Intel Starts Production of Next-Generation Haswell Microprocessors.

[Idontcare]

Cheap TIM held Ivy back, and IIRC its pretty much been confirmed that Intel is going back to solder. And with the extra time of 22nm under their belt plus other overclocking friendly features being added in to the new architecture, I'd almost be willing to bet 5GHz will finally be a new norm for the enthusiast willing to spend ~$100 on cooling

However I'd bet against IPC improvement being that high, it'd wager it will be closer to 10% at best

It isn't the quality of the TIM that holds IB back; rather, it is a matter of how thick the TIM layer itself is.

Also, unless Intel has internally changed their decision on the matter, last time an Intel individual talked with me about the topic the plan of record was to package Haswell the same way as they package IB (no solder for desktop socketed cpus).

 

[ShintaiDK]

Ok, I'll change my wording then, IB on 22nm FinFET isn't tuned for higher clocks than SB managed on 32nm. Point in case, I don't expect Haswell to be different in this area and I haven't read otherwise.

Haswell and IB uses different 22nm transistors. If that will change anything as such tho is to be seen.

 

[Fx1]

But only barely in the case of http://www.anandtech.com/show/6194/asus-ux31a-putting-the-ultra-in-ultrabooks/8 for example.

I doubt Haswell will bring much of an improvement to desktops. 22nm FinFET isn't tuned for high clocks and Haswell will not bend physics.

Except the Macbook has way better cooling.

 

[Smoblikat]

Hopefully this chip will be an upgrade to the 2500K. Ivy was a no-grade (unless you are using a craptop).

 

[Nemesis 1]

But only barely in the case of http://www.anandtech.com/show/6194/asus-ux31a-putting-the-ultra-in-ultrabooks/8 for example.

I doubt Haswell will bring much of an improvement to desktops. 22nm FinFET isn't tuned for high clocks and Haswell will not bend physics.

Haswell isn't tring to bend physics. And linking to an article about IVB comparring to haswell is of no value , Haswells GPU and CPU no longer share the same power plane as they did with IVB. Also comparring IVB on 3d 22nm to Haswell isn't correct either. SB was built for the 32nm process. IVB is simply a shrink with small improvements. Haswell is made from the ground up for 22nm process and Efficiency. Haswell running at 4ghz and above will perform hust fine . It What Haswell does at 15watts and below that matters this round. That we have to wait to see.

 

[khon]

They dont overheat. And they produce LESS heat. Meaning your laptop should be more comfortable to use than for example a sandy bridge mobile chip. And there is no TIM/IHS on mobile chips.

They do overheat, otherwise they wouldn't need to throttle, which most of the high end laptops do when you fully stress the CPU.

As for the comparison to Sandy Bridge you're correct that they produce less heat, but laptops have gotten a lot thinner than they were 2-3 years ago, so heat is actually more of an issue now than it was then. And if you want to take it a step further and go for a tablet format, it becomes an even bigger issue.

Anyway, what I'd like from Haswell is improved IGP, lower idle power consumption, and lower average power consumption during use. CPU performance can stay the same for all I care.

 

[Homeles]

Hopefully this chip will be an upgrade to the 2500K. Ivy was a no-grade (unless you are using a craptop).

You must have a different definition of "upgrade" from the rest of the world.

 

[SiliconWars]

You must have a different definition of "upgrade" from the rest of the world.

Not really. I bought this 2500K after Ivy was launched. I figured Ivy wasn't worth the extra $5 due to its inferior power consumption and heat when overclocked.

 

[ShintaiDK]

A 4.7Ghz IB is around a 5Ghz SB

 

[SiliconWars]

A 4.7Ghz IB is around a 5Ghz SB

With furnace-like heat, yes.

 

[Idontcare]

With furnace-like heat, yes.

What point the hyperbole?

There is no comparing the operating temperatures of either processor to that of a furnace (TJmax is well below the operating temp of a furnace), and IB consumes less power than SB at any given clockspeed.

 

[scannall]

Glad to see that it appears to be on time. I skipped both 'Bridges' as my OC'd i7-920 never ran into anything it couldn't handle. I suppose I could wait until Broadwell, but I'm itching for something new fairly soon.

 

[blackened23]

With furnace-like heat, yes.

The temperatures between SB and IB are not directly comparable because Ivy has a wider thermal envelope than SB. Furthermore, the temperatures in your chart are hardly surprising because:

1) Ivy bridge is designed to run at 1 volt and some change, so you're over volting it a ton at 1.35v. So you're comparing a SB with a small overvolt to an IB with a huge over volt. Not exactly an apples to apples comparison.

2) The die size is substantially smaller than SB. Because of the heat dissipation properties, the temps of the chip will be higher when overclocked.

None of this matters because the chip is designed to tolerate higher temperatures.

 

[Homeles]

With furnace-like heat, yes.

You don't understand how physics works.

Thermodynamics. Read about it.

 

[Piroko]

@FX1 Certain configs of the '11 MBP had overheating issues and the retina isn't exactly cool to the touch either when loaded. And that's with one model with exceptional validation compared to the norm.

Haswell isn't tring to bend physics. And linking to an article about IVB comparring to haswell is of no value , Haswells GPU and CPU no longer share the same power plane as they did with IVB. Also comparring IVB on 3d 22nm to Haswell isn't correct either. SB was built for the 32nm process. IVB is simply a shrink with small improvements. Haswell is made from the ground up for 22nm process and Efficiency. Haswell running at 4ghz and above will perform hust fine . It What Haswell does at 15watts and below that matters this round. That we have to wait to see.

Am I the only one who has trouble with extrapolating his point?
Even if they don't share the same power plane, they still share the same TDP and cooler. Haswell isn't made from the ground up either afaik but an iteration to Ivybridge. And Ivybridge should already be adapted fairly well to the 22nm process tech or else we wouldn't have seen its power consumption drop this much in the first place.

 

[SiliconWars]

What point the hyperbole?

There is no comparing the operating temperatures of either processor to that of a furnace (TJmax is well below the operating temp of a furnace), and IB consumes less power than SB at any given clockspeed.

Your graphs do not appear to have much in common with tech press results who found temperatures far higher at any given clockspeed and runaway power on IVB after 4.8-4.9 GHz.

 

[SiliconWars]

The temperatures between SB and IB are not directly comparable because Ivy has a wider thermal envelope than SB. Furthermore, the temperatures in your chart are hardly surprising because:

1) Ivy bridge is designed to run at 1 volt and some change, so you're over volting it a ton at 1.35v. So you're comparing a SB with a small overvolt to an IB with a huge over volt. Not exactly an apples to apples comparison.

2) The die size is substantially smaller than SB. Because of the heat dissipation properties, the temps of the chip will be higher when overclocked.

None of this matters because the chip is designed to tolerate higher temperatures.

Yes 5 Celsius more. However TPR found this -

http://techreport.com/news/22859/why-is-ivy-bridge-so-hot-and-bothered

To keep things even, we aimed for similar levels of power consumption for the two CPUs, which should translate directly into similar levels of heat to be dissipated. Fortunately, we were able to reach rough power parity at equal clock speeds. At 4.9GHz, the Ivy-based system draws 236W at the wall socket, and the Sandy rig pulls 231W.

Turns out the difference in temperatures between the two was quite a bit wider than the 5°C difference between the thermal throttling thresholds. Our Ivy CPU flirts with 100°C at those settings, while the Sandy Bridge chip is currently running at about 80°C in the same test system.

I dunno why people are pretending that IVB isn't a hot and power hungry chip when overclocked. There's a good reason why the OC community steers well clear of it.

 

[Idontcare]

Your graphs do not appear to have much in common with tech press results who found temperatures far higher at any given clockspeed and runaway power on IVB after 4.8-4.9 GHz.

That is because the "tech press" spent almost no time attempting to optimize the power curves for either processor. Are you all that surprised or easily fooled?

How many "tech press" authors are industry-experienced engineers with PhD's to know when they are doing things right versus half-assed out of ignorance?

There is a big difference between saying "here is what IB can do" versus saying "here is all I bothered to do with IB".

If I had any confidence in the "tech press" then I wouldn't spend my time duplicating their efforts with my own tests. Think about it.

 

[SiliconWars]

What do you mean by "optimize the power curves"?

The tech press is never going to go into that kind of detail regardless. How many of them dropped Brazos to 1.15V, how many of them bothered with K10 stat for Llano and Trinity?

The only way to judge each is out of the box and in tech reviews. Who's to say your IVB sample isn't a very good one and your 2600K bad?

 

[Idontcare]

Who's to say your IVB sample isn't a very good one and your 2600K bad?

My results are not exceptional in any regard when compared to my colleagues who have taken their time to optimize the power curve as well.

But if you are relying on any "tech press" as your basis for claiming insight into the underlying device physics that come from the process node itself, then you are the one who is doing themselves a great disservice.

Out of the box IB results tells you nothing as to how a non-lidded IB will perform (anything mobile), nor does it tell you how the 22nm process tech itself performs (which means you have no real grasp on how Haswell will leverage it either, per the thread's topic).

 

[Clorky]

What about pricing? Where prices start?

 

[rgallant]

Yes 5 Celsius more. However TPR found this -

http://techreport.com/news/22859/why-is-ivy-bridge-so-hot-and-bothered





I dunno why people are pretending that IVB isn't a hot and power hungry chip when overclocked. There's a good reason why the OC community steers well clear of it.

-that link is based on a crap ES[3Q 2011?] sample it seems , so can you trust any base numbers or conclusions from those test ?

 

[Fx1]

If haswell is a bust then we are screwed because who can justify buying a 3 gen old CPU in a new system

 

[Homeles]

If haswell is a bust then we are screwed because who can justify buying a 3 gen old CPU in a new system

What would make it a bust? And your ridiculous assertion assumes that Ivy Bridge wasn't an improvement over Sandy Bridge, which only a lunatic would argue — especially in the face of the evidence that has been presented on this very page.

 

[Fx1]

What would make it a bust? And your ridiculous assertion assumes that Ivy Bridge wasn't an improvement over Sandy Bridge, which only a lunatic would argue — especially in the face of the evidence that has been presented on this very page.

No one claims that Ivy is better than Sandy. 1-3% IPC improvements and a big thermal problem is what Ivy was. 22nm has brought more issues than it fixed.

Haswell has had big expectations since people first heard about it. If it fixes 22nm and offers better performance with the overclocking headroom of Sandy then its a winner.

Especially if IPC improves a decent margin to