Why Google and Amazon could end up cooking their own chips

It used to be that the ante for being a serious player in the technology game was your own data centre. Or several. On this basis, Google, Facebook, Amazon, Microsoft and very few others have constructed hugely expensive data centres that centre much of the web's activity on them. But among these web giants, it may no longer …

[Anonymous Coward]

...buy TSMC.

[Anonymous Coward]

Nah

Give it a couple more months and buy AMD once their management have finished crashing it into the dust.

[Piro]

Re: Nah

But AMD already span off their manufacturing arm as Global Foundries, right?

[HMB]

Re: Nah

"Give it a couple more months and buy AMD once their management have finished crashing it into the dust."

Do you think that by then someone might be competent enough to write an HDMI output driver for a graphics card that doesn't screw up the output to an HDMI TV?

Neither ATI nor AMD seem to be capable of doing this. (The drivers do this mind boggling stupid thing of resizing the output for overscan when there is none for DTV and then resizing back, completely messing up the 1:1 pixel mapping.)

[Charles 9]

Re: Nah

"(The drivers do this mind boggling stupid thing of resizing the output for overscan when there is none for DTV and then resizing back, completely messing up the 1:1 pixel mapping.)"

They fixed that already. Use the "standard" rather than the "optimized" resolutions and it goes back to 1:1. The optimized resolutions probably come because not all DTVs are digital displays and thus may actually have overscan issues.

[Charles Manning]

Why buy into a dying technology

x86 is not the way to go in the future and that means buying AMD would be pointless for the server side of the biz.

64-bit ARMs and custom processors (ARM + FPGA) make sense.

[J. R. Hartley]

Commodore and MOS worked well.

[Voland's right hand]

Amazon has another ace up its sleeve

Amazon has been incredibly successful in creating wholesale elements out of key infrastructure components for its business. AWS started as its shopfront hosting service, ditto for most of its other cloud services. It now provides fulfillment, shipping, financial services, etc for companies that are competing with it at some level. All of that is profitable and all of that has customers.

So even if Amazon itself does not have the volume based on its own devices, it can and will establish yet another Amazon wholesale product. SOC as a Service anyone?

[Anonymous Coward]

There's a certain irony here; remember Apple was one of the original shareholders in what was then ARM Ltd.

[IGnatius T Foobar]

Network Computing

I for one am happy to see all of this happening. It's the dream of Network Computing finally come to fruition. In the not too distant future, every upright desktop monitor will contain a SoC with a full Android stack running on it -- plugging in a computer will be optional.

[Alan Denman]

Samsung - designed to fail.

Samsung's biggest mistake was allowing Apple to take part in the slightly modified designs of the Samsung Exynos range of ARM Socs. Add to that the creative lazy hysteria of the media and one would think they were all Apple designs.

It is only in version A6 where Apple has been the main instigator.

Luckily for Apple buyers , competition from experts like Samsung will eventually drive forward Apple ARM takeup.

So whilst Apple tread water in using Coretex A9 chips, Samsung and others have already moved onto using Cortex A15 chips.

[karlp]

Re: Samsung - designed to fail.

Saying that Apple is using A9 cores is quite disingenuous is it not?

I could just as truthfully say that they are already using A15's.

In reality both, and neither, of those are true.

http://www.anandtech.com/show/6292/iphone-5-a6-not-a15-custom-core

In my world A15 shows interesting promise for certain computing workloads, but I can't help but feel that Apple did precisely the right thing by incorporating what is good about A9 and A15 in their own package that is laser focused on power efficiency and performance as hand in hand requirements.

My guess is that, at least for a time, others (including amazon) may decide they need the same design latitude in order to drive excellent user experiences.

[Anonymous Coward]

Re: Samsung - designed to fail.

Its so funny reading what Fandroids who know nothing about Apple chip design write about the subject because it reveals the extent to which their comments are driven by wish fulfilment rather than based on any knowledge or evidence. The truth is no-one outside Apple and their closest partners knows for sure what the Apple A6 processor is based on but the best evidence points to it being a fully custom build that is neither A9 or A15. AnandTech have done a full analysis here and also point out why they believe the approach they appear to have taken has many advantages over both Cortex A9 and A16. You can read about it here:

http://www.anandtech.com/show/6292/iphone-5-a6-not-a15-custom-core

Other commentators have simply pointed out the evidence is already in it is a kick-arse solution:

http://seekingalpha.com/article/883901-apple-s-iphone-5-chip-design-seriously-impresses-is-the-competition-asleep

I know this doesn't fit with your narrow world view that Apple don't design or do anything themselves, but the reality is they are one of the largest employers in the world of chip designers, period. Yes that's right.

[Anonymous Coward]

Google to go for the Jugular ?

And go the whole hog and buy ARM. That would piss on MS's chips for RT ;)

[Anonymous South African Coward]

History repeating itself?

IBM tried to lock out other vendors from its mainframe biz back then.

Microsoft also tried to lock out other vendors from Windows (most notably Word/Excel competitors).

Should be interesting to see what happens going forward...

[Merrill]

Custom VLSI could reduce data center cost and power

Instead of having rack after rack of general purpose CPUs spinning endlessly through the same bits of code for low level functions, custom VLSI could do it a lot cheaper and more efficiently.

[Yet Another Anonymous coward]

Sounds reasonable

They want a bit of wire with a SATA plug on one end and a cat-5 socket on the other and an ARM in the middle.

Why does Amazon want 1000s of Xeons spinning away serving up e-books?

"In ten years, computers will just be bumps in cables." Gordon Bell (Feb 90)

[Big_Ted]

Re: Sounds reasonable

Because they have a massive cloud business

Plus for the Kndle Fire's they need tons of capacity for streaming services.

Both them and Google are into data in a massive way and I can see Google getting more into hardware with Moto now part of the business.

Makes perfect sense to me.

[David Halko]

What's old is new again!

It is very interesting that in the not-so-distant past, workstation & server manufacturers were all on Motorola processors and they moved to custom designed RISC chips. There was IBM PC and clones on Intel.

SPARC offered an open (royaltya free) rchitecture for dozens of designers and chip, board & system manufacturers - from embedded, to desktop, to to servers, to high-end SMP systems, to massive MPP systems. Anyone could design, burn, and build... everyone did.

The market seemed to consolidate back down to proprietary Intel chips, following the PC vendors.

Now, it seems vendors are moving back over to custom manufacturing, ARM seems to be SPARC with royalties. ARM controls the architecture, more aggressively than SPARC, but more loosely thatn Intel.

It should prove to be interesting into the future.

[John Savard]

Too Late

Now, if only Apple could have bought Motorola, the incompatibilities of PowerPC Macs and Intel Macs could have been avoided!

[Anonymous Coward]

They didn't have the money in 2004/5

and acquisitions like that weren't Steve's style.

In retrospect, it wasn't a bad decision to make the crossover, though I see how it annoyed many users.

[Ramazan]

Re: PowerPC Macs

PowerPC perfomance per watt was and still is inferior, Steve was absolutely right to dump that shit

[John 62]

Re: PowerPC Macs

I don't think it was all that different at the time of the switch, but it seems IBM had no interest in competing with Intel in the commodity processor space. Apple alone didn't have enough volume to make it worth the effort for IBM. But neither did many people realise just how serious Intel was about performance per watt (Pentium 4 was still fresh in everyone's memory back then) or the performance gains Intel achieved with the Bridge series of chips.

[chris lively]

No surprise

Nothing about this surprises me.

ARM has been making huge in roads into chip design and manufacture over the past several years. They have quality designs, great performance and with the capability of essentially "white box" labeling them, it absolutely makes sense for companies like Apple and nVidia to use ARM instead of buying from Samsung / Intel.

Which, is essentially what's going on. These companies are licensing designs from ARM, then sending them off to the "best" fabricator they can get. Of course "best" is defined by some combination of production quality, price etc.

Point is, they aren't actually "making" their own chips. Instead, they are just licensing the designs with minor tweaks and sending them to production. The days of hiring one firm to handle all of that are going away.

[TkH11]

IC Design

I'm surprised Apple didn't already employer ASIC designers and use other manufacturers, that's been the main stay of the IC industry for decades. FPGAs which other people have mentioned, may not be the way to go, if you want the most transistors on a die, if you want to keep the cost down to as low as possible, you don't want to waste space on the die (cost is related to the silicon area), if you want the lowest power consumption and highest speed then you opt for a full custom design mask-programmable technology, not FPGA. All depends on production quantities of course.

[Daniel B.]

I'll cheer...

For anything that makes us finally move off of the crappy x86 arch. If the car industry were like the PC one, we'd all still be driving VW Beetles, Kombis & Trabants.

[Ken Hagan]

Re: If the car industry were like the PC one

Ah, but those Beetles, Kimbis and Trabants would have warp drive, big bigger on the inside than the outside, and would be free with your breakfast cereal.

Intel seem to be doing fine, despite their crappy arch. I've been hearing about how ARM were going to take over the world for almost as long as I've been hearing about Linux on the desktop. Given what Intel do to batteries and what Microsoft have done to my desktop, I'd be delighted if either scenario came to pass, but after 10 years or more, I'm still waiting.

[Unicornpiss]

Vertical integration makes sense

...If you can swing it. And Google and Amazon have the resources to try. It worked for Commodore (MOS Technologies) in the 80s, and allowed them to deliver a price point that most others couldn't touch.

My only question is, how much can they actually save in this day and age, with chips as cheap as they are now? A very different time than when there were only a few players.

[Charles 9]

Re: Vertical integration makes sense

If power is not your thing, then yes ARM solutions are cheap as...well, chips. However, Google and Amazon have multimedia and more in mind and therefore need something with a little more oomph (one thing not always mentioned is that when it comes to raw computing power on demand, ARM needs help, usually GPUs or other DSPs). So there is still a need for an optimal chip: one capable of delivering the most power for the least power at the lowest price. That area is still considered bleeding edge so there are costs involved. Vertical integration helps to minimize those costs, so to Google and Amazon, it's a boon.