Analysis: Haswell "Refresh" Is Not Broadwell

[Hulk]

Broadwell is coming in 2014, just not for desktops. Desktops go straight to Skylake in 2015.

Okay, got it. Kind of like Nehalem (quads only) to Westmere (duals only). Somebody else around here mentioned that and it makes sense now.

I wonder why though? I mean if they are producing 14nm Broadwell I would assume the yields would permit some higher leakage parts for the desktop where power isn't as big a concern. Kind of makes me think Broadwell will show basically no performance improvement over Haswell, just GPU and power improvements. So Broadwell looks to be a straight up shrink of Haswell with an improved GPU right?

 

[meloz]

What do you guys think? Am I full of it or does what I wrote make sense? Thoughts?

Broadwell is coming in 2014, just not for desktops. Desktops go straight to Skylake in 2015.

You have not addressed one issue: why are they releasing Z97 and H97 chipsets if Broadwell is not being released? Who will buy these chipsets without Broadwell? And the primary market for both these series of chipsets (Hx7 and Zx7) has traditionally been desktop motherboards not notebooks and mobiles.

So either the roadmap is old, fake or intentionally lying to spread disinformation. Or desktop Broadwell is coming. I just do not like how people focus on one part of roadmap which best helps builds their fantasies, while completely ignoring other. Many tech blogs and 'news' sites have made a fool of themselves these past few days.

I also do not believe in the fairy tales that we will all be using large socket Haswell-E, or that the two sockets will be converging. The large socket is a premium product, a decent socket 2011 CPU + MB costs close to $1000. And in future will likely cost even more because of lack of rivals in this arena. Whereas in the PC market most PCs -entire PCs- are sold for less than $1000.

So unless we are all getting a cash bonus from Intel to help facilitate pruchase of large socket system, the mainstream and enthusiasts will stick with small socket.

Socket 2011 and its future successors are for the 'extreme' enthusiasts and workstation guys. And it makes no sense for Intel to try and narrow the gap between the two sockets (that is the most nonsensical thing I have read in recent days). They segmented these two in the first place because the power (TDP), compute power and economics of the two segments were so different. They serve two entirely different segments.

The large socket also forms the basis (or at least shares most of the engineering work) with Xeons, so unless Intel can convince server customers to buy CPUs in which half or more of the die is lying unused (iGPU).........honestly I feel stupid trying to explain this.

Maybe they will do both. Refresh Haswell (for Pentiums and Celerons) and also release Broadwell. The reason they have not mentioned Broadwell anywhere in the roadmap might be that its launch date is still floating, totally dependent on how much orders Intel get for Airmont.

If they price Airmont right and it gets big ODM orders, Intel might not have the capacity at 14nm to manufacture Broadwell in sufficient quantities alongside Airmont. And if Airmont is a failure or a 'modest success' (due to pricing or whatever), Broadwell might be bought to market sooner.

 

[Arachnotronic]

Okay, got it. Kind of like Nehalem (quads only) to Westmere (duals only). Somebody else around here mentioned that and it makes sense now.

I wonder why though? I mean if they are producing 14nm Broadwell I would assume the yields would permit some higher leakage parts for the desktop where power isn't as big a concern. Kind of makes me think Broadwell will show basically no performance improvement over Haswell, just GPU and power improvements. So Broadwell looks to be a straight up shrink of Haswell with an improved GPU right?

A buddy of mine who worked on Broadwell strongly hints that there will be some performance enhancements/cleanup, Ivy Bridge style.

 

[ViRGE]

You have not addressed one issue: why are they releasing Z97 and H97 chipsets if Broadwell is not being released? Who will buy these chipsets without Broadwell? And the primary market for both these series of chipsets (Hx7 and Zx7) has traditionally been desktop motherboards not notebooks and mobiles.

I don't know whether any of this roadmap is true or not. But Intel uses the same chipsets on the desktop and mobile. So if Broadwell is coming to mobile in 2014, then why would you not release the new chipset on the desktop at the same time?

 

[Pandamonia]

The problem with 22nm and 14nm is that they arent made for performance. They are low power lithography.

I once saw a chart which showed how badly heat scaled with voltage and performance scaled with mhz. 32nm was the last high performance process intel made.

I have a 45nm i7 950 and the increase to 32nm was massive. IPC, clock speed and temps all saw a huge benefit.

22nm was intels first move to low power design not just in CPU but also process design. 14nm is the follow up to that same path that intel started with 22nm. If Intel doesnt drop a desktop 14nm chip there is a VERY good reason. The reason will be that 14nm will clock worse than 22nm and IPC wont improve which will mean that intel will be selling CPU's which perform worse than the year before. The K series CPU's will be worthless. No one will buy an i7 5770 if it performs worse than the 4770.

The only reason Haswell is worth while is because there is IPC gains. So Broadwell will offer no benefit to desktop users at all without IPC because 14nm is going to overclock like Conroe. Crap!

 

[SlickR12345]

I was expecting to see 6 to 8 core models from Intel this season, but what they did was downgrade, instead of upgrade.

I mean 4 cores have been the norm for at least 2 years now, even low end are now getting 4 cores, so I would have thought they would at least try and make 6 cores the new norm with great single threaded performance that they have over AMD.

But then again AMD has been really bad with the performance and they need to stop up their game so that Intel has pressure to deliver.

 

[jpiniero]

I don't know whether any of this roadmap is true or not. But Intel uses the same chipsets on the desktop and mobile. So if Broadwell is coming to mobile in 2014, then why would you not release the new chipset on the desktop at the same time?

Because Broadwell has the chipset on package for all models (on Haswell only the ULV has this); it was deemed not worth the effort to make it also work with the external chipset which also is compatible with Haswell.

I'm guessing Skylake will bring the chipset ondie.

 

[ShintaiDK]

Because Broadwell has the chipset on package for all models (on Haswell only the ULV has this); it was deemed not worth the effort to make it also work with the external chipset which also is compatible with Haswell.

I'm guessing Skylake will bring the chipset ondie.

H97 and Z97 are not laptop chipsets.

 

[meloz]

Just because Intel have not explicitly stated that Broadwell will be released in 2014 does not mean that they have decided to not release it until 2015. Why is this basic concept so hard to understand? They can always release it in latter part of 2014, which was the original expectation / schedule anyway.

Mayhaps Intel are growing extra paranoid (not about competition, but analysts and shareholders) and do not want to announce Broadwell until later. Saving something positive for later. Who knows.

Then again all these roadmap slides might have been created by some wicked troll who must know us enthusiasts all too well. If so, hope he is enjoying his success.

EDIT:

One more thing. Whether you believe the slides to be true or false, in neither case can they be used to arrive to the logical conclusion that Broadwell will not be released in 2014.

In case you think the slides to be true, obviously the chipset schedule tells us that Broadwell is coming. In case you think the slides are false, there is insufficient data to draw any conclusion.

And this whole desktop-laptop does not matter to the topic at hand. The takeaway is if there are new chipsets there is a new processor. And while most sales -particularly OEM- will go to Intel BGA, we know of at least two chipsets that will make it to motherboard manufacturers. And that's good enough for me.

 

[ViRGE]

Because Broadwell has the chipset on package for all models (on Haswell only the ULV has this); it was deemed not worth the effort to make it also work with the external chipset which also is compatible with Haswell.

I'm guessing Skylake will bring the chipset ondie.

Interesting. I thought it was Skylate that brought the chipset on-chip for the entire lineup, but in retrospect I may have that confused. Regardless, if Intel is rolling out new chipsets and new features in those chipsets, one would expect they roll it out to desktops and laptops regardless of whether desktops get a new CPU or not. Especially if there's a die shrink involved, as that would mean they're going to be ramping down production of that node.

H97 and Z97 are not laptop chipsets.

By model number, no. But traditionally Intel uses the same silicon for both, even if they give them different names (and disable different features).

 

[galego]

The problem with 22nm and 14nm is that they arent made for performance. They are low power lithography.

I once saw a chart which showed how badly heat scaled with voltage and performance scaled with mhz. 32nm was the last high performance process intel made.

I have a 45nm i7 950 and the increase to 32nm was massive. IPC, clock speed and temps all saw a huge benefit.

22nm was intels first move to low power design not just in CPU but also process design. 14nm is the follow up to that same path that intel started with 22nm. If Intel doesnt drop a desktop 14nm chip there is a VERY good reason. The reason will be that 14nm will clock worse than 22nm and IPC wont improve which will mean that intel will be selling CPU's which perform worse than the year before. The K series CPU's will be worthless. No one will buy an i7 5770 if it performs worse than the 4770.

The only reason Haswell is worth while is because there is IPC gains. So Broadwell will offer no benefit to desktop users at all without IPC because 14nm is going to overclock like Conroe. Crap!

Good analysis, coincides with this:

When Intel chose to integrate a voltage regulator into Haswell, it made a decision that would improve the chip's low power capability, but increase thermal density. That was always going to hurt high-end scaling. In other words, Intel made a decision that prioritized mobile and low power consumption over high desktop performance. As the benefits of scaling to new process nodes fall, Intel is repeatedly choosing to focus what benefits it can grasp on the low power consumption side of the equation.

http://hothardware.com/News/Intel-B...tops-Until-2015-Reality-Likely-More-Complex-/

Just because Intel have not explicitly stated that Broadwell will be released in 2014 does not mean that they have decided to not release it until 2015. Why is this basic concept so hard to understand? They can always release it in latter part of 2014, which was the original expectation / schedule anyway.

Mayhaps Intel are growing extra paranoid (not about competition, but analysts and shareholders) and do not want to announce Broadwell until later. Saving something positive for later. Who knows.

Then again all these roadmap slides might have been created by some wicked troll who must know us enthusiasts all too well. If so, hope he is enjoying his success.

EDIT:

One more thing. Whether you believe the slides to be true or false, in neither case can they be used to arrive to the logical conclusion that Broadwell will not be released in 2014.

In case you think the slides to be true, obviously the chipset schedule tells us that Broadwell is coming. In case you think the slides are false, there is insufficient data to draw any conclusion.

And this whole desktop-laptop does not matter to the topic at hand. The takeaway is if there are new chipsets there is a new processor. And while most sales -particularly OEM- will go to Intel BGA, we know of at least two chipsets that will make it to motherboard manufacturers. And that's good enough for me.

Part of the leaked roadmap seems to be confirmed by "one of our trusted sources in the motherboard industry"

http://techreport.com/news/24915/le...haswell-refresh-in-2014-no-broadwell-in-sight

 

[Pilum]

If they price Airmont right and it gets big ODM orders, Intel might not have the capacity at 14nm to manufacture Broadwell in sufficient quantities alongside Airmont. And if Airmont is a failure or a 'modest success' (due to pricing or whatever), Broadwell might be bought to market sooner.

I agree that capacity is probably the key issue here. But it's not only about Airmont, but also Knights Landing, the 14nm Xeon Phi. If they want to launch that mid-2014, it'll gobble up lots of 14nm capacity - these dies are huge. But if Intel ramps 14nm capacity aggressively, we still may see 14nm Broadwell DT in late 2014. But as you said, that may depend on the market success of the various 14nm products.

 

[OBLAMA2009]

call me cynical but i wouldnt be surprised if the "refresh" is a downgrade in speed, like maybe an annoucement of atom based i7's. theyre running out of ideas in cpu design so theyre trying to slow things down

 

[meloz]

Part of the leaked roadmap seems to be confirmed by "one of our trusted sources in the motherboard industry"

http://techreport.com/news/24915/le...haswell-refresh-in-2014-no-broadwell-in-sight

When was the last time Intel released a new series of chipsets for an old processor? Not a rhetorical question, I am just curious.

If Intel are being greedy / cheap and just pushing back Broadwell to 2015 (or cancelling it altogether), why would they bother with the expensive and time consuming task of verifying all new chipsets for existing CPUs? Is SATA Express that important?

Just continue to sell Haswell for 24 months, it is not as if their competition is offering SATA Express or the small-socket market is dying for it.

 

[NTMBK]

When was the last time Intel released a new series of chipsets for an old processor? Not a rhetorical question, I am just curious.

It was fairly common back in the P4 days, if I remember correctly. And as has been pointed out above it is the same silicon as the mobile chipset, so it makes sense to release it- may as well let the desktop share some new features, even if they stay on 22nm.

 

[SammichPG]

On the bright side AMD will have time to catch up and 64bit ARM cores should start eating the low end server market as well.

Interesting times ahead.

 

[cytg111]

Interesting times ahead.

.. slow times ahead

 

[Arachnotronic]

The problem with 22nm and 14nm is that they arent made for performance. They are low power lithography.

This is complete bullshit. Intel has multiple flavors of its manufacturing process. Xeons, high end Core, and Xeon Phi are built on the high performance process (P1270) and the low power parts are built on the low power SoC process (P1271).

 

[Pandamonia]

This is complete bullshit. Intel has multiple flavors of its manufacturing process. Xeons, high end Core, and Xeon Phi are built on the high performance process (P1270) and the low power parts are built on the low power SoC process (P1271).

28nm from TMSC and GloFo are offered in 28nm HP and LP (and other) variations since they make GPU's in HP and SoC's for smartphones on high K low power process nodes. TSMC clearly nailed 28nm for all its customers.

Intel has been about low power useage and has only low power low leakage 22nm 3D transistors to reduce leakage. As you push higher voltage the leakage increases massively and thats why clock speeds are worse than 32nm.

Intel has been adding power saving features to desktop CPU's including VRM's and a ton of other features that are totally useless on desktop. This is because desktop and mobile chips are just different bins of the same chip. Intels 22nm is focused on low power and not high clock speeds. Clearly evidenced by the poor show from Ivy and Haswell in the desktop arena.

The difference between the two process types at TSMC are the type of SOI substrate used and the different methods used for producing SHP and SLP.

Unless you can show me intel using a method of 22nm lithography to produce a Super High Performance 22nm chips then im willing to change my view. Fact is that i have yet to see intel do anything other than bin their chips for mobile, server and desktop.

 

[crashtech]

Intel has zero competition in the high end, why would we expect anything else? I have no particular affinity for mobile devices, but even I can see why they are doing what they are doing. It makes perfect sense. Luckily we will have server derivatives to fall back on for our desktops for the foreseeable future.

 

[mrmt]

The difference between the two process types at TSMC are the type of SOI substrate used and the different methods used for producing SHP and SLP.

TSMC and SOI? Did I miss something?

 

[Torn Mind]

Perhaps they'll "refresh" the IGP on Haswell. lol.

 

[SammichPG]

.. slow times ahead

Software might cath up hardware after years of "hardware abundance" that allowed software vendors to put something together knowing that each new cpu would be substantially faster.

I'm hoping for:
- better multicore programming (there are huge performance increases to be had in many programs)
- better instruction usage (afaik Skyrim used x87 code instead of sse!)
- HSA/opencl/whatever makes possible using gpus in a more transparent way (an example would be handbrake using gpu encoding without any quality penalty or limitation in encoding settings)

I guess that even without any hardware advance we would see great performance gains just with better software for some years.

 

[Pandamonia]

TSMC and SOI? Did I miss something?

Sorry that is meant to say GloFo

 

[Blandge]

Unless you can show me intel using a method of 22nm lithography to produce a Super High Performance 22nm chips then im willing to change my view. Fact is that i have yet to see intel do anything other than bin their chips for mobile, server and desktop.

Nobody cares about what your views are. We care about what's true. You can continue to believe false things, but it is a fact that Intel uses is a different process for high performance and low power designs. Unless you can show me a CPU that is faster than Haswell and Ivy Bridge, then I think they probably qualify as "Super High Performance."