[Fudz] Intel 2011 extreme Core i7 is 990X

[aigomorla]

Fair enough.

I've always wondered though, why are the linewidths different for SSDs, CPUs and GPUs, anyone know ?

I mean CPUs are on 45nm going to 32nm, while GPUs are on 40nm going to 28nm, and SSDs are on 34nm going to 25nm.

Im sure IDC can tell ya..

 

[Soleron]

Fair enough.

I've always wondered though, why are the linewidths different for SSDs, CPUs and GPUs, anyone know ?

I mean CPUs are on 45nm going to 32nm, while GPUs are on 40nm going to 28nm, and SSDs are on 34nm going to 25nm.

CPUs and GPUs could be made on the same node (e.g. both 90nm) but the economics of GPUs have worked out so that half-node shrinks like 55nm and 40nm have made sense more recently.

CPUs are made on a higher performance process (higher clockspeeds, more power use, more expensive to develop) than GPUs typically. This is changing with on-die graphics and Fusion.

SSDs are very much simpler logic than CPUs and GPUs; they don't need much custom design and the repeating unit is very small. So they can be done on a smaller node sooner. If you tried to do CPUs on 25nm now the yields would be abysmal. The way companies like Intel/GF usually do it is create a series of test chips for a new node over the years leading up to it. For 32nm we saw a test wafer with SRAM cells, then a test wafer with generic logic, then actual CPUs, over the course of two years. SSDs are on the simpler end of that scale.

 

[A_Dying_Wren]

I didnt know fabs had such leeway in linewidths. I thought a 45nm facility would absolutely only produce 45nm but could potentially churn out GPUs, CPUs and SSDs.

I must be mistaken

 

[Scali]

Hell, i'll think we'll be looking at stock processors hit 4.0Ghz by next year.

Heh, the first 3.8 GHz stock Pentium 4 was introduced in November 2004... so it's only taken them about 6 years to make that final 200 MHz step

 

[extra]

now why would they do that when there ramping there 22nm fab process in lue of SSDs?

Once SSD market is semi saturated, then the SSD fabs will change over to CPU fabs.

I don't think this is accurate...I could be way wrong though! I thought that the 25nm SSD's that Intel was going to be making did not use flash produced by intel... ??? In fact, I'm almost positive if it is in fact being made by Micron, in some sort of Intel/Micron partnership. I don't think the SSD and CPU fabs have anything to do with each other in this case...pretty sure.

 

[aigomorla]

I thought that the 25nm SSD's that Intel was going to be making did not use flash produced by intel... ???

http://www.pcper.com/article.php?aid=863&type=expert

i thought this fab was owned by intel?

doesnt look much different from a cpu fab.

Guys i could be wrong... as i said when i was last talking to my friend, thats the gist of hints he would give me until he was allowed to speak more to his testers.

 

[Idontcare]

I didnt know fabs had such leeway in linewidths. I thought a 45nm facility would absolutely only produce 45nm but could potentially churn out GPUs, CPUs and SSDs.

I must be mistaken

If you don't work in the industry it is easy to get this impression from the way the press and the media talk about technology transitions and process nodes. So don't feel bad or think you missed out on something obvious.

Fabs can in fact, and routinely do so, produce products on multiple nodes in parallel at the same time.

Think of it like McDonalds or Burger King. They have a "burger station" in the back, even if you haven't worked at a fast food restaurant you've probably been to one enough times to have basically witnessed how they prepare the food.

Now this "burger station" has all the condiments needed to make different burgers. Ketchup, pickles, cheese, etc. And from that one burger station they can create all the different items on the menu. Some of those items use the same ingredients. Order a chicken sandwich or hamburger and the people in the back are going to need to use the same mayo and the same lettuce for both.

Process nodes and fabs are a lot like this. Each piece of equipment in the fab, we call them "tools" in the industry, is capable of doing many different processes, we actually call these "recipes" in the industry.

Think of it like a blender or a mixer, the recipe calls for the specific ingredients to be added to the mixer - 1 cup flour, 1/2 cup sugar, etc - depending on what it is you intend to make with the batter. Pancakes would use different ingredients and quantities than say biscuits or a cake.

So we use these "tools" (the blender/mixer) combined with specially engineered "recipes" on any given batch of wafers (we call them "lots") depending on what process node the wafers belong to and what step in the process flow the wafers are at. A batch of 90nm wafers may need a 60 second dip in phosphoric acid while the next batch of wafers to show up at the tool are 45nm wafers that need a 15 second dip in phosphoric acid (or could even be a different acid altogether, but still processed on the same tool).

A mature fab that has been around for 10yrs will probably have 5 or 6 nodes running in parallel.

Now a BIG difference comes in when we talk about different types of process nodes. Memory versus logic. Flash, dram, cpus.

A process node developed to make dram cannot be used to make flash, you need specialized features to make flash that are not present in dram. We call these "components" in the industry. Same with logic.

A logic process node can't be used to make dram because the logic node will have been developed without trench capacitor components. If you wanted to make dram with the logic node you would need to further develop all the specialized components that make dram different from cpus.

That McDonald's burger station could be used to make Arby's roast-beef sandwiches but they don't have the necessary ingredients (sliced roast beef versus ground hamburger) to create a roast-beef sandwich.

edit: this is a somewhat dated graphic by now, needs some updating, but you see all those nodes that are still in production at TSMC? Those all run in parallel in about 3 different fabs. The older nodes aren't qualified in the newest of fabs of course, but they just add the newest node to the pre-existing fabs by installing a few unique and enabling tools (like immersion litho) in the pre-existing fab, requalify all the existing tools with the new recipes and start ramping production.

 

[Idontcare]

Heh, the first 3.8 GHz stock Pentium 4 was introduced in November 2004... so it's only taken them about 6 years to make that final 200 MHz step

lol yeah but look at those xtor counts climbing 10x while maintaining that clockspeed and power-consumption levels!

http://www.pcper.com/article.php?aid=863&type=expert

i thought this fab was owned by intel?

doesnt look much different from a cpu fab.

Guys i could be wrong... as i said when i was last talking to my friend, thats the gist of hints he would give me until he was allowed to speak more to his testers.

Don't forget that IM Flash is a joint-venture. You could say Intel owns that fab in the same sense that you could say AMD still owns the Dresden fab (just a smaller percentage of it than in Intel's case) but it operationally it really functions as a discreetly separate business. Management decisions are made daily that have little to do with Intel.

Take the process node development, handled entirely outside the R&D infrastructure that exists within Intel, or the one that exists within Micron for that matter.

 

[aigomorla]

And u guys all see why IDC is an elite.

LOL...

EH Moderator?

IDC when did u get mod?

Why is it that i never noticed this until now and i am a mod?

:X


Anyhow keeping on note on the 990X, i think i will get to review one b4 Anand does like the 980X.
So Check the forums, and u shall see it one day...

You could say Intel owns that fab in the same sense that you could say AMD still owns the Dresden fab (just a smaller percentage of it than in Intel's case) but it operationally it really functions as a discreetly separate business. Management decisions are made daily that have little to do with Intel.

Oh no i understand, but its from maturing these fabs, that they will get a boost in the 22nm cpu fabs.
If they start pumping out fully cycled 25nm fabs, im sure they can adopt that to 22nm cpu fabs.

What im trying to say and what my friend hinted on is, that intel is in a 2 way road block waiting on what to do.

1. Fight BD with full force.
2. Try to slam to market with SSD's so public ignores BD.

Either way, the 2 paths my friend was hinting resulted in a node shrinkage.

But he did made it very clear, intel wont lose.

But IDC we can see intel is already attempting 2, its just a matter of time to see if 1, will be implemented.

 

[Idontcare]

IDC when did u get mod?

Why is it that i never noticed this until now and i am a mod?

:X

lol, yeah like maybe 18hr ago

http://forums.anandtech.com/announcement.php?f=8&a=18

It was a conspiracy to keep you out of the loop, special request made by me

 

[aigomorla]

lol, yeah like maybe 18hr ago

http://forums.anandtech.com/announcement.php?f=8&a=18

It was a conspiracy to keep you out of the loop, special request made by me

ROFL video cards??

Your way too technical for video cards, and your maturity level is way high for the red vs green going on in there.

It would be like you showing off a classic shelby cobra roadster in front of a bunch of civic ricers!

Good luck in that section tho.

 

[Idontcare]

dam aigo, aint nothing sacred with you bro!

But seriously now, tell us how you really feel lol

 

[Daedalus685]

ROFL video cards??

Your way too technical for video cards, and your maturity level is way high for the red vs green going on in there.

It would be like you showing off a classic shelby cobra roadster in front of a bunch of civic ricers!

Good luck in that section tho.

Take that Asian made cars!!

 

[aigomorla]

dam aigo, aint nothing sacred with you bro!

But seriously now, tell us how you really feel lol

i love that section in all honesty.

But there is just way too much red vs green going on.

More so then red vs blue here.

Actually the red vs blue would be like a picnic compared to red vs green over there.

 

[Martimus]

Congrats IDC. Although I fear what moderating the Video Forum will do to you. I have already seen how it changed Keys writing style over the last couple years.

 

[bryanW1995]

yes, red vs green is pretty intense some times. hopefully idc will hand out a few lengthly vacations and keep the riff raff in line.

 

[tweakboy]

I guess they want to make it their flagship desktop product of Q1. Then Q2 and Q3 we will see new northbridge Im shure and the i9 core making its debut or am I off here ? Is i9 a 2011 product or not ?

 

[aigomorla]

yes, red vs green is pretty intense some times. hopefully idc will hand out a few lengthly vacations and keep the riff raff in line.

lol which is why i love my section the most.
The only thing we can really argue about is.

Megahalem vs Nocunta D14.

And i wouldnt even call that a really intense situation.

Oh we also have our occassional "fan" boys.. and i mean people who toot fan X is the best once in a while.
But i like Ehume's thread, and it basically makes people understand.

I guess they want to make it their flagship desktop product of Q1. Then Q2 and Q3 we will see new northbridge Im shure and the i9 core making its debut or am I off here ? Is i9 a 2011 product or not ?

to be honest, i will repeat what everyone has been asking me.

I have yet to see even a 2011 lineup.
You guys what me to shoot out ETAs? :ninja:

I dont work for intel, and my friend isnt saying much about 2011.
I dont expect consumers to see it until enterprise has it.

 

[Daedalus685]

lol which is why i love my section the most.
The only thing we can really argue about is.

Megahalem vs Nocunta D14.

And i wouldnt even call that a really intense situation.

Oh we also have our occassional "fan" boys.. and i mean people who toot fan X is the best once in a while.
But i like Ehume's thread, and it basically makes people understand.



to be honest, i will repeat what everyone has been asking me.

I have yet to see even a 2011 lineup.
You guys what me to shoot out ETAs? :ninja:

I dont work for intel, and my friend isnt saying much about 2011.
I dont expect consumers to see it until enterprise has it.

I think from now on we should say y2011 and s2011 for the year's lineup and socket's line up respectively... I'm confused... Couldn't they have used a few less pins!!

Also no rush on the ETA... I don't expect anything worth upgrading to until well into q3 next year... so take it easy! Is there any info for sure that s2011 will be a consumer socket at all? Not that I would mind buying a quad channel MB... but I'd reckon the price woudl be prohibitive to a lot of folks.

 

[Martimus]

I think from now on we should say y2011 and s2011 for the year's lineup and socket's line up respectively... I'm confused... Couldn't they have used a few less pins!!

Also no rush on the ETA... I don't expect anything worth upgrading to until well into q3 next year... so take it easy! Is there any info for sure that s2011 will be a consumer socket at all? Not that I would mind buying a quad channel MB... but I'd reckon the price woudl be prohibitive to a lot of folks.

lol. This is the first time I realized that Socket 2011 is scheduled to be released in year 2011.

 

[Idontcare]

I think from now on we should say y2011 and s2011 for the year's lineup and socket's line up respectively... I'm confused... Couldn't they have used a few less pins!!

We could just use the existing nomenclature that we routinely used in the forum back in the LGA775 days. Let 2011 be 2011, everyone is comfortable with years being years.

But it makes a lot of sense to address socket 2011 as LGA2011, most folks here would know exactly what that meant, even new folks would know what that meant if they happened to be coming from another forum.

Someone coming here from HOCP seeing us talking about y's and s's would be even more at a loss. (no that's not a dig at HOCP forum members maybe )

 

[Daedalus685]

We could just use the existing nomenclature that we routinely used in the forum back in the LGA775 days. Let 2011 be 2011, everyone is comfortable with years being years.

But it makes a lot of sense to address socket 2011 as LGA2011, most folks here would know exactly what that meant, even new folks would know what that meant if they happened to be coming from another forum.

Someone coming here from HOCP seeing us talking about y's and s's would be even more at a loss. (no that's not a dig at HOCP forum members maybe )

heh, well I didn't mean we have to use y's and s's.... just that '2011 road map' on its own confounds me at times . But hey, something to be said for anand tech's very own nerd code speak

 

[Idontcare]

heh, well I didn't mean we have to use y's and s's.... just that '2011 road map' on its own confounds me at times . But hey, something to be said for anand tech's very own nerd code speak

heh, you and me both! in fact the very reason I even opened the article on Fudz to read about the 990X was because I thought he was claiming the 990X was going to be an LGA2011 chip which would then mean (my thinking it out in the microseconds of eternity that spanned while my mouse was still moving to click the fud article link) that LGA2011 was coming out sooner, much sooner, instead of later.

Then to my dismay I realized he meant 2011 as in the year 2011. Lame.

 

[aigomorla]

Rofl..

dont worry guys.. i am to please.

and im sure you were happy with my last gulftown preview.