Intel Not Taking Any Chances with Core i7

[MacRumors]

NY Times reports on Intel's extensive testing of the Core i7 (Nehalem) processors which officially launched today. The company has invested over $500 million annually to test these chips in an effort to avoid any show-stopping bugs.

This cautious attitude comes from experience for Intel who recalled their Pentium chips in 1994 after a widely publicized bug in its floating point calculation unit.

After the Pentium flaw, Intel also fundamentally rethought the way it designed its processors, trying to increase the chance that its chips would be error-free even before testing. During the late 1990s it turned to a group of mathematical theoreticians in the computer science field who had developed advanced techniques for evaluating hardware and software, known as formal methods.

Even with such testing, Intel says it would be impossible to evaluate every possible scenario. As an additional safety net, Intel has included software in the Nehalem chips which can be changed after they ship.

Apple is expected to eventually adopt versions of Intel's Core i7 in future Macs.

Article Link: Intel Not Taking Any Chances with Core i7

[kabunaru]

When can we expect Nehalem in a Mac?

Nehalem is probably one of the biggest upgrades a Mac will ever see.

[wrldwzrd89]

This is good news - Intel needs to be careful, since these chips represent a radical design change compared to the previous generation.

As for when we'll see these in Macs - I suspect Macworld 2009 will give us our answer.

[Asar]

we probably wont see these until spring/summer

[sterlingindigo]

Haven't they tested all the other new releases since 1994?

[sangosimo]

I can't wait to get my hand on this and build a new workstation. I am going to have a 16 core monster machine.

[AidenShaw]

Originally Posted by sangosimo

I can't wait to get my hand on this and build a new workstation. I am going to have a 16 core monster machine.

You'll have 16 logical processors - but only 8 core (unless you plan on using some of the very expensive MP server boards and chips).

The dual socket systems can have two quad core processors, with two logical processors per core, for a total 16 logical processors.

Also note that if you usually have 8 or fewer computable threads, you'll almost always be better off disabling hyper-threading. If you usually have 16 or more computable threads, then turn hyper-threading on.

In theory, a good scheduler could handle the situation, but in a dynamic workload it's a difficult problem to get right all the time.

[FJ218700]

Originally Posted by wrldwzrd89

. . . Intel needs to be careful, since these chips represent a radical design change compared to the previous generation.

which is why I'll wait an extra few quarters for the 32's in the mobile version.

[themoonisdown09]

Originally Posted by kabunaru

When can we expect Nehalem in a Mac?

Nehalem is probably one of the biggest upgrades a Mac will ever see.

The Mac Pros could see them around MacWorld 2009. I wouldn't expect iMacs/Macbook/Macbook Pros to see them until Fall 2009.

[Firefly2002]

Originally Posted by kabunaru

When can we expect Nehalem in a Mac?

Nehalem is probably one of the biggest upgrades a Mac will ever see.

One of the dumbest things I've ever heard. 68040? The PowerPC? The 604? G3? G5?

We won't have it for a bit.

Anyway, I don't see the point of this. The issue with the FPU in the original Pentium wasn't an enormous deal and not worth 500 mil.... most bugs are almost impossible to even stumble upon.

The G5 yes.

Back when the G3, and G4 were new, they were the chips to beat.

The G5 was too, at least when it first came out (quickly ceased to be true) especially clock for clock, though AMD was pretty close... Intel of course wasn't even in the ballpark.

Don't forget the rest of the PowerPCs... they were all also the chips to beat (and were unbeatable)

[kabunaru]

Originally Posted by Firefly2002

One of the dumbest things I've ever heard. 68040?

Why so nitpicky?
I meant one of the biggest upgrades the Intel Macs will ever see.

[foobarbaz]

Originally Posted by MacRumors

As an additional safety net, Intel has included software in the Nehalem chips which can be changed after they ship.

I think that has been standard since, like, the day after the FDIV bug...

[uricmu]

I'm not sure why this is news.

I used to work for a validation unit in the past, and they obviously conduct some random testing and random testing obviously can't cover any scenarios.

Newer architectures or architectures with new components or instructions obviously get longer tests or new tests compared to simple steps between processors with same architecture (since then you can just do regression testing).

[Spades]

Is this somehow different from the updatable microcode that's been used in processors for over a decade?

[macsmurf]

Originally Posted by uricmu

I'm not sure why this is news.

I used to work for a validation unit in the past, and they obviously conduct some random testing and random testing obviously can't cover any scenarios.

Newer architectures or architectures with new components or instructions obviously get longer tests or new tests compared to simple steps between processors with same architecture (since then you can just do regression testing).

Well, the news is pretty sparse on details but mention formal methods, which is a completely different approach from 'random' testing. Formal methods consists of creating a mathematical model of the system, proving that the model have certain (interesting) properties, and, to some extent, arguing that the model is a reasonable approximation of the actual system.

Formal methods were thought to provide a whole new way of doing software development in the 1980's but have since fallen from grace somewhat. I was not aware that it was a big thing in hardware development still. If someone has some references I would love to see them.

[uricmu]

Originally Posted by macsmurf

Well, the news is pretty sparse on details but mention formal methods, which is a completely different approach from 'random' testing. I was not aware that it was a big thing in hardware development still. If someone has some references I would love to see them.

In the late 90s they were doing both since you can obviously get more coverage with random testing than with formal methods. Formal methods were useful or testing certain behaviors and protocols or stateful components such as memory controllers. I would venture a guess that they are still doing the same with Nehalem and that the NY piece is more of a PR thing than anything new.

[iMacmatician]

Originally Posted by kabunaru

When can we expect Nehalem in a Mac?

Mac Pro: H1 2009
iMac, MacBook Pro: H2 2009 (quad-core), H1 2010 (dual-core)
MacBook: H1 2010
Mac mini: H1 2010 or later

[kabunaru]

Originally Posted by iMacmatician

Mac Pro: H1 2009
iMac, MacBook Pro: H2 2009 (quad-core), H1 2010 (dual-core)
MacBook: H1 2010
Mac mini: H1 2010 or later

Why does a quad-core seems more plausible to come first than a dual-core?

[iMacmatician]

Originally Posted by kabunaru

Why does a quad-core seems more plausible to come first than a dual-core?

Because Clarksfield is coming in Q3 2009 and Auburndale is coming in January 2010.

[HerrRolf]

Originally Posted by MacRumors

The company has invested over $500 million annually to test these chips in an effort to avoid any show-stopping bugs.

WOW! That much money just to test a product. This seems astronomically high.

[dagamer34]

Originally Posted by HerrRolf

WOW! That much money just to test a product. This seems astronomically high.

With a revenue of $38.3 billion per year, it's easy to see why spending $500 million to prevent major mishaps is an afterthought for Intel. You have to remember that a widescale product failure like nVidia had recently affects much more than the bad chips sent out. Your stock price definitely feels the pain and takes a while to recover. Plus, manufacturing chips always leads to some defects, so it's better to catch them early and resell them at speeds they can run at rather than see them fail in the wild.

[The Tall One]

Sooooooo when can I get my new MacPro with awesome core chips? After MacWorld? Sweet.

[telecomm]

Wow, this is big news. I bet we'll see other chip companies start to do extensive product testing too.

[SpinThis!]

I'm guessing we won't see new Mac Pros at Macworld which has traditionally been a consumer products oriented show. I have a suspicion if Apple does announce new Mac Pros, they'll probably be available in February or March or even do an announcement shortly thereafter in the year.

This is brand new tech—it's not just a motherboard revision but a whole new direction—like transitioning over from Mac OS 9 to Mac OS X and I'm sure Apple is going to want to test Mac OS X as much as possible before it ships any Core i7-based kit. Apple needs to provision/design new motherboards, graphics cards, etc. Especially for Apple Pro customers, that's not something you want to rush. Apple might even go so far as to wait until Snow Leopard ships to deliver i7 goods to really show off what those chips and its new OS is capable. But it will be worth the wait I'm sure.

In the mean time, expect a quad core iMac and new Cube at Macworld.

[iMacmatician]

Originally Posted by SpinThis!

In the mean time, expect a quad core iMac and new Cube at Macworld.

Based on what?