How Qualcomm Tried and Failed to Steal Intel's Crown Jewel | Hacker News Reader

How Qualcomm Tried and Failed to Steal Intel's Crown Jewel

bloomberg.com

75 pointsItsTotallyOn41 comments

>> However, Chandrasekher had earlier brokered a deal with the Chinese province of Guizhou to fund part of Qualcomm’s server chip work. In return, the local government demanded the transfer of chip designs and exclusive rights to sell the processors in China.

This is common and the reason people say we don't have free trade with China. Many companies have made such short sighted deals which will be the death of US economic strength if it's allowed to continue.

China is not banning Qualcomm from selling chips in China if Qualcomm doesn’t give up its design, but simply offers a deal that Qualcomm can choose to accept or not. One may argue whether the deal is good for US economy strength, but it is still free trade.

> simply offers a deal that Qualcomm can choose to accept or not

With China's culture of corruption, "or not" comes with a giant asterisk of "will face harassment from the communist party"

What you said is true in many cases, and it should be stopped. However it does not apply to Qualcomm. Firstly yunnan is an underdeveloped region in China, its provincial governement doesn’t really have the power to harass Qualcomm at a meaningful level. Also Qualcomm has strong leverage against Chinese government because its products are hard to replace. Recall what happened when Qualcomm stopped selling its products to ZTE.

It will end up applying to them regardless of providence. Companies need to simply stop selling to Chinese companies. It's not worth it due to IP theft. But hey the Chinese government will just steal it anyways.

Would it still be free trade if all your design and IP you give up to not the Chinese JV but "Chinese" in general, and they then uses all all those learning into something of their own?

And no, you cant sue them.

It’s a breach of contract, and it’s wrong. However, when Qualcomm sells its tech, it should factor in this counterparty risk and decide if it is still worth it.

I’m not defending this kind of inexcusable behavior; I’m just suggesting free trade means Qualcomm or any private business can make its own decision.

As long as I'm not misunderstanding you, it seems you are confusing free trade with free to make a decision or not. Free trade means you are able to sell (or buy) things across borders free from government intervention. In this case, the government says you can only sell if you pay a form of tax (paid by IP rather than money).

In particular, what people believe is unfair is how these barriers are applied unequal. If Chinese companies want to sell to the US, it's free of such hinders. The other way around, on the other hand, it's not. (obviously this is not true considering the latest tolls and what not, but that's the core premise anyway).

Btw. Compare this to a situation when somebody is asking you to kill someone, or they'll kill you. You're free to make a decision, but not necessarily free as what people usually refer to when talking about freedom.

Source: https://en.m.wikipedia.org/wiki/Free_trade

I do agree with you on the general notion of free trade. But in this particular case, there is no mention of any barrier imposed on qualcomm products; the deal is about jointly creating a new product. What I disagree is claiming free trade while suggesting qualcomm should not do certain deals for national interest reasons.

Ah, I did misread you! I would guess it's rather complicated though, as it might be part of a bigger "agreement" with the Chinese government in order to ensure smooth business in general - but that's of course only speculation on my part.

>> But in this particular case, there is no mention of any barrier imposed on qualcomm products; the deal is about jointly creating a new product.

There is no mention, but often what happens is a company wants to do business in china and is told no while being presented an offer like this.

>> What I disagree is claiming free trade while suggesting qualcomm should not do certain deals for national interest reasons.

I agree with you that a company is free to make decisions any way it wants within the law. This is why we have ITAR to protect against weapons tech going to the wrong places. But if the US took a similar position with regard to other tech besides war-making, these deals wouldn't be legal. I wouldn't suggest we're having an economic war, but all nations are in economic competition and the stakes are potentially very high.

It's a shame because the Centriq[0] was objectively an excellent and very fast chip. Last year I was doing make -j46 kernel builds[1], but even single core performance felt better than Xeon. And then they dropped it out of the blue in about May this year.

[0] Codenamed Hawker and Amberwing, based on the Falkor core: https://en.wikichip.org/wiki/qualcomm/microarchitectures/fal...

[1] https://rwmj.wordpress.com/2017/11/20/make-j46-kernel-builds...

A colleague of mine did a performance deep dive as well: https://blog.cloudflare.com/arm-takes-wing/

It's sad to see Qualcomm ditch such a good platform.

Did they actually publish the data sheets for this one? Is there mainline kernel support?

There is mainline kernel support. I don't know about how much is published as data sheets but it's basically an ARMv8 processor with GICv3 which boots using completely standard SBSA/SBBR. It currently boots using a standard Linus kernel.

They gave out a bunch of servers to many companies including Red Hat (where I work), but I don't think they ever made it to retail. That's a shame because they are/were spectacular CPUs.

Bizarrely breathless and short-sighted article when Intel has fallen so far behind in process shrinks and AMD multi-core is on the rise.

The article does have an incredibly narrow focus. It's as if the author doesn't know AMD exists, when in fact AMD is a far far larger threat to Intel due to its chips native support of the x86 instruction set.

This is an objectively bad article. I'm still trying to figure out exactly which November the author is talking about. But I do think there is an interesting story in there if it was written by a better journalist with some/any actual insider information.

Agreed.

I'm still trying to figure out which Qualcomm product the author is talking about. Does it exist? Was it bad? Did people buy it? If not, why?

I don't believe the author's justification as to why Qualcomm was in the best position to make a success of an ARM server chip - there's a quote from a Gartner analyst and the assertion that you need a product road-map. What? Is that it? Surely the important thing is to make a good chip, make it cheap and produce enough to meet demand. Do that consistently for two years and the big cloud vendors will start buying. I don't think you need to be Qualcomm to achieve that.

Why couldn't a startup company break into this market? For example, this one https://www.phoronix.com/scan.php?page=news_item&px=Ampere-e...

Maybe this:

“the Qualcomm Centriq 2400 features up to 48 Qualcomm Falkor cores—our custom Armv8-based CPU core”

https://www.qualcomm.com/products/server-processors

Agreed. AMD is the real threat. Lisa Su has done everything right in turning around AMD. AMD was on the verge of bankruptcy. Lisa Su took it over and now AMD is come back to life.

You could also say it was this guy: https://en.m.wikipedia.org/wiki/Jim_Keller_(engineer)

And now he is at Intel.

Intel doesn’t have an architecture problem. They have a manufacturing one. As good as Jim Keller may be it’s just not his domain.

And, I expect Intel also has a cost-base problem. They have 106,000 employees, compared to AMD's 8,900.

False comparison. Intel is doing a lot more than AMD, those employees are across many different products and business units.

Doing more things isn't necessarily a good thing, doing the right things is.

AMD has a ruthless focus on a couple of markets, a competitive product and clear leadship.

That makes them a threat whatever the headcount.

If EPYC starts gaining ground in the high margin/high volume server market then Intel has to double down and they might, the Core architecture was a response to the AMD64 and a good one.

They aren't out the fight but they are taking a few hits at the moment .

Apple with Cook as CEO is a great examlpe of how much impact a efficient supply chain can have.

Efficient supply chain saved Apple when Steve Jobs was still around.

Now it's their vertical integration and just how good they are up and down the stack compared to the competition.

There is not a lot of vertical integration in Apple's physical supply chain. Screens are sourced from Samsung (or have been last time I checked), phones are assbled by Foxxcon (same as above, might have changed).

Managing all that is what set's Apple appart from others. Besides brand, design and such. I'm in no position to judge the software part.

It’s Bloomberg, so it’s possibly not beggering belief that this is little more than a PR piece for Intel. I’m not saying it’s paid for or anything, but PR agencies are always there with the outline of a story for reporters too lazy to bother with actual journalism. While this piece isn’t recurring like a submarine, it still has thst general stink.

I can’t think of a plausible reason beyond that, to ignore the role or AMD and Intel’s recent struggles.

Qualcomm has been facing multi-front battles in the past few years: Apple Lawsuit, NXP deal nixed, Broadcom hostile take over, China problem, Regulator problems, Board In-fights, pressure from activist investor to "cut cost"/split the licensing business... etc, etc. Making the arm server business from 0 to a commercial hit would be too much to ask. Although it did have the best chance.

I was disappointed to hear about QCOM shuttering the team building the Centriq, it seemed really promising. At least there’s still (on the ARM server frontier) Cavium ThunderX2, with which GigaByte recently started rolling out ARM servers and workstations. (Of course, there’s still also AMD, OpenPOWER, etc. in terms of data center competition for Intel.)

Yes, Cavium has been kicking ass. There's also Ampere's new eMAG thing. And HiSilicon server chips exist… somewhere. (https://www.worksonarm.com/cluster/ apparently has some)

>Qualcomm President Cristiano Amon said the company has "right-sized" the server business for the market opportunities.

I hate this corporate speak. 100% of the time this means that they fucked something up and they're going to can the entire thing because they've re-set their expected market share to 0. The only reason they don't state they've canned it is so that schmuck investors and engineers don't realize the project is dead.

The size they are now isn't "Right-sized" for the opportunity, it's "right-sized" for a project that missed the opportunity, and to pretend that the original investment wasn't the right size is just white washing bad management decisions.

Seems like the tides have turned on this one, with Intel stealing Qualcomm’s breadwinner modem business, thanks to the help of Apple of course. It’s funny how things can change so quickly.

Cavium is actively competing with Intel, offering ARM for the server space. You can rent cavium CPUs from http://Packet.Net for a lot cheaper than Intel

Qualcomm's server unit demise has more to do with management than tech - Cavium built it's processors with fewer dollars than Qualcomm.

More examples of teenage engineering discipline. OpenPOWER is the one you want.

OpenPOWER will never gain traction outside IBM. They are kept on life support by Google et al in order to have pricing leverage over Intel. That's it.

And ARM in the datacenter has a really stellar trackrecord huh? Everything changed post meltdown. Gmail is running on OpenPOWER.

They did it to show that they COULD switch away from x64 if needed to. Have they continued buying OpenPOWER systems at the same rate as they buy x64 hardware?

I don't have any FP&A details, only technical public and private. They are vocal about using it for GPUs. It's a much cheaper and more balanced way to glue Nvidia GPUs versus Nvidia DGX on x86 which also has severe bottle necks at the host bridge point. The private is gmail, which happened much later than the negotiating tactic point. It fits into a number of other goals including a completely open firmware stack. You only have to have cursory headline knowledge of the news for the past 2 years to understand why this is no longer about initial procurement cost.

Reply on news.ycombinator.com