Slashdot Mirror
How Qualcomm Tried and Failed To Steal Intel's Crown Jewel
An anonymous reader shares an article from Bloomberg: In early November, Qualcomm Chairman Paul Jacobs stood on a stage in the heart of Silicon Valley and vowed to break Intel's stranglehold on the world's most lucrative chip business. The mobile internet and cloud computing were booming and the data centers running this digital economy had an insatiable thirst for computer servers -- and especially the powerful, expensive server chips that Intel churns out by the million. Qualcomm had spent five years and hundreds of millions of dollars designing competing processors, trying to expand beyond its mobile business. Jacobs was leading a coming-out party featuring tech giants like Microsoft and HP, which had committed to try the new gear. "That's an industry that's been very slow moving, very complacent," Jacobs said on stage. "We're going to change that."
Less than a year later, this once-promising business is in tatters, according to people familiar with the situation. Most of the key engineers are gone. Big customers are looking elsewhere or going back to Intel for the data center chips they need. Efforts to sell the operation -- including a proposed management buyout backed by SoftBank -- have failed, the people said. Jacobs, chief backer of the plan and the son of Qualcomm's founder, is out, too. The demise is a story of debt-fueled dealmaking and executive cost-cutting pledges in the face of restless investors seeking quick returns -- exactly the wrong environment for the painstaking and expensive task of building a new semiconductor business from scratch. It leaves Qualcomm more reliant on a smartphone market that's plateaued. And Intel's server chip boss is happy.
Less than a year later, this once-promising business is in tatters, according to people familiar with the situation. Most of the key engineers are gone. Big customers are looking elsewhere or going back to Intel for the data center chips they need. Efforts to sell the operation -- including a proposed management buyout backed by SoftBank -- have failed, the people said. Jacobs, chief backer of the plan and the son of Qualcomm's founder, is out, too. The demise is a story of debt-fueled dealmaking and executive cost-cutting pledges in the face of restless investors seeking quick returns -- exactly the wrong environment for the painstaking and expensive task of building a new semiconductor business from scratch. It leaves Qualcomm more reliant on a smartphone market that's plateaued. And Intel's server chip boss is happy.
In the land of the blind, the one-eyed man is king. There's no one else challenging Intel for supremacy - unless AMDs EPYC architecture does the job?
In their effort to move stock, Bloomberg conveniently ignores the speculative execution debacle at Intel. Intel's chips are shit right now.
They need a registered+unregistered ECC-capable DDR4 memory controller capable of driving at least 2-4 sockets and either hypertransport lanes to a PCIe controler, or onboard PCIe 4.0-5.0 lanes ranging from 16 to 48 minimum, depending on the market segment they are trying to capture, ideally supporting bifurcation along all power of two possibilities.
If they did that their chips would be capable of driving PCIMG passive backplane motherboards, actual x86 style motherboards, and the full range of consumer, professional, and industrial grade hardware.
This isn't rocket science. The technologies, licensing, and engineering are all non-trivial, but also well within the capabilities of companies like Qualcomm. The fact that they managed to fumble this bad enough to take themselves down is both technological and political in nature (the trade war with china closely aligns with Qualcomm's failed attempt, doesn't it?)
Seems to have been written (or edited) by someone without a semiconductor background. The biggest question I had was what processor architecture were they building around, something the Bloomberg piece never seemed to answer.
If they mean this it's 48 cores and based on ARMv8. Potentially interesting, but the piece lacks all the technical detail about how Qualcomm intended to position the chip technically against Intel, and what advantages it might have over competing ARM-based offerings.
But "ARM" never even appears in the Bloomberg article...
Lawrence Person (lawrencepersonh@gmailh.com (remove all "h"s to mail)
http://www.lawrenceperson.com/
they just focused on the corporate dealings of qualcomm...
technically if they can focus their efforts for centriq on the 5G edge (fixed in home modems and cell sites with caching, SDN etc) then they have a chance to expand their footprint otherwise intel will own the server and have a 5G modem competing with qualcomm which is not good for corporate returns at all...
AMD has been trying to get an edge. They did for a few years, but lost it.
Server chips are not mobile chips. Qualcomm was in over their head's before they started.
They never had a chance.
Some of us remember how much of the design of the DEC Alpha was stolen by Intel for the Pentium. See https://www.nytimes.com/1997/0.... Between this and the theft of VMS technologies to create Windows NT, DEC went bankrupt and stopped producing new technologies to be stolen.
You don't just "walk in" and take over a huge market.
Who lost the most money in this?
AMDs 7nm Epycs promise to be truly epic while the current 12nm ones are already good value. AMD supposedly only has a 1% share of the server market right now, but there is a whole lot of evaluating going on. I can imagine 15-20 share of that lucrative market by this time next year.
Now what we have is AMD+TSMC vs Intel. I'm not betting on Intel in the long run. I'll repeat my prediction that Intel has no choice in the long run but to follow AMD's leave and go fabless. Now... the emerging TSMC fab monopoly, that's something to worry about. On balance, not as bad as the traditional Wintel monopoly, but still bad. Keep in mind that the lithography equipment business had been a near total monopoly enjoyed by ASML for years now, and that has somehow worked. It's beyond me why that works.
We haven't heard the last of the ARM server effort. Intel dodged a bullet this time by pure accident. There will be more barbarians at the gate, not necessarily Qualcomm, but they are hardly out of the game.
When all you have is a hammer, every problem starts to look like a thumb.
Intel's transition to 10nm is delayed until late next year at best, whereas TSMC is selling (similar-size) 7nm chips en-masse today. Furthermore, Intel is facing a 14nm chip shortage due to their long-term planning on having moved to 10nm already, which is hitting the server chip business hard. Now is the time when Qualcomm should've doubled-down and pushed into the market.
Corruption is convincing someone that the selfless ideal is the same as their selfish ideal.
Intel has slept very long and was happy to gauge its customers with inflated prices. AMD is at least 5 years ahead of Intel now, probably more. Ideal situation would be if longer-term they basically have roughly equal market shares to keep each other honest. What Intel does when they have no competition, we already know. There is no reason to assume a dominant AMD would be much batter.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.
It seems to be a one-true Scotsman argument. ARM chips account for 85% of processors currently, Intel only 15%. Sure they dominate the Windows PC and "Cloud Data Center" markets...... Trouble is, your more likely to be reading Slashdot on a non Windows non Intel device these days, with your office or home based server being ARM based (e.g. a Synology RAID). The "true" Scotsman in this claim has changed from "processors" to "server processors" to "cloud scale data center processors".
Sure Intel still dominates the "cloud" data centers, with Xeons running big assed server racks and Qualcomm don't have market share with their ARM based server.... but that's not how the ARM world works. It's not *one* supplier that overwhelmed Intel in the other markets, it was thousands of other companies making thousands of competing commodity products. Pecking away until Intel is driven from that market.
The big growth in cloud servers for Intel *was* China, but the trade war means Intel gets hit with big phased in price hikes in China, while Chinese companies want to sell ARM based servers. Its not like these trade wars can end, because they never had a win scenario, the PR for the trade war *is* the win scenario for Trump. The war is the win. At best, the adults in the room, might resurrect the TPP and EU trade agreements (which locked China out of markets if it infringed IP) and label them "Trump" agreements to save face, but that's a long shot. Most likely the tarrifs will continue for years and Chinese ARM server makers will take over.
I'm not bullish on Intel. They seem to be complacent and in decline.
How is AMD 5 years ahead? The only thing they have over Intel seems to be their "infinity fabric" interconnect architecture, allowing them to make efficient multi-die packages. It's good, but I can't imagine that's 5 years worth of advancement.
Other than that, AMD is in the same ballpark as Intel, maybe a little bit behind. That's still impressive, real competition after all these years, however AMD better not fall asleep now that they have (hopefully) woken up Intel if they don't want to get passed again.
AMD has similar architectural issues to deal with. The epic EPYC PSP problem is still a mind-numbing issue. AMD has pretty much buried it under the rug, not that Intel's Meltdown/Sceptre efforts have been successful.
Qualcomm (or others) could turn RISC designs into something really useful, but there's a cart-and-horse problem that Oracle/Sun, IBM, Moto, Apple, and others know too well.
Phone and IoT/industrial IoT are still growth areas. Remember that the telcos are going to try to obsolete your entire kit with 5G replacements, more expensive phones with kitchen-sink add-ons and more-- soon.
Yeah, Qualcomm did some pretty stupid things. But AMD needs to keep looking over their shoulder. We all do.
---- Teach Peace. It's Cheaper Than War.
Qualcomm (or others) could turn RISC designs into something really useful, but there's a cart-and-horse problem that Oracle/Sun, IBM, Moto, Apple, and others know too well.
All modern processors are internally RISCy, so no. There is no problem. Everyone is using RISC designs. Did you mean to say ARM?
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
No, the Intel/AMD/others x86-x64 familes are CISC processors, no matter the handling underneath. It came to pass that the layers underneath were recently shown to have exposed RISC core access, but such access is disabled so far as most all humans are concerned.
The instruction set vs execution over clock cycles that determines the practical difference. ARM is based on the 6502 instruction set, adapted for larger memory models and in some cases, math, depending upon the implementation. I consider the ARM family RISC, but not the Intel/AMD families mentioned.
---- Teach Peace. It's Cheaper Than War.
The only thing they have over Intel seems to be their "infinity fabric" interconnect architecture
That is definitely not the entire list. Another huge one: AMD cache design is not vulnerable to Meltdown. I agree, definitely not 5 years ahead. Maybe 1 year ahead in process tech because of Intel's engineering misfortune. Roughly even in CPU design.
When all you have is a hammer, every problem starts to look like a thumb.
No, the Intel/AMD/others x86-x64 familes are CISC processors, no matter the handling underneath.
I didn't say otherwise, so you don't get to open with "No," there. I said they were internally RISC, and they are. It's a fact.
It came to pass that the layers underneath were recently shown to have exposed RISC core access,
Completely, totally, and in all other ways irrelevant. RISC won where it made sense for RISC to win, inside the processor. CISC won where it made sense for CISC to win, in the instruction set.
but such access is disabled so far as most all humans are concerned.
Surrrrrrre it is. If you trust your microprocessor manufacturer to get everything right, anyway.
The instruction set vs execution over clock cycles that determines the practical difference.
Processors are virtually all (outside of embedded) multicore and OoO now, there is no practical difference except that CISC effectively offers instruction set compression.
I consider the ARM family RISC, but not the Intel/AMD families mentioned.
That's why I said they were "Internally RISCy" and not "RISC". But then you went off half-cocked attacking a straw man. HTH, HAND!
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"
Idiots.
I really can come up with only one description for all this childish behaviour: Idiots.
An Investor who thinks he can enter a complex market in less than ten years is simply an Idiot.
An Entrepeneur who promises to enter a complex market in less than ten years is simply an Idiot.
Lets check the market:
Google Mail needed seven years to conquer the market.
Android needed six years to conquer the smart phone.
Linux needed 12 years to vanquish the commecial unices.
Whatsapp took six years to rise.
Facebook even needed nine years to get where it is now.
Twitter needed seven years too.
AMD needed five years to create the Ryzen Architecture. And will need another five years to gain greater market share.
Sooo... what is the moral of the story? If someone promises you he can conquer the world in a year take your money and put it somewhere else.
I really think ARM might in the long run be able to catch a part of the market share from Intel. But it will take many years and many little steps.
Where is the cheap Mini-ITX board with a Snapdron 850 and up to 32BG memory?
Where can I get a cheap Add-In-Board with 128 ARM cores for my PC, something like an Xeon Phi for cheap?
Whats the driver situation for ARM multi media solutions?
Is there any standard how ARM Desktops, Workstations and Servers are working, booting, initializing?
Bullshit. There is nothing except broken unrealistic promises.
"Life is short and in most cases it ends with death." Sir Sinclair
ARM is based on the 6502 instruction set
No. Just no. Have you even looked at either of them?
When all you have is a hammer, every problem starts to look like a thumb.
Ok, not really the 6502 instruction set per se, but rather, similar to the 6502 in that it didn't use DMA.
---- Teach Peace. It's Cheaper Than War.
Your just stringing words together, aren't you?
6502 processor didn't use 'DMA'? DMA is when a peripheral accesses main memory. Apple 2s used DMA, pretty sure C-64 did also.
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
Long day.
Read about it here: https://www.engineersgarage.co...
---- Teach Peace. It's Cheaper Than War.
Also, AMD can compile custom chips, including a variable amount of powerful GFX. See all the consoles with AMD CPUs.
Most ACs are not even worth the keystrokes to insult them. Be generically insulted by this and ignored otherwise.