone manufacturers will switch to Huawei / Samsung / MediaTek, but Apple can absorb a $5 or $10 increase in per chip cost for better performance and power efficiency since it is vertically integrated.

And also Qualcomm seems to have been forced into giving away all of their server chip designs to the Chinese, too:

https://www.theregister.co.uk/2018/12/10/qualcomm_layoffs/

That's what you get from frothing at the mouth over "entering the huge Chinese market" by giving away all of your IP and tech through coerced "joint ventures" only to later see yourself kicked out of the market and some multi-billion dollar Chinese competitors selling your tech much more cheaply.

And there's this sweet irony, too:

https://www.bloomberg.com/news/articles/2019-01-04/u-s-star-...

If you think all of this is far-fetched, do read on Nortel and Huawei's "joint venture", which led to the demise of Nortel and the domination of Huawei in Canada:

https://www.lightreading.com/nortel-huawei-kill-jv/d/d-id/62...

https://www.assemblymag.com/blogs/14-assembly-blog/post/9063...

https://www.networkworld.com/article/2223272/cisco-subnet/60...

That's not to downplay the idiocy of Qualcomm's board and leadership, which also played a huge role in all of this, as well as Qualcomm being involved in multiple antitrust lawsuits. The board and leadership should have been fired at least two years ago. They allowed their greed to stay in the way of the company's technical progress and decided to make money from bullying other companies instead of continuing to improve their products (most of which are just rehashed stock Arm designs these days).

Nortel didn't disappear because of Huawei. Chinese are used as scapegoats for everything these days.

I used to work for Nortel and they went the way of the dodo because they were like Nokia (which ended the same way. I worked with them at some point as well...): Too bureaucratic, too slow, not hungry anymore, poor management.

For their last CEO they ended up paying $12 million to Motorola because that guy breached the non-competition agreement he had when he took the job at Nortel... Just an example.

I don't think you can dismiss the Chinese element in Nortel's downfall quite so easily. Undoubtedly Nortel made missteps themselves, but rarely does a large organization fail for just a single reason. In Nortel's case it definitely appears that China played a part. I think it can reasonably be argued from either side on whether or not China's activities served as a tipping point.

It's not 'China'. It's new competitors that blew them out of the water, like Apple and Samsung did to Nokia, for the reasons I listed in my previous comment.

Except a major one of those new competitors appears to have been a Chinese company that benefited from misappropriated trade secrets and IP. As I said, more than one reason for a failure of this scale. As an example, Nortel might have emerged from bankruptcy intact, as many companies manage to do, had it not been for the financial collapse of 08'-09'. That downturn made it more difficult for Nortel to get competitive prices for the assets it was shedding in an attempt to emerge from bankruptcy. Prior to that, a contributing factor in the decision to declare bankruptcy in the first place was that it could not easily service its debt and retain enough working capital to support operations. It had plenty of cash on the books, but nearly all of it was in other countries and repatriating it presented very thorny problems. [0] As I said, no one simple reason.

[0] http://www.ottawacitizen.com/business/Part+Last+things+first...

"misappropriated trade secrets": I read that as Nortel was poorly managed and handed those on a plate. That's business, if you're not smart others will be more than happy to abuse you. In this case Nortel essentially asked a competitor to manufacture their products from them... What could possibly go wrong?

Your last point regarding cash is also an example of poor management in my book.

It's too easy to always find scapegoats.

By misappropriated trade secrets I'm referring to the fairly well documented hacking of Nortel, not the willing hand-over of IP. [0][1][2]

I also think it's simplistic to say their cash issue was purely an issue of poor management, as it resulted in part due to the unforeseen financial collapse at the time.

Of course you can try to define both of those as "poor management" as well: Nortel should have protected itself better from hackers. Or they should have been prepared for a financial collapse on a scale not seen for 80 years.

Under this type of reasoning, any problem at all for any company anywhere could be defined as "poor management" and that all companies fail from poor management. The problem is that this overly simplifies rather complex issues. It's a bit like saying "Everyone dies of heart failure." Because in every instance of death the heart has stopped beating. In every instance of a company failure, its managers failed to act in a way that might save the company, therefore management was poor.

It neglects the difference between proximate causes and distal causes. It ignores that, for example, the heart may have stopped because a bullet hit an artery and a person lost too much blood. Back to companies, it ignores the difficult questions of "Why & How did management fail? What, if anything, could they have done differently given the knowledge they had at the time? Would those things have been enough?"

[0]https://www.washingtonpost.com/business/technology/report-ch...

[1]https://www.theregister.co.uk/2012/02/15/nortel_breach/

[2]https://www.cbc.ca/news/business/nortel-collapse-linked-to-c...

Mike Z, is that you?

Nope :) I have nothing to do with the comms industry directly. I just really enjoy (to the extent that's the appropriate word) the puzzle that is massive organizational failure. It's a fascinating topic. Sears & Toys R Us are two more recent examples that have enthralled me. The road from massive mult-billion dollar corporation to defunct is filled with branch points. Any one of them might have avoided or delayed disaster. You're correct of course in a general sense that mismanagement is a frequent component. In my opinion though, that may be a sufficient condition in many cases, but it's not always a necessary condition.

A great example, maybe my favorite, where poor management was really a primary issue was the death of Circuit City. There as they faced increasing competition from internet sales, Best Buy, and other competitors, they sought to cut costs in part by firing the highest paid floor sales staff, replacing them with lower paid, usually minimum wage workers. Some of those higher paid workers may have been dead weight, but on the whole they represented the most experienced, most knowledgeable floor staff best able to help customers (or more cynically sell them on stuff they didn't need with techno-jargon). In any case, it was a very short jump from there to revenue falling off a cliff, and bankruptcy. So poor management, absolutely foreseeable, directly caused their downfall. Though even there the poor management was partly a result of externalities like increased online shopping putting previously non-existent pressure on them.

I think the sears story's most important point is that they canceled their general catalog sales in 1993, the year before amazon was founded.

So 100% lack of foresight, but the real screw-up was by 1997 or so everybody and their dog was trying to sell everything under the sun online. Where was sears, still stubbornly refusing to acknowledge that internet sales were just the modern version of the business model that worked immensely well for sears.

How much would it have cost them to resurrect some form of mail-order/web forms read some business journals about Dell's order and deliver model and start making some side bets and deals with UPS/fedex?

Oh yes, these were absolutely the early decisions that started their demise. In more recent years though, I attribute their demise to the CEO. If not for this they may have weathered the storm another decade, maybe (though less likely) pivoted to a long term midel of profitability. But the CEO, far from simply beimg incompetent, engaged in a protracted pattern of self-dealing to his other financial interests that gradually drained them of resources and productive assets.

And to add to my other comment, Yes: It really is astonishing that Sears did not see the clear analogues between the internet and its traditional catalog sales. But then they'd (as you pointed out) decided to reduce costs a few years earlier by discontinuing their general catalog. Had they kept the catalog it's entirely possible someone would have said "Hey, this internet thing looks popular, how about we put a copy of our catalog up there?"

I'd love to know what their thinking was in the late 90's. I can imagine that management for brick & mortar stores might have pushed back on making inroads on the internet out of fear it would cannibalize their sales, and blocked it internally. Or it could simply have been inertia and a blind spot against noticing that "internet" wasn't a flash in the pan fad.

I was at Nortel (they acquired a startup I was at). I have never seen a more incompetently run company. Some of the stupid stuff they did was mind blowing. An example would be their acquisition of Alteon for near $8b when they could have had Foundry for less, except they would not make Bobby head of enterprise (where he would be useful) because of some old grudge over the Rapid City vs. Centilion beef internally (last gen of stuff they acquired - Bay Networks timeframe).

> I used to work for Nortel and they went the way of the dodo because they were like Nokia (which ended the same way. I worked with them at some point as well...)

Admit it - you're the real common link here... /s

Huawei have contracts from the govt just like lots of the American tech companies. And speaking of govt sponsorship, Samsung is definitely one of them, if you know the history about memory chips competition and how they become one of the monopolies.

Samsung, a monopoly? I don't know why anyone would say that unless you are a mainlander Chinese/Taiwanese.

And what gov't contracts? Unlike what China is doing today, there was no such overt protectionist gov't policies implemented here. Much of the demand for tech products and R&D were driven by the cold war between the WW2 and the fall of Berlin Wall. The stylized fact that there are now fewer than a few US memory makers, namely Micron and Intel, from two dozens in the 70's and 80's demonstrates how little the gov't contract has done to "protect" the industry.

China's policy of forcing foreign companies to give up their IP for access to domestic market and seeding unlimited gov't subsidy is based on the Japanese model in the 80's. Their model however fell apart as Japan's economy declined and the Japanese companies focused on quality, not on mass manufacturing to compete with emerging South Korean competitors like Samsung.

Unlike Japan's large domestic market, South Korea had very little market of their own and was geared toward purely for export. It didn't coerce foreign companies to drop all their patent lawsuits or future lawsuit like the Chinese gov't reportedly demanded Samsung and SK Hynix last week.

Needlessly to say, equating China's outright IP theft to the industry policy of the US or South Korea is a bit asinine.

* Noh, "Samsung, SK Hynix: Pay fines, or drop all existing or future patents lawsuits; China demands South Korean companies to give up their IP", Hankyung Economic Times, Dec 26, 2018 https://translate.google.com/translate?hl=en&tab=TT&sl=ko&tl...

* Kang, JoonKyu, A study of the DRAM industry, 2010, MIT PhD Thesis https://dspace.mit.edu/bitstream/handle/1721.1/59138/6595145...

Samsung as a whole ( Not Just Samsung Electronics ). Many Korean could have argued the other way around, because Samsung controls the government.

Credit where Credit's due, Samsung's Electronics R&D is absolutely one of the best on the planet. But it doesn't clear them as being clean.

They really aren't. You should read anandtech article about sd vs exynos. The exynos kills sd in geek bench but in real life performance sd does better. The issue is these things are in tiny enclosures. They will just start throttling after a while.

In this case it's a server chip so it's a little different.

Qualcomm's only real advantage as a company is killer patent portfolio, and lobbying power to sue every competitor out of the market. There were 10 mainstream Smartphone SoC available on the open market during 3g era, but only 2 in 4g (Mediatek and Qualcomm, with Samsung being out of competition due to their non-compete treaty with Qualcomm.)

It is only because Qualcomm can't go against Mediatek on its home soil, they went after their channel, and clients in courts.

I believe that Huawei's decision to home source its own SoC was also in part dictated by them being afraid of channel risk (i.e. Mediatek going down to lawsuit, or the civil war restarting)

BTW, if anybody remembers the company called Allwinner that was all the rage during tablet boom. I have 100% certain info that they too got into secret non-compete with Qualcomm in exchange for distribution rights of their low-end chips for 10 years.

And there's Broadcomm who simply vanished after 3G

First, wasn't Infineon always on the 3G/4G scene?

> It is only because Qualcomm can't go against Mediatek on its home soil, they went after their channel, and clients in courts.

Qualcomm, or wireless patent holders in general, never go after component makers; as a long-held industry practice, they always go after the handset makers for maximum royalty. That practice has been upheld by regulators around the world -- though there is a lawsuit by the FTC to change that at Apple's behest now. This also presumably allows new entrants to enter the wireless chip business and, therefore, increase competition.

Qualcomm further is not exactly known for lawsuits, though it always seems to be fighting regulators around the world. It reportedly initiated fewer than 3 lawsuits in the past two years -- now compared to Apple's 90+ pending lawsuits.

> First, wasn't Infineon always on the 3G/4G scene?

No, plainly speaking. There was never was a real SoC from them. Intel bought an already dead business from Infineon.

> Qualcomm, or wireless patent holders in general, never go after component makers; as a long-held industry practice, they always go after the handset makers for maximum royalty.

No, anybody with a real business, and not living off patent trolling will not do that, and did not do that. Only QCM really have no head on their shoulders there to risk everything in a retaliatory lawsuit every time.

> That practice has been upheld by regulators around the world

No, completely not the case. I worked in electronics business for close to 12 years now, and I know few things about how patent law actually works in real world.

Qualcomm was show to the door great many times by courts during their peak patent trolling a decade ago, in I think more than 10 countries.

Yes, I'm talking about wireless modems, not SOCs.

I wasn't asking your opinion on what wireless licensing ought to be; this licensing model has long been upheld by regulators and accepted by companies around the world for decades. Even China's antitrust regulator NDRC who fined Qualcomm to the tune of $975M last year supports their royalty model (https://www.winston.com/images/content/1/0/v2/100610/HarrisJ...):

"... Qualcomm will charge royalties for 3G and 4G Chinese SEPs for branded smartphones sold for use in China based on a royalty base of 65 percent of the net selling price of the smartphone and royalty rates of 5 percent for 3G devices and 3.5 percent for 4G phones; ..."

And all wireless patents holders like Ericsson, Nokia, etc use the same royalty basis (ie, system level licensing) and similar royalty rates a bit lower. If you weren't unaware of it, now you know it.

I've been around the software business for 20 years now though it doesn't make me an legal expert on anything, much less transnational IP laws. I just try not to make any empty position without evidence.

> I wasn't asking your opinion on what wireless licensing ought to be; this licensing model has long been upheld by regulators and accepted by companies around the world for decades. Even China's antitrust regulator NDRC who fined Qualcomm to the tune of $975M last year supports their royalty model (https://www.winston.com/images/content/1/0/v2/100610/HarrisJ...):

That's nothing much than a political move to appease USA. There is really no law in China that ever says any of terms named, they (NRDC) made it up on the go. From standpoint of Chinese legal system, there are simply no legal reasoning for that. Because to them it sounds like: "if you don't buy our chips, we will sue you," and that doesn't make any sense to the judges. In Chinese, the judgement effectively read like "we make a concession for you because big papa is behind you."

When Qualcomm was suing Nokia and Siemens a decade ago, German courts, I think, threw out more 10 of their lawsuits out of the court.

China's NDRC's antitrust findings were later affirmed by counterparts in South Korea, Taiwan, and the EU who found the same violations in their investigation in 2016 and 2017. Now, are you also arguing that there is no law in those countries/union and they just made it up on the go like China?

And appease USA? As I noted earlier, Obama's USFTC filed a lawsuit challenging Qualcomm's licensing practices. Who exactly was the NDRC appeasing? Qualcomm or USA? The NDRC's action is contrary to the USA's policy and China isn't going to win new friend by contradicting US's policy goals.

Again, the system level licensing is not only widely accepted and upheld industry practice. Take another instance of the UMTS IPR Association's statement on system level licensing in wireless industry:

" ... The royalty “collection point” shall be the last manufacturer in the manufacturing “chain.”This means that chip and subsystem manufacturers shall be indemnified for sales made toLicensees of certified Essential Patents who are the last manufacturers in the “chain.”Licensees shall not include those manufacturers of component products which are incorporated into final assembled products for which royalties are paid to their respective Licensor(s) ..."

* 3G Patent Platform for 3G Mobile Communication Systems – Definition, Function,Structure, Operation, Governance, UMTS IPR ASSOCIATION, Section 8.2.6 (June 15, 2000)

Huawei also has their own Cloud solution in China, even though it is not anywhere near as big a Tencent and Alibaba's cloud. So I guess they could use those as testbed. Along with 5G Infrastructure, directly competing against Intel. And this is only the first Gen, I am sure they have more to come.

Intel should never have wasted billions into McAfee. May be its next acquisition target would be Nokia or Ericsson, both at around ~30B, likely need 40B.

Competition is good.

The announcement is available on https://www.youtube.com/watch?v=lNY4J3yCzgI

was this fabbed at global foundries?

"Using the cutting-edge 7nm process, the CPU was independently designed by Huawei based on ARMv8 architecture license"

I assume it's the same 7nm TSMC process as for the Kirin 980.

TSMC. Global Foundries basically gave up on smaller processes a few months ago.

GF stopped working on their 7nm process and it wasn't used by any clients.

https://www.anandtech.com/show/13277/globalfoundries-stops-a...

What makes you think it has a link to GF? Just wondering.

Is it better than apple A12?

Define 'better'. This thing has 64 cores vs. 4+4 on the A12X (and 2+4 on the A12). It's built for a different use-case.

I guess you could benchmark the performance of a single core and compare those, but this won't power any mobile equipment and I bet it gets hot too.

It's like comparing a BMW truck to an Audi sports car. Better at what?

s/BMW/MAN

What sorts of lap times can you expect from a Mercedes panel van at the nurburgring? How much can an Aventador tow? These are the sorts of questions answered by benchmarks.

It has 8 memory channels, the A12 has 2. So it's definitely targeted at a different segment.

The top of the line Intel Xeon has 6 memory channnels and the top of the line AMD Epyc has 8 channels.

This title is misleading, should be "Huawei claims to unveil Industry's Highest-Performance ARM-Based CPU"

Well, they probably do have the title for now. What they have done is interesting, even if marketing had to smear it.

I dunno. Fujitsu's A64FX has incredible specifications.

Fujitsu's A64FX is 7nm-class, and has those SVE extensions for incredible throughput. It is expected to be in several supercomputers soon.

I expect that the A64FX is the fastest ARM in the world.

The Huawei specs are still a bit to limited for a full comparison, but it does use the same process.

The big questions involve how Huawei's 8 DDR4 channels compare to Fujitsu's 4 HBM2 banks, how Huawei's 64 cores compare to Fujitsu's 48.

Even if their IP is worse, they seem to have brute-forced their way to the top.

Does it really need pointing out that a headline on huawai.com is a claim by Huawai?

It's less that it's a claim by Huawei and more that the claims haven't been verified independently. Competitive benchmarking has a long history of results which are hard to reproduce and comparisons which “accidentally” left the competing product in an unusual state, so it's prudent to distrust them until they can be verified independently.

It's a press release. They're always worded like that.

I am not sure what industry you are in but if you go read press releases from the likes of most of the networking companies you see all the time “bob networking unveils the biggest flubar...” or the like as standard practice. Heck I have had to edit some of these for technical correctness before the marketing team puts their foot in it..

Where are you looking?

For Cisco, their press releases seem to be here: https://newsroom.cisco.com/pressreleases

I haven't looked through all of them because I'm lazy, but scrolling through them I haven't yet found one that isn't in the 3rd person.

And in case anyone is wondering why they do that: press releases are intended to be used in an article verbatim. Headline and all. Press releases are written from third person perspective because that's how a reporter would write it.

And if you go on Google and search some text from a press release, you'll find tons of articles using the headline and body completely verbatim like it's their own article. This is intentional.

It's absolutely standard to do that for press releases.

>I'm not sure why Huawei did this.

Because that's how you do press releases.

News outlets print them verbatim, and it's traditional to read like that (as opposed to e.g. "we released a CPU" which on a news outlet it would make it seem like the outlet's team released the CPU").

We reduced the title above.

> Based on the TaiShan servers, Huawei Butt also provides elastic butt services

Looking at huaweicloud.com, the "elastic cloud" servers seem to be all Intel :(

Looks like someone uses "cloud to butt" extension...

Of course. What, some people still don't use it?

Hmm, I feel like I've read this before. Oh right, the HiSilicon K3V2, which at the time was touted as the fastest ARM CPU for mobile devices. Except in real usage it was downclocked ~40% (including the memory) to keep temperatures low (still overheated) and the GPU performance was underwhelming. Really nice chip for tinkering with, no limits on any clocks or voltages, you could do anything you wanted with it.

No you haven't. HiSilicon K3V2 is a 6 years old mobile chipset built on ancient 40nm technology with DDR2 RAM, so like every other mobile chipset in existence, it has massive thermal issues due to non-existent cooling—even more so due to the terrible manufacturing process.

This is a server CPU built with modern processes, so thermal problems are out of the way.

https://www.engadget.com/2012/02/26/huawei-adds-home-grown-k...

I'm not dissing on the new chip, it's just that everyone claims their brand new silicon has the highest performance to date in their press releases.

And, well, they're usually right at the time of release. You can't blame them for wanting to market that they have the best specs at the time.

Reply on news.ycombinator.com