Intel and the big.LITTLE concept

[Lonyo]

I just noticed the new pipeline article about BIG.Little on AT http://www.anandtech.com/show/6420/arms-cortex-a57-and-cortex-a53-the-first-64bit-armv8-cpu-cores which builds on the big.LITTLE concept.

Would Intel be able to benefit form this? Obviously with Haswell, they are reducing power requirements, but will there still be (some) power advantage for Atom cores against Haswell cores?

A single Atom core plus 2~4 Haswell cores on a single 22nm die would potentially allow low power use when using the single Atom core, with the Haswell cores being able to do the heavy lifting.
Obviously the software support would need to be there, and with Windows 8 it might not be yet, but going forwards, is it conceptually viable that Intel would consider a similar big.LITTLE concept using Atom cores, or are they more likely to stick with making their big x86 cores simply more power efficient in mobile use situations?

 

[ShintaiDK]

It basicly shows that ARM knows its in trouble. Also the concept is very impractical. It adds to diesize and complexity to solve something that ARM, in this case, cant fix (performance/watt of performance parts.).

Hence why I cant see Intel doing it either.

If you look on the powerbudget changes from 28nm to 14nm. The ARM numbers are anything but impressive.

 

[nehalem256]

I think the Low power Haswell still uses something like 10w of power which is probably impractical for phones/tablets where this concept would be most useful.

Also, going by current performance isn't the atom more a competitor for the BIGarm anyway?

 

[ShintaiDK]

I think the Low power Haswell still uses something like 10w of power which is probably impractical for phones/tablets where this concept would be most useful.

Also, going by current performance isn't the atom more a competitor for the BIGarm anyway?

Yep, even a single Haswell 8W part would be miles ahead.

The competition is 14nm Atoms. And looking ont hose numbers, the 14nm Atoms just gonna molest those ARM cores.

Also I wonder if its a typo on the slide, because TSMC dropped 14nm in favour of 16nm.

But again, both are still 2 years into the future.

 

[Olikan]

small core with big brothers....looks like ARM copying from nvidia
reestruture of data management...ARM copying from apple swift

...this means that ARM can use the IP that the other companys developed?

 

[ShintaiDK]

small core with big brothers....looks like ARM copying from nvidia
reestruture of data management...ARM copying from apple swift

...this means that ARM can use the IP that the other companys developed?

Its not an IP to mix cores so to say.

 

[Exophase]

The whole point of big.LITTLE is to migrate code seamlessly between the big and little cores. You can't do this if the processors have different architectures. Even in ARM's case they had to bring the little core all the way up to every last extension the big core offers.

Intel couldn't even do this with Ivy Bridge and Atom as it currently stands.. For next generation it'll either need Silvermont Atoms with AVX2 support (currently they lack even SSE4) or castrated Haswell without AVX2... Intel has been releasing lower end CPUs without AVX since SB, but I hope they change their mind with Haswell. At any rate they've never been about relegating the best power consumption figures solely to their cheapest low end processors.

Right now ARM's little cores are so tiny that it barely hits the die budget to include them. Intel hasn't come anywhere close in making such a small modern x86 core. At these levels the architectural requirements become much more significant.

And ARM's strategy is totally different from nVidia's. They don't have a "little" core at all, their fifth core is the same size as all the others and is strictly muxed out with one of the other four. The advantage is entirely in manufacturing while ARM's advantage is in design (and manufacturing as well if the SoC vendor decides to pursue this)

 

[podspi]

I just noticed the new pipeline article about BIG.Little on AT http://www.anandtech.com/show/6420/arms-cortex-a57-and-cortex-a53-the-first-64bit-armv8-cpu-cores which builds on the big.LITTLE concept.

Would Intel be able to benefit form this? Obviously with Haswell, they are reducing power requirements, but will there still be (some) power advantage for Atom cores against Haswell cores?

A single Atom core plus 2~4 Haswell cores on a single 22nm die would potentially allow low power use when using the single Atom core, with the Haswell cores being able to do the heavy lifting.
Obviously the software support would need to be there, and with Windows 8 it might not be yet, but going forwards, is it conceptually viable that Intel would consider a similar big.LITTLE concept using Atom cores, or are they more likely to stick with making their big x86 cores simply more power efficient in mobile use situations?

Actually, I think probably a better configuration would be two Haswell + HT cores, and four smaller Atom cores. Similar to how it is looking like Haswell's IGP is going to go very wide, but slow.

It basicly shows that ARM knows its in trouble. Also the concept is very impractical. It adds to diesize and complexity to solve something that ARM, in this case, cant fix (performance/watt of performance parts.).

Hence why I cant see Intel doing it either.

If you look on the powerbudget changes from 28nm to 14nm. The ARM numbers are anything but impressive.

Actually, Intel is doing something similar (if the rumors are true) with its IGP. More shaders at a lower clockspeed is a very similar strategy to this, except in this case you have entire cores dedicated to low power.

Unless you are claiming Intel has somehow managed to design a CPU that can cover over an order of magnitude of TDP. Even if Haswell is designed for 8W ~ 80W, 8W is a lot of power for a tablet or smartphone. While I doubt we'll see such a design in desktops or even laptops, for tablets (which typically for Intel carry a much larger margin anyway) I wouldn't be surprised. Intel has already (ok, not already, but supposedly) shown us it is willing to eat die space for power savings.

Of course, if you are correct and Atom embarrasses ARM's offerings (I won't agree or disagree with you there, we're just too far away from release at this point), then Intel wont' have much reason to go this route.

 

[ShintaiDK]

I just cant imagine ARM fighting 14nm Atoms.

 

[Olikan]

I just cant imagine ARM fighting 14nm Atoms.

what it will bring to table? all i know is OoO x86

 

[ShintaiDK]

what it will bring to table? all i know is OoO x86

It basicly get a 5 year update or how long it is now to its uarch. OoO is one part. Its a major overhaul.

Airmont, will offer a 10x gain in processing power and graphics performance over today’s Atom, and 20x that of the Atom N270. To put that in perspective, the Atom N270 launched in 2008 managed 3.856MIPS when running at 1.6GHz. The Airmont chip will achieve 79,000 MIPS. That’s about the same performance as an AMD Phenom X6 6-core processor today.

It has to be said that these numbers comes from an Intel tablet roadmap. So might be abit different for smartphones.

The benchmarks used are Specint2000_rate and GFX just said...GFX. So also take it with a grain of salt. But still, we talk huge improvements.

 

[Olikan]

The benchmarks used are Specint2000_rate and GFX just said...GFX. So also take it with a grain of salt. But still, we talk huge improvements.

it's a good performance bump... but, nothing surreal, really...

if the jaguar cores deliver the promised, they will be around ~50% slower in ST to a 1055T , and that is in 28nm....
add 2 cores, shrink to 14nm, and profit!

 

[ShintaiDK]

it's a good performance bump... but, nothing surreal, really...

if the jaguar cores deliver the promised, they will be around ~50% slower in ST to a 1055T , and that is in 28nm....
add 2 cores, shrink to 14nm, and profit!

You dont make any sense. ST as in SingleThread?

If thats the case, then you mix singlethreaded specint with multithreaded specint_rate. So your jaguar core setup will come in short, very short.

 

[Olikan]

You dont make any sense. ST as in SingleThread?

If thats the case, then you mix singlethreaded specint with multithreaded specint_rate. So your jaguar core setup will come in short, very short.

yeah, single thread...

 

[cbn]

I just noticed the new pipeline article about BIG.Little on AT http://www.anandtech.com/show/6420/arms-cortex-a57-and-cortex-a53-the-first-64bit-armv8-cpu-cores which builds on the big.LITTLE concept.

Would Intel be able to benefit form this? Obviously with Haswell, they are reducing power requirements, but will there still be (some) power advantage for Atom cores against Haswell cores?

A single Atom core plus 2~4 Haswell cores on a single 22nm die would potentially allow low power use when using the single Atom core, with the Haswell cores being able to do the heavy lifting.
Obviously the software support would need to be there, and with Windows 8 it might not be yet, but going forwards, is it conceptually viable that Intel would consider a similar big.LITTLE concept using Atom cores, or are they more likely to stick with making their big x86 cores simply more power efficient in mobile use situations?

Great question you are asking.

http://forums.anandtech.com/showthread.php?p=32634336#post32634336

Remember all the C and P states? The multiple frequency and voltage steps between? P states are for active and C states are for idle. The lower the power state, the longer it takes to wake up.

By having enough C and P states they can avoid adding another small core CPU just to save power. Race to idle to save power won't work if the CPU core is many times slower, like putting in the Atom CPU. Any potential power that might have been saved might be wasted because it takes longer to execute the task.

 

[jpiniero]

There definitely is room for improvement when it comes to idle power consumption. Looking at Anand's Trinity review, the desktop Trinity and Core systems idled in the 30ish range. A6 and the Atom? Around 1ish.

Is 1ish idle power consumption unrealistic for a desktop processor? Probably, but I think that's a goal of Intel's and why they are not interested in anything like big little. Maybe in a couple years when everything's an SoC.

Actually, Intel is doing something similar (if the rumors are true) with its IGP. More shaders at a lower clockspeed is a very similar strategy to this, except in this case you have entire cores dedicated to low power.

Big little isn't really comparable. BL is far more complicated and requires that the OS/Software (?) supports it.

 

[podspi]

It basicly get a 5 year update or how long it is now to its uarch. OoO is one part. Its a major overhaul.



It has to be said that these numbers comes from an Intel tablet roadmap. So might be abit different for smartphones.

The benchmarks used are Specint2000_rate and GFX just said...GFX. So also take it with a grain of salt. But still, we talk huge improvements.

Does that include the IGP? If not, that is ridiculously incredible. Given that the X6 isn't too far behind the 2600K/3700K in multithreaded applications, I have a hard time believing Atom will be packing that level of processing power.

Either that or Haswell is going to be really, really awesome.

Edit: Also, source? Not disbelieving you, I just haven't seen them posted anywhere I frequent on the 'net...

 

[meloz]

I just cant imagine ARM fighting 14nm Atoms.

Neither can I, for that matter.

ARM are presently in the same place AMD were a few years ago when they had the edge over Intel. AMD did not invest aggressively, and now we can all see the consequences.

ARM have a short window in which to use their profits from current boom and invest heavily in 50-150 watt 64-bit CPUs, because their 'monopoly' in the 1-10 watt CPUs / SoCs arena is only going to last another 3-4 years, if not less.

Intel made things very easy for ARM by sleeping on Atom after the initial success of netbooks, but as their roadmap shows the chipzilla has woken up and intend to be hyper-aggresive on new Atoms from 2013-14.

Only Intel's obsession with absurd profit margins might leave certain niches for ARM, otherwise ARM are going to be pushed back into the embedded space and out of the more profitable gadget market.

On their part ARM are attempting a few things with AArch64 and so on, I am just not certain it will be enough against the likes of Broadwell and Broadwell-E.

To me it looks like Intel will have an easier time of competing and dominating the 1-5 watt segment with its all new Atoms than ARM will have of competing with Intel in the 50-150 watt segment (desktops, servers, workstations) with unproven, first generation 64-bit products. ARM might not have the time to develop a second generation.

Intel have biggest margin on their Xeons, and ARM are attempting to compete there, without a CPU that matches the Xeons in outright performance. And they think they can compete 'indirectly' by densely packaging multiple CPUs.

Trouble is you can only cobble together so many weaker cores / CPUs -freedom fabric notwithstanding- in a 1U or 4U chassis. Or a 44U cabinet. At some point you simply need more powerful cores, and x86-64 Xeons have that market locked.

And it remains to be seen whether these upcoming larger, more powerful ARM CPUs can maintain the performance/watt ratio of their low powered siblings, forget matching any Xeons. There was a discussion at RealWorldTech long ago and Torvalds (and others) made some good points that are just as valid today when we talk of ARM.

With Haswell, Broadwell etc Intel will get even more competitive in performance/watt. At best ARM might force Intel to (temporarily) rationalize prices of certain Xeons.

 

[Arachnotronic]

Neither can I, for that matter.

ARM are presently in the same place AMD were a few years ago when they had the edge over Intel. AMD did not invest aggressively, and now we can all see the consequences.

ARM have a short window in which to use their profits from current boom and invest heavily in 50-150 watt 64-bit CPUs, because their 'monopoly' in the 1-10 watt CPUs / SoCs arena is only going to last another 3-4 years, if not less.

Intel made things very easy for ARM by sleeping on Atom after the initial success of netbooks, but as their roadmap shows the chipzilla has woken up and intend to be hyper-aggresive on new Atoms from 2013-14.

Only Intel's obsession with absurd profit margins might leave certain niches for ARM, otherwise ARM are going to be pushed back into the embedded space and out of the more profitable gadget market.

On their part ARM are attempting a few things with AArch64 and so on, I am just not certain it will be enough against the likes of Broadwell and Broadwell-E.

To me it looks like Intel will have an easier time of competing and dominating the 1-5 watt segment with its all new Atoms than ARM will have of competing with Intel in the 50-150 watt segment (desktops, servers, workstations) with unproven, first generation 64-bit products. ARM might not have the time to develop a second generation.

Intel have biggest margin on their Xeons, and ARM are attempting to compete there, without a CPU that matches the Xeons in outright performance. And they think they can compete 'indirectly' by densely packaging multiple CPUs.

Trouble is you can only cobble together so many weaker cores / CPUs -freedom fabric notwithstanding- in a 1U or 4U chassis. Or a 44U cabinet. At some point you simply need more powerful cores, and x86-64 Xeons have that market locked.

And it remains to be seen whether these upcoming larger, more powerful ARM CPUs can maintain the performance/watt ratio of their low powered siblings, forget matching any Xeons. There was a discussion at RealWorldTech long ago and Torvalds (and others) made some good points that are just as valid today when we talk of ARM.

With Haswell, Broadwell etc Intel will get even more competitive in performance/watt. At best ARM might force Intel to (temporarily) rationalize prices of certain Xeons.

The "profits" from ARM's boom are about $100M/quarter. Intel takes in ~$3B per quarter in SHEER PROFITS and pays ~$1B/qtr in dividends alone - more than ARM's entire revenue for a year.. If Intel decides to heavily invest to keep its lead in the high end (and dominate the low end) - which it has made EVERY indication that it has - ARM and its band of scattered, profitless (STM, AMD, etc.) licensees have no real chance. Qualcomm, maybe, but they make their big margins on the baseband chips more so than the apps processor. But once again, Intel's profits dwarf all of the individual ARM licensees' semiconductor divisions.

I don't really get why people think ARM is magical. *sigh*

 

[ShintaiDK]

I don't really get why people think ARM is magical. *sigh*

Most likely emotional driven.

Its not Intel.
Its not x86.

 

[sm625]

Intel can do it any way they want. They dont even need to use separate cores. They can easily refine power gating to shut down any integer cluster if the scheduler says it doesnt need it. They could power down sections of each cache. It seems they have decided to move in that direction instead. One big smart core. (Or smarter anyway.) Hell, how big is intel's power management cluster? Isnt it the size of some ARM SoCs?

 

[meloz]

The "profits" from ARM's boom are about $100M/quarter. Intel takes in ~$3B per quarter in SHEER PROFITS and pays ~$1B/qtr in dividends alone - more than ARM's entire revenue for a year.. If Intel decides to heavily invest to keep its lead in the high end (and dominate the low end) - which it has made EVERY indication that it has - ARM and its band of scattered, profitless (STM, AMD, etc.) licensees have no real chance. Qualcomm, maybe, but they make their big margins on the baseband chips more so than the apps processor. But once again, Intel's profits dwarf all of the individual ARM licensees' semiconductor divisions.

I think the ARM is presently able to punch above their weight because they have got so many 'partners' -including Apple- all of whom have an interest in driving 64-bit ARM processors forward.

Remains to be seen whether they actually get collabrative on R&D side, though, and how deep the partnership runs.

 

[ShintaiDK]

I think the ARM is presently able to punch above their weight because they have got so many 'partners' -including Apple- all of whom have an interest in driving 64-bit ARM processors forward.

Remains to be seen whether they actually get collabrative on R&D side, though, and how deep the partnership runs.

Majority of, if not all, ARM partners have no real interest in ARM. They will happily change if something better comes along. And certainly Apple, they ditched PPC for the same reasons.

 

[Arachnotronic]

I think the ARM is presently able to punch above their weight because they have got so many 'partners' -including Apple- all of whom have an interest in driving 64-bit ARM processors forward.

Remains to be seen whether they actually get collabrative on R&D side, though, and how deep the partnership runs.

I doubt it. Why? ARM's licensees are competitors to one another. Nvidia's Tegra division certainly isn't going to be helping Qualcomm's Snapdragon or Apple's Ax divisions.

 

[Idontcare]

I doubt it. Why? ARM's licensees are competitors to one another. Nvidia's Tegra division certainly isn't going to be helping Qualcomm's Snapdragon or Apple's Ax divisions.

The aggregate R&D though does play into the creation of entire markets for which companies base their own R&D investments because of the projected TAM.

Design tools for example, if the TAM for Cadence was 50% of what it is today then Cadence would be investing much less into 20nm/14nm optimized design tools.

Sure it isn't one-for-one as if all these guys are marching in lockstep, there is a fair amount of duplicity in the efforts done with the same R&D dollars.

But they all feed together into raising the tide for those businesses that seek to earn a living as suppliers and vendors to the R&D efforts. (look at what ASML did for example, pooling R&D monies from Samsung/Intel/TSMC towards the 450mm effort)