Intel Broadwell Information (Intel Core M)

[Brainiac777]

Today’s preview and Intel’s associated presentation are going to be based around the forthcoming Intel Core M microprocessor, using the Broadwell configuration otherwise known at Broadwell-Y. The reason for this is a culmination of several factors, and in all honesty it’s probably driven as much by investor relations as it is consumer/enthusiast relations, as Intel would like to convince consumer and investor alike that they are on the right path to take control of the mobile/tablet market through superior products, superior technology, and superior manufacturing. Hence today’s preview will be focused on the part and the market Intel feels is the most competitive and most at risk for the next cycle: the mobile market that Core M will be competing in.

 

To that end Intel’s preview is very much a preview; we will see bits and pieces of Broadwell’s CPU architecture, GPU architecture, and packaging, along with information about Intel’s 14nm process. However this isn’t a full architecture preview or a full process breakdown. Both of those will have to wait for Intel’s usual forum of IDF.

 

Diving into matters then, Core M will be launch vehicle for Broadwell and will be released for the holiday period this year. In fact Intel is already in volume production of the Broadwell-Y CPU and production units are shipping to Intel’s customers (the OEMs) to begin production and stockpiling of finished devices for the holiday launch.

 

 

Many of these changes ultimately amount to boosting performance and reducing power consumption to a point where power and heat are where they need to be for mobile form factors, either through process efficiency improvements or through better power management and wider dynamic ranges – boosting where it matters and doing a better job of idling between tasks. However as Intel has discovered they not only need to be able to meet the TDP requirements of a tablet but they need to be able to meet the size requirements too. A particularly daunting task when the entire thickness of a device needs to be under 10mm, and the CPU thinner yet.

 

As a result, coupled with Core M’s performance improvements and power reductions is a strong emphasis on the size of the processor package itself and what Intel could do to reduce it. Intel calls this an outside-in system design, with various parts of Intel focusing on everything from the size of the logic board needed to hold the processor to the thickness of the processor die itself.

 

 

Most of these power savings are coming from the move to 14nm, but there are a few other influencing factors. Core M's minimum voltage has been reduced by 10 percent, and the second version of Intel's integrated Fully Integrated Voltage Regulator (FIVR) helps to save power when the chip is idling. Intel's approach to improving battery life with low-power Haswell chips was to make sure the CPU could get back to being idle as quickly as possible, so further reducing power usage when idle only makes sense.

 

In terms of functionality Broadwell’s GPU has been upgraded to support the latest and greatest graphics APIs, an important milestone for Intel as this means their iGPU is now at feature parity with iGPUs and dGPUs from AMD and NVIDIA. With support for Direct3D feature level 11_2 and Intel’s previous commitment to Direct3D 12, Intel no longer trails AMD and NVIDIA in base features; in fact with FL 11_2 support they’re even technically ahead of NVIDIA’s FL 11_0 Kepler and Maxwell architectures. FL 11_2 is a rather minor update in the long run, but support for it means that Intel now supports tiled resources and pre-compiled shader headers.

 

 

 

Broadwell chips will be Intel's first to use the 14nm process, and, of all the chips that will eventually ship with Broadwell, the low-power Core M family will be the first. These are the CPUs that will be in computers "for holiday and not at the last second of holiday," in the words of Intel CEO Brian Krzanich—we'll look at these chips specifically and then take a step back to look at architectural improvements that should benefit all Broadwell processors.

 

This isn't the first time Intel has talked about Core M—it was first mentioned at this year's Computex in June. As we reported then, Core M is a new name, but it's not necessarily a new class of product. It's just the newest name for Intel's Y-series parts, the slowest and most battery-friendly chips in Intel's CPU lineup to be based on its flagship architecture (they occupy the space between the U-series processors you'd find in a MacBook Air and most Ultrabooks and the Atom chips you can find in lower-end tablets, convertibles, and laptops). Previous Y-series chips used the same Core i3, i5, and i7 badges as other CPUs, but that will no longer be true for Broadwell.

The thing Intel stressed the most for Core M is that the company started with the kind of device it wanted the chip to fit into—a 10-inch fanless tablet 9mm or less in thickness—and worked backward to the technology. Every aspect of these systems, from the CPU and chipset die to the CPU package to the motherboard to the other system components, has apparently been optimized to save space and energy.

 

 

There are a lot of questions about how these power controls will actually work—will they noticeably affect performance? Will OEMs be restricted to certain parts if they want to take advantage of these features? Will the controls be exposed to users, or will it happen strictly in the background? These are all things we'll need to wait to find out about, whether it happens in a future briefing nearer to Core M's actual launch or from observation of actual Core M systems.

 

Now that we've heard a bit more about Core M, we can fill in further gaps in our knowledge about the rest of the Broadwell lineup. Intel says that Broadwell chips are five percent faster, on average, than Haswell chips running at the same clock speed. This IPC gain isn't mind-blowing, but it's not bad given that Broadwell is a refinement of Haswell and not an all-new architecture. 

This is just the first of several Broadwell-related information dumps we can expect over the next few months. We'd expect additional, more specific information about Core M models (along with specific clock speeds, performance numbers, and TDP numbers) at IDF early next month and more specific information about the rest of the Broadwell lineup whenever Intel is ready to give it. Once we have actual systems in our hands, we'll be able to put some of these performance and power consumption claims to the test.

 

 

Sources: 

http://www.anandtech.com/show/8355/intel-broadwell-architecture-preview

 

http://arstechnica.com/gadgets/2014/08/broadwell-is-coming-a-look-at-intels-low-power-core-m-and-its-14nm-process/

[Jobbe03]

 

Can't wait to switch back to intel

[qwertywarrior]

here is me waiting for my 11 inch yoga  laptop  with backlit keyboard , windows 9 and  coreM

[Real_PhillBert]

Ok, you have my attention.

[Yuval]

Well.. who wanted a 4690K anyway.

[atomicmonkey101]

my gosh...

[Bsmith]

as it also goes mobile as in laptop, then i think that desktop will be faaaaaaaaaaaarrr away for me.

[Deletive]

Well.. who wanted a 4690K anyway.

only a 5% ipc improvement over a haswell cpu... 5% is an overclock.  

[SuperfastJellyfish]

Can't wait for the Macbook Air with Core M.

 

TAKE ALL MY MONEY

[patrickjp93]

only a 5% ipc improvement over a haswell cpu... 5% is an overclock.  

That's on the mobile chips. A process shrink on a smaller chip doesn't change as much as it does on a desktop chip (much larger). You also have to remember the sole purpose of Haswell/Broadwell was power efficiency to do R&D for more mobile (phone) processors and improve things for servers where power REALLY matters.

 

Skylake will be a huge IPC improvement, not to mention unifying memory access between the CPU cores and GPU cores.

 

And look at the cache difference. Another .25MB of L3 per core makes a big difference.

[Deletive]

That's on the mobile chips. A process shrink on a smaller chip doesn't change as much as it does on a desktop chip (much larger). You also have to remember the sole purpose of Haswell/Broadwell was power efficiency to do R&D for more mobile (phone) processors and improve things for servers where power REALLY matters.

 

Skylake will be a huge IPC improvement, not to mention unifying memory access between the CPU cores and GPU cores.

 

And look at the cache difference. Another .25MB of L3 per core makes a big difference.

I honestly never thought a 5% ipc improvement was bad, just not worth it if you have haswell already (desktop only) for mobile this will hopefully really good  

[Trik'Stari]

TL:DR.

 

How's it looking against 4690k?

[patrickjp93]

I honestly never thought a 5% ipc improvement was bad, just not worth it if you have haswell already (desktop only) for mobile this will hopefully really good

Oh yeah, changing generations just on die-shrink alone has always been stupid. At minimum I'd say go Haswell-Skylake/Broadwell-Cannonlake, but to me waiting less than 5 generations to switch is too little. It takes a year or two for new instructions to be incorporated into OS and game updates. And now that clock rates are pretty much guaranteed to not increase beyond 4.5 GHz, I'd say it's even dumber now, especially if you have a 6-core chip. The only reason to really change would be if you were on the edge of scientific computing research and needed Broadwell's SIMD 512 (one of only 3 confirmed instruction additions between Haswell and Broadwell).

[Deletive]

Oh yeah, changing generations just on die-shrink alone has always been stupid. At minimum I'd say go Haswell-Skylake/Broadwell-Cannonlake, but to me waiting less than 5 generations to switch is too little. It takes a year or two for new instructions to be incorporated into OS and game updates. And now that clock rates are pretty much guaranteed to not increase beyond 4.5 GHz, I'd say it's even dumber now, especially if you have a 6-core chip. The only reason to really change would be if you were on the edge of scientific computing research and needed Broadwell's SIMD 512 (one of only 3 confirmed instruction additions between Haswell and Broadwell).

Yep and I think this will be the way it will be on the consumer line until the end of the decade (considering intel is pretty much going all the way up to 5/3nm) 

[coder-guy22296]

Can't wait for the Macbook Air with Core M.

 

TAKE ALL MY MONEY

not a bad concept, but imagine one of these in a asus rog G750 , at full load, the fan would barely have to be moving and pretty dang silent, assuming gpu is not under load

[connorpiper]

14nm! 

 

VAIO! Come back to Sony!

[patrickjp93]

14nm! 

 

VAIO! Come back to Sony!

Yawn, wake me up when Cannonlake (10nm) gets here.

[LuckyCharm]

That's on the mobile chips. A process shrink on a smaller chip doesn't change as much as it does on a desktop chip (much larger). You also have to remember the sole purpose of Haswell/Broadwell was power efficiency to do R&D for more mobile (phone) processors and improve things for servers where power REALLY matters.

 

Skylake will be a huge IPC improvement, not to mention unifying memory access between the CPU cores and GPU cores.

 

And look at the cache difference. Another .25MB of L3 per core makes a big difference.

I'm sorry are you an Intel engineer? Where are you getting this information of Skylake being a huge ipc improvement? Before Haswell launched people talked about how it was going to be a huge improvement, it turned out Haswell was only 5-15% better in ipc. Before Haswell refresh launched people talked about how you could do 5ghz on air, well I haven't seen that yet. And by the way mobile chips have pretty similar die sizes to desktop chips, so a shrink will effect them similarly.

[patrickjp93]

I'm sorry are you an Intel engineer? Where are you getting this information of Skylake being a huge ipc improvement? Before Haswell launched people talked about how it was going to be a huge improvement, it turned out Haswell was only 5-15% better in ipc. Before Haswell refresh launched people talked about how you could do 5ghz on air, well I haven't seen that yet. And by the way mobile chips have pretty similar die sizes to desktop chips, so a shrink will effect them similarly.

 

Lucky, it's been documented Haswell is not a good overclocking architecture due to the integrated voltage regulator and the very tight power control within the circuits themselves (hence the huge power efficiency gain). Broadwell is only a die shrink of essentially the same architecture. Skylake is on a new socket (BIG HINT) and brings with it technology Intel felt was worth copying. You'd have to be stupid to think it won't be a great improvement in clock for clock throughput. It's Intel's secondary cycle: performance, efficiency, performance, efficiency. Sandybridge/Ivybridge was performance. Haswell/Broadwell is efficiency. Skylake/Cannonlake will be performance.

 

You don't have to be an Intel engineer to understand the basics of what's going on. Common sense and knowing some basic facts in addition to having sources of reference allows me to say this confidently.

 

Also, the die on a mobile chip is half as thick as a desktop chip, and usually they are a bit more narrow, regardless of the socket or IHS size. Please put ALL the facts on the table instead of making baseless blanket statements.

[The Crazed Child]

I'm looking forward to Iris Pro on the desktop chips. Not that I'll have a use for it anyway, it just seems awesome.

[patrickjp93]

I'm looking forward to Iris Pro on the desktop chips. Not that I'll have a use for it anyway, it just seems awesome.

I mean Hell, Apple managed to force Intel to accelerate and accentuate their iGPU development for the purpose of Retina display, and now so many high-res tablets depend on Iris Pro instead of an AMD APU. People can say Intel's graphics suck all they want. They haven't had the real thing just yet.

[flibberdipper]

[The Crazed Child]

I mean Hell, Apple managed to force Intel to accelerate and accentuate their iGPU development for the purpose of Retina display, and now so many high-res tablets depend on Iris Pro instead of an AMD APU. People can say Intel's graphics suck all they want. They haven't had the real thing just yet.

I would like to see Intel put 256MB of eDRAM on the Iris Pro die and then put it on the package of an entry-level Pentium. Those would sell like pancakes/hotcakes or whatever they're called.

[patrickjp93]

I would like to see Intel put 256MB of eDRAM on the Iris Pro die and then put it on the package of an entry-level Pentium. Those would sell like pancakes/hotcakes or whatever they're called.

What would be the purpose of such a chip though?

 

Edit: low-power academic setting computers with vPro tech attached? It's a large-scale environment frankly perfect for that...hmm... Maybe I should try to have Intel hire me as a market analyst/consultant.

[Deletive]

TL:DR.

 

How's it looking against 4690k?

 On a scale of 1 to 10? 

It's between 'shouldn't worry about it' and 'don't bother'.