Possibility of another 32nm Westmere SKU

[IntelUser2000]

Originally Havendale and Auburndale were going to be the first Nehalem architecture based IGP processors. They were dual cores with HT, and had an IGP as an MCM. We could say that the combined size of Havendale/Auburndale+IGP is similar to the quad core Bloomfield/Lynnfield/Clarksfield CPUs.

Now with 32nm we'll see 32nm 6-core Gulftown on the high-end with X58, and 2 core IGP variants with Clarkdale/Arrandale. Roadmaps don't show it, and it doesn't make sense, where are the options for quad cores and IGP?

Maybe there's a possibility that we'll see 32nm Lynnfield derivative with IGP on MCM. If they only release IGP variants on dual core, Intel will lose vast portion of the IGP market. By creating a 4 core + IGP part, they will cover most of the market and it still won't affect the ultra high end Gulftown.

 

[ilkhan]

Its entirely possible, but we still haven't heard of *any* 32nm quad parts. Until we do... :shrug: I think your scenario is likely, but we'll see what Intel has to say about it. Id like to see a 32nm quad paired with a 32nm GPU. But that won't be for a while. The nehalem/westmere tock/tick generation is rather fucked up as far as timelines go.

 

[Idontcare]

Originally posted by: IntelUser2000
Now with 32nm we'll see 32nm 6-core Gulftown on the high-end with X58, and 2 core IGP variants with Clarkdale/Arrandale. Roadmaps don't show it, and it doesn't make sense, where are the options for quad cores and IGP?

It makes sense to me from a product differentiation standpoint.

Enthusiast class will have 3x the cores and threads as mainstream class. No amount of overclocking on a cheap mainstream part is going to bridge the performance gap to the much more expensive 32nm LGA1366 SKUs.

We'll see a quadcore Westmere the day Intel becomes concerned with the possibility of AMD fielding fast enough PhII's (or BD if that comes early) as to compete with the top-end of the 2C/4T westmere SKU's.

 

[IntelUser2000]

True, but there is also the pricing problem. Initially with Lynnfield as being the quad core part, we'lll see it occupying the $150 to $500 segment, with Clarkdale going at $150 and below. After release of Gulftown, where will be the $150 to $500 parts? Lynnfield on 45nm?? What are they doing making such a vast gap? What about the P55 + Lynnfield buyers that will have no 32nm upgrade path? Throw the P55 and buy H55, go with Clarkdale that's dual core, or buy an X58 with Gulftown?

 

[Idontcare]

Everything publicly stated and discussed by Intel would say that is exactly the plan until Sandy Bridge comes along.

Look how long they milked 65nm Q6600 well past 45nm debut. I don't claim to know anything about their strategy, only that to date they have been fairly consistent and logical insofar as I have cared to think about it.

Quad core + IGP? There's no great super stellar reason to be stealthy about the roadmap there, 95% of the cat is out of the bag already, what possible element of surprise would be leveraged by a stealth launch of a quadcore w/IGP?

Personally, just to be pragmatic about it, from a TDP perspective I can see why a quadcore (capable of 8 threads) combined with a 45nm IGP all under one IHS would potentially put unacceptable TDP limits on the clockspeed of the quadcore itself.

Maybe once a 32nm IGP is MCM'ed under the IHS, or who knows where Larrabee intersects all this with Sandy Bridge in another year.

 

[IntelUser2000]

Could be just a stealthy release to keep the competitor antsy. They must be really confident about dual core Westmere performance if they choose not to release a quad core version.

From a TDP perspective, it has to do with "We could say that the combined size of Havendale/Auburndale+IGP is similar to the quad core Bloomfield/Lynnfield/Clarksfield CPUs."

Like top end at the time of release + IGP is not acceptable in terms of die size and TDP, but a lower end that has combined TDP and die size of a top end might be.

For example on 45nm:

4 cores die is roughly equal to 45nm dual core CPU + IGP

On 32nm:

6 cores die is roughly equal to 32nm quad core CPU + IGP

The entire Northbridge is said to be 10-12W so its not that big of a deal in theory. We'll see how it pans out.

 

[TuxDave]

If you were the type to buy a quad core in a 2 core/4 core market, wouldn't you be the type to buy a 6 core in a 6 core/2 core market? But yeah if the price gap is significantly bigger than the 4 core - 2 core gap, I guess I could see a problem.

 

[ilkhan]

I figure we'll see i3 from $50-$200, i5 from $150-500, and i7 from $300-1000.
32nm quad + 32nm iGPU makes sense to me as an i6 branded part, but thats still just opinion.

 

[21stHermit]

Do you suppose that the IGP for Clarkdale will be the same IGP for the next release of Atom, Pine? with an IGP inside the package with Atom?

As I understand the new Pine? Atom will show late 2009, about the same time frame as Clarkdale.

 

[ilkhan]

um. doubtful. Atom (dual core) is about 52mm^2, probably the same size as the clarkdale GPU itself.

 

[Idontcare]

Originally posted by: ilkhan
um. doubtful. Atom (dual core) is about 52mm^2, probably the same size as the clarkdale GPU itself.

I suspect that GPU in this photo is a wee bit larger than 52mm^2...considering that it is sitting next to a dual-core Westmere which is prolly ~90mm^2 in its own rights.

 

[ilkhan]

ok so 2.5x bigger than the atom dual core. Its huge in comparison.

 

[Idontcare]

Yeah...which really makes you wonder what the heck does Intel have under the hood for that IGP given that it's probably as large or slightly larger than a penryn (wolfdale) chip (or an ATI RV740 for another comparison).

When you think of what a wolfdale can accomplish computation-wise, and the RV740, one wonders just what Intel is shooting for performance-wise with that 45nm IGP coupled with westmere in clarkdale.

 

[ilkhan]

Intel's IGP goals are fairly well known, they don't want to be in the shitter anymore. Id say for anything up to low res crysis the clarkdale IGP will be fine. Clarkdale will be an effective gaming CPU when paired with a high performance GPU. The IGP isn't it.

My caring about clarkdale is only to the extent that its the same as arrandale. Leave gaming to discrete desktop chips. It'll work better. I've yet to find a valid need for a gaming laptop, and NO need for high performance laptop IGP. For arrandale its almost exclusively for running aero, and low demand 3D.

A laptop needs a low power GPU (IGP). A non-gaming desktop needs a low power GPU (IGP). A gaming desktop needs a high performance GPU (not an IGP). Yeah this'll work in a pinch, but thats not the primary design goal. If intel can squeeze RV740 level performance from it, that'd be amazing. But I doubt they have the TDP budget for that.

 

[Idontcare]

Well one would hope for the shareholder's sake that there is a rationalization why they are packing such a large (more costly) IGP into clarkdale versus one that is half the size and 1/4 the cost.

If it is merely going to be fast enough for everything but gaming, then it could accomplish that with half the xtor budget too.

As you say, there is a threshold below which it becomes an "all or none" situation...its either fast enough that no one cares to displace it by adding a discrete GPU solution or its slow enough (regardless how slow) that it does get displaced.

So why build-in a large-die high(er) cost chip just to eat into clarkdale's gross margins?

One would hope it required the xtor budget (and thus mm^2 budget) and cost because it actually meets some non-arbitrary performance target that enables the customer to extract value from it versus an IGP that was half it's performance...and hopefully that performance target from the large IGP is critically enabling of some level of application capability. One would hope :laugh:

 

[Denithor]

Right now a fast dual-core will handle the majority of games but we are seeing (rapidly) increasing support for quads (and in many cases - not just support - necessity).

I wonder how these new 2 physical/4 virtual core processors will handle the games that need three to four cores to run properly?

Because I for one would like to make 32nm my entrance into the i5/i7 architecture - but they don't appear to be planning the 4/8 chip I would like. So I'm either going to be stuck with a 45nm 4/8 (hot), a 32nm 2/4 (perhaps insufficient going forward) or a 32nm 6/12 (nutso expensive).

 

[CoinOperatedBoy]

Originally posted by: Denithor
Right now a fast dual-core will handle the majority of games but we are seeing (rapidly) increasing support for quads (and in many cases - not just support - necessity).

I wonder how these new 2 physical/4 virtual core processors will handle the games that need three to four cores to run properly?

Because I for one would like to make 32nm my entrance into the i5/i7 architecture - but they don't appear to be planning the 4/8 chip I would like. So I'm either going to be stuck with a 45nm 4/8 (hot), a 32nm 2/4 (perhaps insufficient going forward) or a 32nm 6/12 (nutso expensive).

Out of those options, I'd choose "hot".

 

[Idontcare]

Originally posted by: CoinOperatedBoy

Originally posted by: Denithor
Right now a fast dual-core will handle the majority of games but we are seeing (rapidly) increasing support for quads (and in many cases - not just support - necessity).

I wonder how these new 2 physical/4 virtual core processors will handle the games that need three to four cores to run properly?

Because I for one would like to make 32nm my entrance into the i5/i7 architecture - but they don't appear to be planning the 4/8 chip I would like. So I'm either going to be stuck with a 45nm 4/8 (hot), a 32nm 2/4 (perhaps insufficient going forward) or a 32nm 6/12 (nutso expensive).

Out of those options, I'd choose "hot".

Guys don't forget that it's 2C/4T 32nm goodness...you likes 5GHz+ on air?