Haswell: News, Rumors & Reviews

[archibael]

And, as we learned from the BD fiasco, lets take the estimates comming from Intel in regards to app compatability with a grain of salt for now, shall we? I'm always skeptical of the marketing department...

As well you should. Speculation is fun (it's what this thread is all about), but you should always demand proof.

 

[mayankleoboy1]

also, the medfiels based devices are said to be quite more expensive.
who wants a expensive phone that has less than 1% of the apps of the cheaper competitior?

 

[Ubrales]

As well you should. Speculation is fun (it's what this thread is all about), but you should always demand proof.

Known as "Trust, but verify!"

 

[Blandge]

A lot is made of the idea that ARM is "good enough" in the performance space, but architecturally more well suited to lower power, giving it an inherent advantage in tablets/phones. This may be a fair assessment. However, is there a point at which IA is "good enough" in the low power space, and still architecturally more well-suited to performance and/or software compatibility? This is, in my opinion, a question that will define Intel's forthcoming execution in the mobile market.

This is a good point, but I think Intel's brand carries a lot of weight as well. If the performance is "good enough" and the Android app compatibility is also "good enough" than I think that Intel will sell smartphones decently based on brand name recognition alone.

When we look at the tablet segment with Win8 we can see a clear advantage to Intel in terms of app and OS compatibility. I think the biggest opponent Intel will have here is cost, but I would hazard to guess many Windows users would gladly pay $600 for a tablet with the capabibility of a full Windows OS behind it. It's hard to make a case that iOS or Android have much of an advantage over Windows in terms of content creation, and I can easily see Win8 tablets becoming common for a work laptop replacement.

 

[Shadow703793]

I think the biggest opponent Intel will have here is cost,

Keep in mind that Intel has experience with die shrinking. After recouping R&D cost, Intel could potentially sell these chips for very, very, very little cost thus bringing full per product price down. If you want to see an example of this happening, look at Atom. The profit margin Intel has with Atom is pretty big and the main reason for this is die shrinking/small die size.

 

[Blandge]

Great point. Considering that Intel will eventually be using 450mm wafers, potentially hundreds more Atom dies can be made per wafer than Core or Xeon. The new fab being built in AZ is being sized for 450mm

 

[tecmo34]

@Shadow703793... Glad to see you around!!

 

[Shadow703793]

@Shadow703793... Glad to see you around!!

lol. Won't be here for too long. I'm only here because reddit is down…

 

[tecmo34]

So that is where you are hanging out nowadays... Cool... Well, good to have you back even for a short time.

 

[fazers_on_stun]

Mobile: Intel Will Overtake Qualcomm In Three Years

Surprised nobody has linked this yet, what with all the Medfield discussion.

Chris makes some pretty good points - Qualcomm will have to go to OoO eventually to get higher performance with their ARM-design, and Intel has almost 2 decades of experience in that area. Also, looks like Qualcomm barked up the wrong tree when they went with GloFlo for 28nm - no HKMG due to yield issues with gate-first.

 

[Haserath]

Been focused so much on graphics lately. I sort of skipped over that even though I noticed it. After reading Anand's piece on the Medfield platform, I knew Intel was going to have a major advantage against the ARM players.

Krait is already starting off with an OoO -unified reservation station- architecture. Waiting to see the performance and power use of the arch.

If Krait/A15 designs aren't a good deal faster than the Atom Intel has shown at CES, Intel will easily be able to bring Atom up to par. All of the optimizations through the x86 years can be incorporated into Atom for any performance boost necessary. Graphics for the Atom are still being worked on, I suppose. Once Intel has graphics down, the ARM players won't even stand a chance.

Everyone keeps shouting Intel only has a few more shrinks left to go. Well, I don't even know if ARM players will last until our current transistor tech is obsolete. I'll bet that even at the end of the road; someone will bring out the Bulldozer to clear a new one.

 

[fazers_on_stun]

^ Which brings to mind another point Chris' article mentioned - Intel will be effectively two nodes ahead of Qualcomm when they bring out the 22nm version of Atom with trigate. Qualcomm will be stuck at 28nm poly gates it seems, so they won't realize much in the way of power reduction as they would with HKMG. Intel might even be able to go OoO and keep the lead in power usage, in which case they would leap ahead of the competition.

From what I've read, Out-of-Order processing, with all that branch prediction, speculative processing, etc., takes a lot of juice to execute, but it can increase performance tremendously. Which is why AMD's Bobcat design is faster than Atom, but uses much more power.

 

[fazers_on_stun]

Intel Reports Record Year

After the bell on Thursday, chip giant Intel (INTC) reported a record year in terms of revenues. The company posted decent revenue numbers in the fourth quarter, and they beat the midpoint of their lowered guidance when the company warned a few weeks ago. Shares were up about 1.3% after reporting numbers.

Here's how the numbers looked for the fourth quarter:

Revenues of $13.89 billion, beating midpoint of lowered guidance ($13.7 billion). Analysts called for $13.72 billion.
EPS of 64 cents (GAAP), beating analysts' call for $0.61.
Gross margins (non-GAAP) of 65.5% and GAAP of 64.5%. Both numbers met revised guidance from company.

For the full year:

Revenues of $54 billion (GAAP), beating analyst expectations of $53.84 billion.
EPS of $2.39, beating analyst expectations of $2.37.
For the first quarter of 2012, Intel provided revenue guidance of $12.8 billion, plus or minus $500 million. That is exactly where analysts had the company pegged, at $12.8 billion.

*Yawn* - the Intel Money Machine continues to churn, grinding up the competition..

 

[Haserath]

The fact that Intel has its own fabs that are so far ahead of the competition gives Intel the power(pun intended) to conquer the competition. It's not a very fair comparison when Intel will have their shiny, new, fresh out of the lot 22nm tri-gate up against last year's 28nm poly gates. If they don't even have HKMG, that's a nail in the coffin right there.

Out-of-order does seem like it would be a bad design decision for extreme low power designs. It reduces the dependancy on the memory and cache subsystem, but it increases power immensely for the core. It seems like it would be a better idea to enhance the memory handling of the ARM chips instead of going out-of-order. Though the better the algorithms get, the more efficient the design will be overall.

Hyperthreading really makes sense for low power in-order designs, at least to me. It allows for 'out-of-order', parallel in actuality, execution without all of the complexity=greater efficiency boost on in-order designs than out-of-order designs.

Graphics look like big, in-order, parallel processors. This is why they're so efficient compared to x86 processors. Get rid of all of that out-of-order complexity while keeping the instruction execution efficiency, and you have huge amounts of processing power. They don't reach as high on sustained flops thanks to being in-order, but they make up for it in raw processing power. In fact, Knight's Corner looks like a GPU with x86 instruction support.

Intel's execution is superb... I would really love to see them create a fresh architecture and move on with it... if only it could gain support.

 

[Chad Boga]

Everyone keeps shouting Intel only has a few more shrinks left to go. Well, I don't even know if ARM players will last until our current transistor tech is obsolete. I'll bet that even at the end of the road; someone will bring out the Bulldozer to clear a new one.

Who is "Everyone"?

The only people I have seen doing that have been exceptionally uninformed.

Intel's execution is superb... I would really love to see them create a fresh architecture and move on with it... if only it could gain support.

There was Itanium.

 

[esrever]

intel can make it until 11nm with current projected tech which will be a long ways away. By that time there will probably be newer technology that would replace the current laser on silicon action for making chips.

 

[4745454b]

The only people I have seen doing that have been exceptionally uninformed.

I consider myself to be well informed, and I don't think they can shrink forever. I guess it depends on what you mean by "few" doesn't it? They will reach a point where you can't get any smaller. (unless they find a way to make electrons smaller. or start using light.) Our current way of doing things will reach an end. I think I heard once it was around 8nm.

Edit: Forgot to add that I think it will be a bit funny if/when someone hits the "do over" button. Comes up with brand new way of doing things, but needs to be fabbed at 130nm or something like that. 130nm CPU faster then 22nm CPU because of exotic new material.

 

[Chad Boga]

The only people I have seen doing that have been exceptionally uninformed.

I consider myself to be well informed, and I don't think they can shrink forever. I guess it depends on what you mean by "few" doesn't it? They will reach a point where you can't get any smaller. (unless they find a way to make electrons smaller. or start using light.) Our current way of doing things will reach an end. I think I heard once it was around 8nm.

No doubt you would have heard it was around 8nm, you probably also would at one point have heard it was going to be at 22nm, 32nm, 45nm, 65nm etc.

And who said they can shrink forever? My comments were directed at people who claim that Intel only has about 3 shrinks left, and this was going to allow the industry to catch up to them.

Good God, how stupid are they.

 

[esrever]

well at 9 nm they are going to be using just a couple atoms. silicon just won't work beyond that point.

 

[4745454b]

The comment you quoted said few, not three. And I've heard how past Xnm wasn't going to be possible, but with 8nm there is physics behind it that you can't argue against. (that whole rock down a hose problem.)

I am quite positive that Intel will find a way to get the most out of the current tech that they can. But I'm also sure they won't be able to just shrink forever. They will need to come up with something new.

 

[gamerk316]

Mobile: Intel Will Overtake Qualcomm In Three Years

Surprised nobody has linked this yet, what with all the Medfield discussion.

Chris makes some pretty good points - Qualcomm will have to go to OoO eventually to get higher performance with their ARM-design, and Intel has almost 2 decades of experience in that area. Also, looks like Qualcomm barked up the wrong tree when they went with GloFlo for 28nm - no HKMG due to yield issues with gate-first.

Gate first does have its performance advantages though. It is a risk, but if GloFo settles down a bit, it might be one that pays off.

 

[gamerk316]

Chad, theres a difference between "not being able to reach xnm because the technology isn't there" and "not being able to reach xnm because the laws of physics as currently understood do not allow it".

That being said, I do expect to start seeing some work on optical processors in teh VERY near future. And to be fair, we do have some quantum computers out already...[If D-Wave goes bankrupt now, after finally getting a product, after a decade of near infinite funding, I will laugh, a lot].

 

[Chad Boga]

Chad, theres a difference between "not being able to reach xnm because the technology isn't there" and "not being able to reach xnm because the laws of physics as currently understood do not allow it".

The real issue here is the belief that Intel's manufacturing lead is going to run out soon because of these laws of physics.

As a previous poster eluded to, if we and when we can't go any further with silicon, we will switch to a different substance/process, and if the nodes on this new substance are larger, that doesn't mean Intel's manufacturing lead is going to end.

For the foreseeable future, Intel is going to be increasing the gap between itself and all its rivals on the manufacturing front.

There is no imminent catch up coming to save anyone.

 

[gamerk316]

^^ Frankly, I'd argue that most companies would be glad to let Intel handle all the R&D for new gating techs, then jump on board whatever the best one is...two nodes back. Heck of a lot cheaper for them in the long run.

If Intel needs a two process lead to be competitive [be it performance, power, or whatnot], then something is fundamentally wrong with the design compared to the competition.

 

[Chad Boga]

^^ Frankly, I'd argue that most companies would be glad to let Intel handle all the R&D for new gating techs, then jump on board whatever the best one is...two nodes back. Heck of a lot cheaper for them in the long run.

If Intel needs a two process lead to be competitive [be it performance, power, or whatnot], then something is fundamentally wrong with the design compared to the competition.

Where did that notion come from?

Intel is going to take their process lead and extract greater profits with it.