Intel Quark architecture, "1/10th" the power use of Atom

[NTMBK]

If they will work with 3rd parties to make semicustom designs, I have a proposal! Strap a massive vector unit (say 512 bit) to the core, put ~60 of them on a ringbus, give it a GDDR5 memory bus, and call it a graphics card...

[trollface.gif]

 

[mrmt]

Isn't Intel going after ARM A15 with Atom and ARM A7 with Quark?

 

[NTMBK]

Isn't Intel going after ARM A15 with Atom and ARM A7 with Quark?

More like the M4.

 

[JDG1980]

Intel's corporate culture requires massive gross profit margins on everything. This is going to be a big handicap if they want to edge out ARM, since low-end ARM SoCs are dirt-cheap. And there are a lot of embedded applications where neither x86 compatibility nor the absolute best performance are required.

 

[knutinh]

Perhaps there are low-power/low-cost applications where x86 offers superior debugging/optimization/application support? (I don't claim to know)

If I was putting a low-power cpu in something like ballistic missiles or pacemakers, I would worry more about mapping designer intent into tested code into cpu execution than cost.

What is the number of developers capable of writing decent C# in Visual Studio compared to the number of people capable of writing decent code using C and exotic tools?

-k

 

[Sweepr]

1/5 of Silvermont core size, probably <1mm². Its fun to see Intel scaling down and ARM scaling up.

 

[simboss]

Perhaps there are low-power/low-cost applications where x86 offers superior debugging/optimization/application support? (I don't claim to know)

In the embedded world, ARM offer far better tools than Intel, simply because they are actually used in far more devices and by far more developers.
Add to that the fact that the ARM ISA is much simpler than x86, I don't really see any advantage for Intel in this field.

If I was putting a low-power cpu in something like ballistic missiles or pacemakers, I would worry more about mapping designer intent into tested code into cpu execution than cost.

For exactly the same reason, ARM is clearly better positioned than Intel.
For ballistic missile, I am not sure the low power is really a problem anyway, neither is cost, and afaik Intel may be a solution (but not Quark) because it actually needs a lot of processing power.

What is the number of developers capable of writing decent C# in Visual Studio compared to the number of people capable of writing decent code using C and exotic tools?

So you think C# in Visual Studio is more exotic than GCC for C/C++?
In the embedded world, C# is the most exotic thing you can find, and I would not let someone that prefers C# to C code for my pacemaker :whiste:


I don't really see what Intel could bring to this field at the moment, they might do but that would take a loooooooong time, probably even longer than what it would take to ARM to get to the server market.
The design cycle for these product are really long, most of them have just started to use ARM cores few years ago.

If the wearable computing thing takes off as fast as the smartphone market did, they might have a chance as this has more commonalities with the consumer side of things (short design cycles, performance is more a factor, ...).
But on the really embedded world (aka pacemaker or washing machines), the market is very different and I don't see what Intel could bring to it.

 

[MisterMac]

In the embedded world, ARM offer far better tools than Intel, simply because they are actually used in far more devices and by far more developers.
Add to that the fact that the ARM ISA is much simpler than x86, I don't really see any advantage for Intel in this field.



For exactly the same reason, ARM is clearly better positioned than Intel.
For ballistic missile, I am not sure the low power is really a problem anyway, neither is cost, and afaik Intel may be a solution (but not Quark) because it actually needs a lot of processing power.



So you think C# in Visual Studio is more exotic than GCC for C/C++?
In the embedded world, C# is the most exotic thing you can find, and I would not let someone that prefers C# to C code for my pacemaker :whiste:


I don't really see what Intel could bring to this field at the moment, they might do but that would take a loooooooong time, probably even longer than what it would take to ARM to get to the server market.
The design cycle for these product are really long, most of them have just started to use ARM cores few years ago.

If the wearable computing thing takes off as fast as the smartphone market did, they might have a chance as this has more commonalities with the consumer side of things (short design cycles, performance is more a factor, ...).
But on the really embedded world (aka pacemaker or washing machines), the market is very different and I don't see what Intel could bring to it.

Absurd ammounts of chips to sell at near cost prices due to the retarded cost of keeping fabs up in this day fo age.

I would expect quark to have been designed with margins around 10% or there of.


Intel is clearly understanding that the profit can be made from the "cloud servers" while the grunt R&D cost can be recouped on the topline from ultra lower power chips.

 

[simboss]

Absurd ammounts of chips to sell at near cost prices due to the retarded cost of keeping fabs up in this day fo age.

I would expect quark to have been designed with margins around 10% or there of.


Intel is clearly understanding that the profit can be made from the "cloud servers" while the grunt R&D cost can be recouped on the topline from ultra lower power chips.

Again, the embedded market is VERY different, it's not about cores and process, it's about IOs and integration, tools and ease of use (including support)....
Cost is also obviously a factor, but it is in any business, so it does not really make any difference.

You can't really just churn out chips at cost and hope someone will use them. In server and PC it works like that mostly because:
- all PCs and server in the world uses the same abstraction layers (mainly OSes). You can't do that in embedded, the overhead is too big.
- x86 has dominated the market, so you had no choice.
- The HW interfaces are well standardised and mostly self describing (PCI, USB...). So no need to develop anything new to switch from AMD to Intel.

In embedded, each new project requires to develop a lot of HW dependant code, mostly bare-metal, so you better have a simple, easy to program chip and good cheap tools and support.
That cost money and time to design, plus time to convince everyone that your solution is better than someone else's.

 

[Phynaz]

"If Intel gets big into fully synthesizable circuits then they will do for synthesis tools what they did for computational lithography, and everyone will benefit from Intel's push for solid and reliable software design tools"

could u pls explain this in english. thanks

This please!

 

[Phynaz]

In embedded, each new project requires to develop a lot of HW dependant code, mostly bare-metal, so you better have a simple, easy to program chip and good cheap tools and support.
That cost money and time to design, plus time to convince everyone that your solution is better than someone else's.

Did you miss the part in keynote where Intel said they did exactly this?

http://www.eetimes.com/document.asp?doc_id=1319447&

 

[Exophase]

You guys are jumping to conclusions that Intel is really even positioning this for the microcontroller market (where Cortex-M series for instance sells). The mixed signal and power stuff you want there isn't necessarily a good fit for Intel's 22nm or even 32nm process. They showed the thing for use in smartwatches (where applications SoCs are currently used), that's a different position and sounds more like going against Cortex-A5 level stuff. The fact that the reference board has wifi and 3G on it always points in the same direction - you will rarely find a microcontroller eval board with these things.

And getting down to say, Cortex-M0 level of transistors is just unfeasible, not while supporting anything like full x86. Even full ARMv7 is too much for that level. It's too much even for Cortex-M3/M4, for that matter. I guess it's possible that they're targeting a reduced ISA, but I doubt it.

Some amount of assembly code still shows up now and then in embedded stuff, as well as libraries - personally I'd be reluctant to go with something like this in embedded due to a total lack of other people making similar cores. That and I don't see Intel suddenly offering a huge product catalog with lots of different variations for your particular needs like ST, Atmel, Freescale, TI, etc etc do today. Maybe if they really do end up licensing the core to big microcontroller makers and it floods the market, but it's hard to see why they'd be so compelled to change to it.

Also seems like no one is even asking about performance..

 

[Fjodor2001]

So each Quark x1000 processor core would be 1.12 mm2 if built on Intel's 32nm if I am reading this correctly.

So it's bloated compared to the 0.45mm2 ARM Cortex-A7 that has been around for years...

 

[ShintaiDK]

So it's bloated compared to the 0.45mm2 ARM Cortex-A7 that has been around for years...

Wrong conclusion. Its 1/5th of 22nm Silvermont.

 

[Exophase]

Did Intel even say 1/5th the size of Silvermont? EE Times says Quark is on 32nm right now. HotHardware says "1/5th the size of current Atom processors" which could mean Saltwell since Silvermont is not actually in a shipping product yet.

 

[ShintaiDK]

Did Intel even say 1/5th the size of Silvermont? EE Times says Quark is on 32nm right now. HotHardware says "1/5th the size of current Atom processors" which could mean Saltwell since Silvermont is not actually in a shipping product yet.

Silvermont.

Update: Intel has sent over a quick clarification on size; the size comparisons are meant to be core-to-core rather than SoC-to-SoC. Specifically compared to the 22nm Silvermont core.

 

[Zodiark1593]

Could the Quark make a case for itself in automobile computer systems, and future diagnostic tools?

 

[ElFenix]

Any bets on what the architecture is like? I'm going to call "tweaked Pentium".

that's the (still) current atom architecture. though i suppose if you cut the pipeline stages back down to 5 from 16 you might be able to save some transistors.

 

[Exophase]

that's the (still) current atom architecture. though i suppose if you cut the pipeline stages back down to 5 from 16 you might be able to save some transistors.

While Saltwell is dual-issue and in-order like the original Pentium the similarities end there. It's a very different design with no common ancestry.

Quark does sound Pentium derived due to the mention of the ISA being "Pentium-compatible." There were instructions added to Pentium Pro (P6) like cmov, why else would they draw the line at Pentium ISA if there was no design relationship?

 

[Vesku]

While Saltwell is dual-issue and in-order like the original Pentium the similarities end there. It's a very different design with no common ancestry.

Quark does sound Pentium derived due to the mention of the ISA being "Pentium-compatible." There were instructions added to Pentium Pro (P6) like cmov, why else would they draw the line at Pentium ISA if there was no design relationship?

I'm kind of wondering if it's a variant of the pentium derived core used in the Xeon Phi.

 

[MisterMac]

I'm kind of wondering if it's a variant of the pentium derived core used in the Xeon Phi.

I was contemplating this too - in fact being a PHI Core itself with some added x86 legacy ancestry on top - minus the massive fat vector units.

 

[Fjodor2001]

Wrong conclusion. Its 1/5th of 22nm Silvermont.

So which core has the biggest die area, if on the same process node, the Cortex A7 or the Quark? And how many percent bigger is the larger one of them compared to the other one?

Also, what about the performance and power efficiency?

 

[NTMBK]

Is Quark the Claremont chip which Intel demoed a while back? http://semiaccurate.com/2012/12/20/intel-explains-claremont-the-near-threshold-solar-pentium/

 

[simboss]

Did you miss the part in keynote where Intel said they did exactly this?

http://www.eetimes.com/document.asp?doc_id=1319447&

Saying it is not exactly doing it...

Last week, HVAC giant Daikin got one industrial reference board using a Quark chip and including WiFi and 3G support. Kevin Facinelli, executive vice president for operations at the company, dialed into the board from the IDF event here to show it is working.

You would hope Intel is able to get few design win with a new product that they design themselves and probably sell nowhere near what it cost to develop and maintain. they did exactly the same with CloverTrail...

Building a CPU with better efficiency or performance can be assessed as soon as it is launched with the usual set of benchmark/reviews.
The ease of use and support, each company has to try it first, decide if it is not only good enough to use, but so good that you want to jump to a new type of CPU.
Again, this takes a VERY long time, although Intel is not a pure newcomer to the market, it's clearly not the leader.

The other thing in these market is that multiple sources of components are important, what happens for your product if Intel decides that your CPU is not relevant anymore for its business?

But again, this only apply to real embedded devices, as Exophase said the market for lower power consumer devices (aka smartwatch and co) seems like a more likely target for two reasons:
- It's still withing reach of x86.
- There are no established player in this field yet as the field itself is new. So the support and tools are yet to be designed.

 

[Idontcare]

Is Quark the Claremont chip which Intel demoed a while back? http://semiaccurate.com/2012/12/20/intel-explains-claremont-the-near-threshold-solar-pentium/

Oh man! I think you nailed it, I really do.