Sign in with
Sign up | Sign in
Your question
Closed

Is Intel losing its process lead?

Tags:
  • CPUs
  • Processors
  • Intel
  • Cortex
  • Product
Last response: in CPUs
Share
February 24, 2010 6:49:46 PM

Wow! AMD was supposed to have the new 28nm Global Foundries plant in New York done by 2012 but apparently they are using some new tools in other existing plants. Those ARM chips will be great for small applications and they will be small and use very little energy and heat. Nice find.
Score
0
February 24, 2010 7:35:54 PM

Whats interesting here is, from 32/28 on down, all GF products will have HKMG, which really brings the power advantage even with Intel
Score
0
Related resources
February 24, 2010 7:47:11 PM

And what would HKMG be? Something with heat and power?
Score
0
February 24, 2010 8:06:02 PM

Oh. Cool.
Score
0
a b à CPUs
February 24, 2010 9:16:16 PM

I'm pretty sure that the process used for ARM and other low power devices is different from the process for high power devices. Intel's been on 28nm with their flash (for their SSDs) for a while for example. Intel is maintaining their process lead just fine from all current indications - it appears that AMD will not be putting out a 32nm CPU until the beginning of 2011 at the earliest.
Score
0
a b à CPUs
February 24, 2010 9:35:21 PM

cjl said:
I'm pretty sure that the process used for ARM and other low power devices is different from the process for high power devices. Intel's been on 28nm with their flash (for their SSDs) for a while for example. Intel is maintaining their process lead just fine from all current indications - it appears that AMD will not be putting out a 32nm CPU until the beginning of 2011 at the earliest.



+1

Nothing to see here people. Just more spin from JDJ. Move along.
Score
0
February 24, 2010 10:02:47 PM

This is the nonsense everyone see from you JDJ you have no balance,
your green through and through and sometimes you just can't hide it.
Score
0
February 24, 2010 10:27:04 PM

cjl said:
I'm pretty sure that the process used for ARM and other low power devices is different from the process for high power devices. Intel's been on 28nm with their flash (for their SSDs) for a while for example. Intel is maintaining their process lead just fine from all current indications - it appears that AMD will not be putting out a 32nm CPU until the beginning of 2011 at the earliest.


Yep yep, they are on the 45 now, till they get the 28 its gonna be in no less than 10 years!
Score
0
February 24, 2010 10:32:01 PM

All Jaydee was post a link, some info and a question. Nothing to get flamed about.
Score
0
a b à CPUs
February 24, 2010 10:57:25 PM

Nice pickup jd ... I see you have been over visiting Charlie too.
Score
0
February 25, 2010 12:39:23 AM

So just because Jaydee posts a link about a company (spun off by AMD fyi) actually producing 28nm cpus in h2 he is a fanboy? Wow... anything that can be slightly better than Intel is bs huh?
Score
0
February 25, 2010 12:41:22 AM

Yea, but its not where I got the info from.
Interesting no doubt. Too bad its not just memory either. Tho Im sure therell be cache involved somewheres within all those trannys.
Having a 28nm cpu counts too doesnt it? Or is that only for certain makers, like AMD ot Intel?
Score
0
February 25, 2010 12:43:10 AM

My point here is, people are saying that GF is losing ground, when they may have the first 28nm, or smallest cpu node to date by then end of the year.
Whens Intels 22 coming?
Score
0
a b à CPUs
February 25, 2010 1:18:18 AM

JAYDEEJOHN said:
Yea, but its not where I got the info from.
Interesting no doubt. Too bad its not just memory either. Tho Im sure therell be cache involved somewheres within all those trannys.
Having a 28nm cpu counts too doesnt it? Or is that only for certain makers, like AMD ot Intel?

It counts, but as I said, I believe it is a slightly different process to make things like flash memory and ARM cpus compared to things like regular CPUs and GPUs. That's also why AMDs 32nm is behind this 28nm stuff (and you can't really compare the different process to AMD or Intel's progress on 32nm or 22nm).
Score
0
a b à CPUs
February 25, 2010 1:22:42 AM

JAYDEEJOHN said:
Yea, but its not where I got the info from.
Interesting no doubt. Too bad its not just memory either. Tho Im sure therell be cache involved somewheres within all those trannys.
Having a 28nm cpu counts too doesnt it? Or is that only for certain makers, like AMD ot Intel?



Trannys scare me, LMAO.
Score
0
February 25, 2010 1:30:02 AM

Actually, the main difference between cpu/gpu etc is the qualifying/end process. It takes forever for a cpu to pass, whereas a gpu can much quicker.
Thinking youd need a cpu to be more exact on a puny lil chip like a i3 or whatever compared to a 3.2 billion tranny chip (down boys) well, just the size alone tells you youd better know what youre doing using 3.2 billion.
How many transistors does a 920 have? And dont count cache either, as thats most of what most of a cpu is, isnt it?
Score
0
a b à CPUs
February 25, 2010 3:43:21 AM

Cache counts - it uses transistors as well, and is just as prone to defects.

The Bloomfield chips use around 850 million IIRC, with Lynnfield being slightly more.
Score
0
February 25, 2010 3:58:46 AM

So, comparing a part thats a 40% the size of a large gpu, and to say is far superior, well, count yields and have no problems, it is important to have a great process, and dont be fooled by hype.
Its not exactly the same, but looking at LRB and how its been delayed played a part in all this, and it was held back on an old process as well.
Even if its repetitive, like LRB or any die made, the larger it is, the harder it is to make.
So, making something at least 2 1/2 times larger on a smaller process is something.
Its surely more than "theyll never catch up, and they cant either" , words heard here.
Thing is this, Intels direction is driven only by Intel. Theres businesses that wants lower nodes, unlike Intel, as you mentioned flash.
Now, its money that drives this, not Intels lead per se.
GF has all the money it needs, so the "theyll never catch up" thing is bogus, and this is the first we will see of this, with more coming.
To say Fermi doesnt require a super process at 3.2 billion trannys, having ECC and being made for HPC is ridiculous, and biased as well.
When the gpgpu units come along, itll be reaching into cpu segments, and is the primary reason for LRB, so this "its not a cpu" argument fails right there
Score
0
a b à CPUs
February 25, 2010 4:02:21 AM

GPU processes are difficult, and yes, they can be (in most ways) considered similarly to a CPU process (though there are slightly different considerations due to the greater number of transistors and lower clock speeds in GPUs). What's not as comparable is something very small and low power, like a FLASH chip or an ARM CPU.
Score
0
a b à CPUs
February 25, 2010 4:19:08 AM

Another troll thread from AMD fanboy.

Get another non-mod account and see if others will treat you as nice when you post something like this.
Score
0
February 25, 2010 4:22:11 AM

Remember, ARM needs to talk, as much as it can, it has no real competitors in its own niche per se. Now, having a 2.5 billion tranny gpu coming out early next year at 28nm sounds good as well?
Possibly by the end of this year as well.
Understand what Im saying here. Intel drives Intel, and if it doesnt fit its tic tock, it wont happen, theres no need, they dont want a Osborn effect.
Not so with GF and its customers, as each of those customers have their own particular needs, much more diversification than Intel, and so too nodes requirements.
Im just trying to point a few things out again I think people have missed, like I occasionally do heheh.
The reasons why theyre usually missed is because of fan blindness. Dont know if its kool aid or what, but its all business, not "Intels business, then the rest".
So, we know GF has more money potential for R&D than Intel, has a wider need for a greater diversification of nodes, be they low power, high speed or anything in between.
These requirements may push GF past Intel at some point in process, and certainly catch up to them, if needed.
ARM needs these now, the ATI chips are a mystery for competitions sake, so we dont know their release window.
No crazy arguments here, just real money making potential here, and money talks, it always has

PS Your cache defects can be huge with little effect on the chip overall, where these change with the rest of the cpu or gpu as well
Score
0
a b à CPUs
February 25, 2010 6:13:09 AM

If they can get a 2.5 billion transistor GPU working on 28nm this soon after the problems with 40nm (which was actually pretty much equivalent to 45nm in gate pitch), I'll be pretty surprised. I'd absolutely love to see it, but I don't expect it to happen.
Score
0
February 25, 2010 6:31:43 AM

TSMC is having all sorts of trouble, as pointed out:
Ability to Ramp and Time-to-Volume: Leakage
emerged as a major yield limiter at the 45/40nm
node, preventing other leading foundries from
ramping production (See figure 2).
This painfully underscored the fact for foundry
customers that a risk production date on a roadmap
is one thing, ramping production is quite another.
Risk production was offered by the other leading
foundry in October 2007, yet two years later yields
remain insufficient for what should be a mature
technology. In contrast, GLOBALFOUNDRIES
surpassed the 25K per quarter wafer shipment
mark with high yielding devices in less than 3
quarters. GLOBALFOUNDRIES supports a leader
in x86 CPU’s that ramps large complex high
performance die to drive low defect density with
high manufacturability and yields. Initial 32nm CPU
products will be sampling in 2010. The 32nm HKMG
ramp will precede the 28nm HKMG ramp of other
GLOBALFOUNDRIES customers by about one
quarter.
http://www.globalfoundries.com/pdf/GF_HKMG_TrifoldBroch...

This is a solid poke at TSMC, and they deserve it, tho, GF had better come thru as well
Score
0
a c 172 à CPUs
a b å Intel
February 25, 2010 8:08:54 AM

JAYDEEJOHN said:

So, we know GF has more money potential for R&D than Intel, ...

How do you figure?
Score
0
a b à CPUs
February 25, 2010 10:36:21 AM

The only reason GF isn't providing 32nm SOI right now is due to AMD not being ready for it.
Score
0
February 25, 2010 10:59:20 AM

Here's the problem with your MONEY will make everything better thinking.
1) P4 intels chip wasn't the best of chips and intel has the money.
2) Phenom 1 with the help of GF money didn't make it a good chip either.
Now GF haven't made anything as of yet that would make anyone believe
they are the best at making cpu's. JUST ALOT OF BLIND FAITH ON YOUR PART JDJ.
Score
0
February 25, 2010 11:35:20 AM

Like I said before GF throw 750million at phenom did it help?
Score
0
a b à CPUs
February 25, 2010 12:11:52 PM

No, they didn't.

How many billions did intel throw at Larrabee btw? Anyway why are you arguing architecture when it's about process?
Score
0
February 25, 2010 12:23:48 PM

If this would have been titled "Is GF ready to take a big swipe at TSMC?" I doubt anyone would have argued the positive impact going forward. The process for flash memory and ARM chips I would imagine is apples and oranges to CPU's. The title suggested that they were in a process lead in regards to Intel even though they have spawned a gazillion variants of their broken chips at the 45nm process under different names, different core counts and (admit it Intel guys) very aggressive prices.
Score
0
February 25, 2010 12:29:59 PM

JDJ is saying GF will be ahead of Intel because they have the money, I'm saying
that's not true at all.
Score
0
a b à CPUs
February 25, 2010 12:33:21 PM

Intel still have the process lead as they can pack more cache transistors into a smaller area, switch them faster, have less latency, and they have less leakage at idle, and under load.

AMD can design them, but can't quite execute at the process end, thus their product is still a bit less powerful ...

Argueably Intel's overall design is still slightly more efficient and therefore has a higher IPC too.

They burned out a lot more engineers designing the i7 though ... the R&D budget is much higher.

There is no shame for AMD in that ... in fact their response is to chase more cores rather than making a more efficient processor in the interim.

Providing they can continue to shrink the process they will still progress ... to a point of diminishing returns ... until they do some serious work on their cores.
Score
0
a b à CPUs
February 25, 2010 2:12:30 PM

This whole thread is yet another JDJ speculative "AMD gonna do dis" or "AMD gonna do dat" thread, except now for GF. I have previously posted links showing that GF might find it harder to do HKMG than what JDJ suspects. At any rate, the facts are that right now Intel is shipping CPUs with 2nd gen HKMG on 32nm, whereas AMD/GF are shipping 45nm CPUs with zero gen HKMG (i.e., none), of which yields seem to be significantly poorer compared to Intel's due to the high availability of cripple cores.

And isn't TSMC still manufacturing 100% of AMD's GPUs?
Score
0
a b à CPUs
February 25, 2010 2:19:45 PM

Fail fazers.

AMD are shipping 12 core server cpu's at 45nm while intel can't even make 8 cores at the same clocks. The reason the Phenom II X2's exist is for market segmentation, everybody knows these are quads just waiting to be unlocked in the bios.

Of course you were the one who said AMD would be 18 months behind intel moving to 32nm, right? :D 
Score
0
February 25, 2010 2:35:26 PM

OK, thanks all for the laugh. jennys the only one of the last few posts thats gotten this at all.
It has absolutely NOTHING to do with AMD, other than maybe a few of their products.
Its not about arches, its about process, reread the OP.
It isnt about AMDs arch etc, or AMDs fabs, its GF going to 28nm early, making them the lowest node available.
Its about any, umm any? company that wants to push forwards, whether its cpus, gpus or whatever we use heheh
I dont care about AMDs usage, nor Intels, this is about 28nm, and last time I checked, its smaller than 32nm.
The R&D factor can easily be seen as a no brainer here, if GF needs a bump outside their already competitive investments, daddy ATIC will be there, oh, and by the way, ATIC could buy Intel several times over and still be looking good, so no, money isnt a problem here.
As Ive said, its about what the customer wants. If they want more, and Intel just isnt there, and doesnt want to create an osborn effect, they wont be either, if someone wants a smaller process, and enough do, itll happen.
Acting like ARM or 28nm HKMG isnt important is not sensible, or doesnt make sense, since this is what ARM wants, and we will see what a superior process with HKMG does for their march into their particular markets.
If their product turns out to be very desirable, it effects other businesses as well, which may also want in on this as well, or even something more. Not just Intel roadmap, but many, thats the point here, and will drive GF to be very competitive
Score
0
a b à CPUs
February 25, 2010 2:50:33 PM

jennyh said:
Fail fazers.

AMD are shipping 12 core server cpu's at 45nm while intel can't even make 8 cores at the same clocks. The reason the Phenom II X2's exist is for market segmentation, everybody knows these are quads just waiting to be unlocked in the bios.

Of course you were the one who said AMD would be 18 months behind intel moving to 32nm, right? :D 



This is a lie and you know it. The reason x2's exist is to increase the yields. If it was pure market segmentation ALL x2's/x3s would be unlockable and fall into the bin qualifications of its x4 counterpart.
Score
0
a b à CPUs
February 25, 2010 3:51:58 PM

jennyh said:
AMD are shipping 12 core server cpu's at 45nm while intel can't even make 8 cores at the same clocks.


IIRC Beckton is for the 4P & higher market, is monolithic - not "glued together double cheeseburger" MCM, and has a larger number of QPI interconnects, so it is a far more complex CPU than the Magny Cours, which AMD admits is just stop-gap until BD comes out.

Quote:
The reason the Phenom II X2's exist is for market segmentation, everybody knows these are quads just waiting to be unlocked in the bios.


LOL - sure, Jenny, sure - all those X2's & X3's are unlockable, and "market segmentation" is obviously the reason as AMD clearly loves to waste SOI wafer space since it only costs on average 30% + more than the strained silicon that Intel uses. :p 

And of course all those "But my X3 won't unlock" threads here on THG are just spintel paid pumper lies to discourage bottom-feeders from buying crippled CPUs :D .

Quote:
Of course you were the one who said AMD would be 18 months behind intel moving to 32nm, right? :D 


Pretty much, although that was some time ago when AMD's roadmaps were saying BD would be their first 32nm CPU and wouldn't appear until 2H of next year. Of course, AMD roadmaps are far less reliable than, say, Rand-McNally's, so who knows at this point. What is known, however, is that GF is saying one thing about HKMG gate-first on their website, obviously to attract customers, and yet complaining to IBM that gate-first is not as good as Intel's gate-last and wanting to switch at 22nm, if not sooner...

Hmm, which to believe - marketing crapola or questions from the design engineers??
Score
0
February 25, 2010 3:56:29 PM

Actually, its a combination, filling segments and offering more, good/positive marketing costing little if nothing, plus filling the segment.
To me, a brilliant move on AMDs part
Score
0
a b à CPUs
February 25, 2010 3:59:42 PM

someguy7 said:
This is a lie and you know it. The reason x2's exist is to increase the yields. If it was pure market segmentation ALL x2's/x3s would be unlockable and fall into the bin qualifications of its x4 counterpart.


Wrong. Dual cores sell MUCH more than quads, and AMD doesn't have any Phenom II part except for the X4. There is no native dual core Phenom II, these parts can only be made from X4's. A very small minority of X2's will not be unlockable, but it will be a *very* small minority now.

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=373...

Quote:
The Phenom II X2 is nothing more than a Phenom II X4 with two cores disabled. Originally these cores were disabled because of low yields, but over time yields on quad-core Phenom IIs should be high enough to negate the need for a Phenom II X2. This is most likely why AMD removed the Phenom II X2 from its official price list. It's also why the stranger Phenom II derivatives are also absent from AMD's price list. All that's left are Phenom II X4s pretty much.


AMD's 45nm is well over a year and a half old and yields will have been in the 90%+ for months. They still have to sell dual cores, that means disabling an X4.
Score
0
February 25, 2010 4:02:53 PM

Wonder how long itll be before Intel copies GF and AMD before they go SOI?
Score
0
a b à CPUs
February 25, 2010 4:15:25 PM

According to GF's brochure

Quote:
In contrast to the unity of GLOBALFOUNDRIES and its partners, the so-called ‘Gate Last’ approaches are not unified. For example, a leading
integrated device manufacturer of x86 CPU’s has two widely different approaches at 45 and 32nm.
At 45nm the High K dielectric and initial layer of gate metal are deposited BEFORE the dummy poly; this process is truly a hybrid of ‘Gate First’
and ‘Gate Last’.


GF moving to gate last? I really doubt that. They also said they were having none of intels 22nm issues, which makes me think that intel will be switching to GF's methods at 22nm. :) 
Score
0
a b à CPUs
February 25, 2010 4:43:08 PM

jennyh said:
Wrong. Dual cores sell MUCH more than quads, and AMD doesn't have any Phenom II part except for the X4. There is no native dual core Phenom II, these parts can only be made from X4's. A very small minority of X2's will not be unlockable, but it will be a *very* small minority now.

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=373...

Quote:
The Phenom II X2 is nothing more than a Phenom II X4 with two cores disabled. Originally these cores were disabled because of low yields, but over time yields on quad-core Phenom IIs should be high enough to negate the need for a Phenom II X2. This is most likely why AMD removed the Phenom II X2 from its official price list. It's also why the stranger Phenom II derivatives are also absent from AMD's price list. All that's left are Phenom II X4s pretty much.


AMD's 45nm is well over a year and a half old and yields will have been in the 90%+ for months. They still have to sell dual cores, that means disabling an X4.


No. You are wrong like usual.

First of all. There are Phenom II's that are not x4. AMD Phenom II X2 555 Deneb 3.2GHz 2 x 512KB L2 Cache 6MB L3 Cache Socket AM3 80W Dual-Core Desktop Processor - C3 Revision - Retail. Yes they all are originally intended to be X4's but they are sold as X2's and x3's and they are Phenom II's. So to say it is not a Phenom II part is just silly. It is not a native dual core or whatever but it is a Phenom II part. And it is branded as so.

It is not a very small minority either. All you got to do is look on the forums. I do not know what you call a very small minority so it could be all subjective. But with you it never is. It is just more AMD FUD.

Of course the yields have got better. And it is "harder" for AMD to actually fill the market segment with the chips that failed. I am not disputing that. I am not even disputing that AMD is taking some perfectly good x4's and disabling cores to meet the demand. This is my problem. You said this "The reason the Phenom II X2's exist is for market segmentation, everybody knows these are quads just waiting to be unlocked in the bios." And that is flat out lie. In your response to me you have proved your original statement false with this "Originally these cores were disabled because of low yields," from the anand article. Again the chances of unlocking the newest X2's are greater then in the past no doubt about it. But it is not very unlikely that a chip will not unlock and be stable.

And now you say that AMD has a 90 percent yield on its X4's? What dream world did you pull that number from. Just stop it already with all the lies and FUD.


You just can't be happy that the tech company that you love has a good product. You still have to exaggerate and make stuff up. It is rather pathetic.

Score
0
a b à CPUs
February 25, 2010 5:12:28 PM

What kind of yields do you think these are at? If TSMC can yield Cypress at 80% at a size of 334mm2 on their garbage 40nm, what makes you think GF can't yield Phenom II at 258mm2 a lot higher?

Get a clue please someguy.
Score
0
February 25, 2010 5:20:24 PM

At this stage, nearing 90% isnt unreasonable, give or take a few % points, which touted as either a loss of opportunity for AMD, as some have claimed, or brilliant option, as Ive said, in reality, it all comes down to margins.
If AMDs margins are decent for them doing this, whats the problem? Anybody doesnt like purchasing a dual core with a good possibility of getting 3 or 4 cores operational?
Knowing its a roll of the dice heheh, its not a bad marketing approach
Score
0
a b à CPUs
February 25, 2010 5:25:21 PM

JAYDEEJOHN said:
Yea, but its not where I got the info from.
Interesting no doubt. Too bad its not just memory either. Tho Im sure therell be cache involved somewheres within all those trannys.
Having a 28nm cpu counts too doesnt it? Or is that only for certain makers, like AMD ot Intel?


Using that logic, Intel/AMD are falling behind in CPU development in general, because of Azul Systems 54-core processor, Sun Microsystems has a 8 core 64 thread processer, etc.

Nothing to see people, move along.
Score
0
a b à CPUs
February 25, 2010 5:42:01 PM

gamerk316 said:
Using that logic, Intel/AMD are falling behind in CPU development in general, because of Azul Systems 54-core processor, Sun Microsystems has a 8 core 64 thread processer, etc.

Nothing to see people, move along.

What about Intel's 80 core, 1TFLOP, 65nm CPU that they developed for research purposes in 2007? It only used 62 watts too. Clearly, everyone else is falling behind...

As for 22nm? Intel has a working 2.9 billion transistor test chip (and that was back in September). I don't see anything remotely comparable from AMD/GF.
Score
0
February 25, 2010 5:43:56 PM

So, how many people will use your example compared to mine? Thought so.
How many segements will this 1 chip compete with AMD/Intel? Thought so.
How many other examples qualify for competition of which were refering to here? thought so.
Name 1 gpu for instance? thought so.
Name 1 net top etc.
No, money talks, and if the customers demand it, itll put the squeeze on Intel despite their tic tock approach, regardless.
Thats my point here.
Is it possible no one will want these lower nodes quicker? yes, but it isnt what we see with things going forwards.
You bring in a 1 solution fits all to discredit whats happening here, and are ignoring whats happening here.
Your solution not only doesnt fit, lets talk price for example eh? Or just how people tend to use your scenario over, say, Atom? Or the next gen gpus? Or who knows what?
Money will drive the market, it always has. Iff the demand is there, GF will catch Intel, no "lets be stupid and hide our heads in the sand" will work.
Everyone thats a Intel fan has stepped out and tried many differing ways as to how GF is non consequintial, or wont come up with monies, or will only fall further behind, or their approach is bad etc.
All of this doesnt matter, as the customers will drive the market as they want it, despite Intel and its current pace.
Score
0
a b à CPUs
February 25, 2010 5:52:44 PM

This will be GF's/AMD's first ever half-node too. Pretty impressive stuff from GF, they are clearly going to be #1 foundry for the forseeable future.
Score
0
Related resources
!