Google May Be Designing Chips In Threat To Intel

[Sequences]

Who needs x86? Whatever software google is using it's a recompile away from running on a new ISA.

MIPS or ARM doesn't really matter.

If people didn't switch from x86 to Itanium 10 years ago, what makes you think they'll switch to something else now?

 

[jvroig]

If people didn't switch from x86 to Itanium 10 years ago, what makes you think they'll switch to something else now?

I think you might have slightly misunderstood his message. He wasn't talking about a massive industry-wide ISA switch.

Instead, he just means Google will develop CPU's for their own internal use, running only their own programs. In that very limited scenario, recompiling software to a new ISA (whatever ISA they settle on for their in-house server CPU) can make sense.

 

[Imouto]

If people didn't switch from x86 to Itanium 10 years ago, what makes you think they'll switch to something else now?

ARM is cheaper and less power hungry even a full node behind?

 

[nismotigerwvu]

ARM is cheaper and less power hungry even a full node behind?

I think you are confusing an ISA and an implementation. There's absolutely nothing about the instructions that comprise ARMv7 or ARMv8 that make them more power efficient than current X86 (well X64 + AVX and well, you get what I mean). This efficiency advantage comes at the architectural level. For example, Krait, A15 and Cyclone all target much lower TDP levels and higher efficiency than Atom or Bobcat/Jaguar as with added transistors for extra computing power comes the cost of efficiency. There's absolutely no reason why someone like Qualcomm couldn't produce a Steamroller/Broadwell class ARMv8 implementations, however they would also run at Steamroller/Broadwell level TDP levels. The inverse is also true (AMD/Intel launching Krait-class X86 hardware), however neither is likely in the near future due to experience. Either of these scenarios would represent uncharted waters for everyone involved and despite all the talent at ARM/Intel/AMD it is just REALLY difficult to start from scratch on something like this. This explains why you see the ARM sector slowly creeping up in power (both computing and draw) while Intel and AMD creep down. The battleground will actually be fought somewhere in between the "comfort zones" of both sides, which my crystal ball predicts will be in some tablet/convertible form factor and/or in servers.

 

[John Connor]

and black magic does not exist either.

It did when I played COD. LOL! I cheated my butt off and my name was Blackmagic.

if Google were to make an x86 based chip and even if it was lower priced than Intel I still wouldn't buy it.

 

[Sequences]

ARM is cheaper and less power hungry even a full node behind?

A bike is cheaper and less power hungry than a car, but I'm still buying a car when I need to go the distance. Judging by the amount of distance Google seems to be going with their analytics, they probably have a long way to go.

And the assumption that they can actually recompile everything... nope.

 

[ShintaiDK]

If Google was only to use these for their servers. I cant see the ROI in it. Specially not with ARM where you can get almost any kind of design from 3rd party.

 

[jpiniero]

The idea is that they would take an ARM design and add some custom hardware instructions which would be useful in speeding up PageRank and/or their data mining efforts.

 

[Tsavo]

If Google was only to use these for their servers. I cant see the ROI in it. Specially not with ARM where you can get almost any kind of design from 3rd party.

They can offload a lot of compute from energy inefficient LCC's (large compute cores) to energy efficient SCC's....that alone would save them buckets of cash on electricity and cooling. They don't need LCC's for everything. And then there's the $$$ they save by not doshing out on millions of Intel parts every year.

They can also bolt on some ASIC logic onto those ARM parts to further reduce their expenditure on Intel/Electricity/Cooling. HTTP accelerated codecs, anyone .

There are far more good reasons for them to do this than there are bad ones.

 

[ShintaiDK]

They can offload a lot of compute from energy inefficient LCC's (large compute cores) to energy efficient SCC's....that alone would save them buckets of cash on electricity and cooling. They don't need LCC's for everything. And then there's the $$$ they save by not doshing out on millions of Intel parts every year.

They can also bolt on some ASIC logic onto those ARM parts to further reduce their expenditure on Intel/Electricity/Cooling. HTTP accelerated codecs, anyone .

There are far more good reasons for them to do this than there are bad ones.

Why not get this done by a 3rd party custom ARM maker? Its much cheaper, no headcount, no missing IP.

 

[Arachnotronic]

The idea is that they would take an ARM design and add some custom hardware instructions which would be useful in speeding up PageRank and/or their data mining efforts.

No ARM licensee can modify the ARMv8 ISA.

 

[Tsavo]

Why not get this done by a 3rd party custom ARM maker? Its much cheaper, no headcount, no missing IP.

Custom logic that no one offers/IP of their own that they don't want anyone else to gawp at.

Also, this is a long-term move. They aren't going to be running out of demand for servers any time soon, and the cost of a small fleet of ARM engineers is easily offset by the savings of reducing Intel/electricity/cooling costs over the long run.

Beyond that, it's not like they will be fabbing this stuff themselves. CPU designs to TSMC/etc then the boards made by Foxconn/etc. I'm pretty sure they have their own custom server stuff made for them already, so this is a pretty small step for them in terms of knowledge and manpower.

Heck, just reducing their reliance by 50% on Intel CPU's for 1 year will almost pay for the Agnilux acquisition itself.

They'll be saving shetloads of money by doing this...which is why they are doing this.

 

[Imouto]

I think you are confusing an ISA and an implementation. There's absolutely nothing about the instructions that comprise ARMv7 or ARMv8 that make them more power efficient than current X86 (well X64 + AVX and well, you get what I mean). This efficiency advantage comes at the architectural level. For example, Krait, A15 and Cyclone all target much lower TDP levels and higher efficiency than Atom or Bobcat/Jaguar as with added transistors for extra computing power comes the cost of efficiency. There's absolutely no reason why someone like Qualcomm couldn't produce a Steamroller/Broadwell class ARMv8 implementations, however they would also run at Steamroller/Broadwell level TDP levels. The inverse is also true (AMD/Intel launching Krait-class X86 hardware), however neither is likely in the near future due to experience. Either of these scenarios would represent uncharted waters for everyone involved and despite all the talent at ARM/Intel/AMD it is just REALLY difficult to start from scratch on something like this. This explains why you see the ARM sector slowly creeping up in power (both computing and draw) while Intel and AMD creep down. The battleground will actually be fought somewhere in between the "comfort zones" of both sides, which my crystal ball predicts will be in some tablet/convertible form factor and/or in servers.

You have to do it really wrong to make an ARM product that fits your needs consuming more power than a bloated X86 counterpart. But yeah, maybe because all ARM implementations are focused on low power use you can assume that anything that Google is planning will consume less power and be cheaper than Intel even taking in mind the development cost. Would be pointless the other way around.

Intel can kiss goodbye to a $500 million customer.

 

[ShintaiDK]

Custom logic that no one offers/IP of their own that they don't want anyone else to gawp at.

Also, this is a long-term move. They aren't going to be running out of demand for servers any time soon, and the cost of a small fleet of ARM engineers is easily offset by the savings of reducing Intel/electricity/cooling costs over the long run.

Beyond that, it's not like they will be fabbing this stuff themselves. CPU designs to TSMC/etc then the boards made by Foxconn/etc. I'm pretty sure they have their own custom server stuff made for them already, so this is a pretty small step for them in terms of knowledge and manpower.

Heck, just reducing their reliance by 50% on Intel CPU's for 1 year will almost pay for the Agnilux acquisition itself.

They'll be saving shetloads of money by doing this...which is why they are doing this.

You make it sound like developing a new CPU is close to being free. Even if we look at Apple. Its a multi billion cost for a new CPU to have even a marginal chance. And then you have to manufactor it afterwards. Not to mention extensive server validation and platform development.

And how much would they save? Silvermont already beat ARM cores on the performance/watt case. But even then, why spend billions on a risky R&D move instead of just buying from one of the many ARM makers.

I doubt Googles server farm is even close to be able to handle such a development cost, just for inhouse usage. Google got between 1-2 million servers (Estimated they had 900K in 2011.). Microsoft to compare got just around 1 million.

 

[jpiniero]

If people didn't switch from x86 to Itanium 10 years ago, what makes you think they'll switch to something else now?

Intel only targeted servers with Itanium. If they were to do a new architecture, they would have to deliver processors that run the entire line to make it work. Obviously, it would be much easier now than it would have been 10 years ago, with the dependency on MS gone.

 

[cytg111]

If people didn't switch from x86 to Itanium 10 years ago, what makes you think they'll switch to something else now?

Content dude, content.
Saying the android ecosystem dont have content? Is windowsrt failing? Why? (hint, difference not being arm)

Not saying either way, but it is not as simple as "if not 10 years ago" .. thats just ...... periods...... .. .

 

[cytg111]

You make it sound like developing a new CPU is close to being free. Even if we look at Apple. Its a multi billion cost for a new CPU to have even a marginal chance. And then you have to manufactor it afterwards. Not to mention extensive server validation and platform development.

And how much would they save? Silvermont already beat ARM cores on the performance/watt case. But even then, why spend billions on a risky R&D move instead of just buying from one of the many ARM makers.

I doubt Googles server farm is even close to be able to handle such a development cost, just for inhouse usage. Google got between 1-2 million servers (Estimated they had 900K in 2011.). Microsoft to compare got just around 1 million.

you doubt.. and i guess that at google hq's they got for more qualified and visionary ppl to make those calls. Suppose x86's move on android is not to googles, or partherships, liking? enemy of my enemy is my friend and all that. [Stuff] is not that simple or we would all be billionaires on the stockmarket... You a billionaire ShintaiDK?

Callouts/personal attacks are not allow. Nor for that matter is profanity
-ViRGE

 

[ShintaiDK]

you doubt.. and i guess that at google hq's they got for more qualified and visionary ppl to make those calls. Suppose x86's move on android is not to googles, or partherships, liking? enemy of my enemy is my friend and all that. Shit is not that simple or we would all be billionaires on the stockmarket... You a billionaire ShintaiDK?

Unless you show me some official Google information. Then nobody at Google HQ made any calls in this matter.

And headless speculation and rumours originating from fairyland is not exactly in small supply.

 

[Intervenator]

http://seekingalpha.com/article/1896211-intel-dont-believe-bloombergs-google-rumor

Lets put this thread to a rest. Read my link. Enough said?

 

[ShintaiDK]

http://seekingalpha.com/article/1896211-intel-dont-believe-bloombergs-google-rumor

Lets put this thread to a rest. Read my link. Enough said?

:thumbsup:

 

[scannall]

http://seekingalpha.com/article/1896211-intel-dont-believe-bloombergs-google-rumor

Lets put this thread to a rest. Read my link. Enough said?

The kind of talent you need to pull that off aren't people you get with a general help wanted ad. That being said, it doesn't make economic sense as their volume wouldn't be enough to justify the expense. If they need a custom ARM it would make a lot more economic sense to give Qualcomm a call.

 

[jpiniero]

http://seekingalpha.com/article/1896211-intel-dont-believe-bloombergs-google-rumor

Lets put this thread to a rest. Read my link. Enough said?

Seeking Alpha can be interesting, but you have to take that site with a large grain of salt. The title makes it look like Intel denied the rumor, but that's not the case.

 

[Imouto]

Seeking Alpha can be interesting, but you have to take that site with a large grain of salt. The title makes it look like Intel denied the rumor, but that's not the case.

Disclosure: I am long INTC, BRCM.

 

[Intervenator]

Read the article, and form your own opinion on it. Everything he says seems reasonable to me. If you want to nitpick over irrelevant pieces of information, be my guest.

Also, the name of the company that an article is based around is almost always in the title.

 

[Intervenator]

Quote:
Originally Posted by Intervenator View Post
http://seekingalpha.com/article/1896...s-google-rumor

Lets put this thread to a rest. Read my link. Enough said?
The kind of talent you need to pull that off aren't people you get with a general help wanted ad. That being said, it doesn't make economic sense as their volume wouldn't be enough to justify the expense. If they need a custom ARM it would make a lot more economic sense to give Qualcomm a call.

______________________________________________________________________________________________________________________________________

They need an entire team to design their own server chips... Hundreds of people will need to be hired if they want to get this done in a timely matter. You think they could do this without making the public aware that they are hiring?