HKMG overated?

[JAYDEEJOHN]

I find the latest rumored speeds of the upcoming AMD cpu (Debeb) intriguing, not just because hopefully AMD may have a new, fast chip coming, but its NOT using HKMG, and the rumors are the chip can go to 4Ghz on air. I can remember as far back as 2006, when certain members here, and some whove left, ascertained the HKMG mantra, at 45 nm and less, youll have to have it, or youre voltage wont be controlled within the chip, and youll have leakage galore.


Well, this seemed to be confirmed, as Intel was humming away with their new 45nm quads and duals, using HKMG, while AMDs 65 quads ate power, abd the old K8 duals did as well. But now, it would seem if the Deneb chip can do this, and from whats also been rumored, and more to fact with Shanghai, the power usage and thermals both look great, showing very low leakage, which goes against everything Ive read and heard about HKMG. I know it helps, but I feel it JUST helps, and has been overhyped, and also used against AMD, IBM etc in various forums and site writeups. Heres a link that shows some interesting info http://www.eetimes.eu/semi/212002481?pgno=3

Quote:" How did AMD reach 45 nm without resorting to HKMG? Actually, the first 45-nm device on the market was Matsushita's UniPhier, and it also used conventional poly plus oxy-nitride dielectric. Just like AMD, Matsushita brought immersion lithography online to meet the scaling requirements of 45 nm. In fact, UniPhier boasts the tightest metal pitches we have seen in a logic chip at 138 nm. The transistor performance requirements of the UniPhier design were lower than a high-performance server processor. The end game for Matsushita was basically circuit density"

So, while Intel certainly does employ nice processes in thier cpu's, they arent the only ones whore innovative, but to me, it seems they ARE the ones that get all the credit, even for things that apparently dont amount to much at 45nm

 

[JAYDEEJOHN]

For those that dont or didnt read the article, one of the most interesting quotes :" The transistor drive current for AMD's 45-nm devices is much lower than that of the Intel HKMG transistors. But power consumption is quickly becoming a high priority for server chips. AMD's transistors exhibit very low channel leakage. Our transistor benchmarks indicates that leakage current is less than one-third of the value measured on AMD's 65-nm process. It's also significantly lower than the Intel 45-nm HKMG process. In fact the Ion/Ioff ratio for AMD's PFET is nearly 10 times better than that for the Intel PFET. "

This is significant, and shows AMD was serious when they said 65nm would be a quick stopover, and they planned 45 to be their main focus, which is obviously looking like it.

When AMD does finally get around to using HKMG, we will see even greater performance from them, which should not only lower power usage, but possibly show even greater ocing abilities as well. 5Ghz on air anyone?

 

[blackwidow_rsa]

well with ibm behind the wheel you can only suspect good things, they do everything...

 

[JAYDEEJOHN]

True, but it wasnt IBMs decision to forego HKMG, and still show a great cpu. IBM still isnt using HKMG either, and its beeen construed as a weakness, like they cant do it. Im saying, its possibly too early to use it, just like when Intel said going to a native quad at 65nm wasnt a good idea, maybe using HKMG at 45nm before youve gotten everything out of youre current arch shouldnt be used, but it looks like AMD has done just that finally, and even without it, its looking pretty good, and should look even better with it

 

[piesquared]

Nice post JDJ. You gotta believe AMD's 45nm tech with their HKMG gate first approach is going to pay dividends. Like BlackWidow mentioned, IBM and the alliance are a force to be wreckoned with. That process, along with AMD's own sauce, is pretty amazing, all things considered.

As for their current tech, I think the writing was on the wall, it's just that Intel's marketing machine overwhelmed the senses, and put theirs on center stage.

 

[JAYDEEJOHN]

Theres rumor that PhenomII is another R4xxx series surprise, and from where I heard it from, I hate saying this, but I may have to agree with em, just because of who said it

 

[zenmaster]

However, we know all the rumors are false.........
How? AMD tells us so.............

Simply Put, If the AMD Chips could run that fast that cool.....
They would be selling them at higher not lower speeds than the current chips.

This way they could charge a price that turns a profit and not a loss.

But hey, what does AMD know about their own chips.
AMD is just an Intel Fanboy to make such claims about their own product.

 

[piesquared]

I think back of how many times in the past that people have written AMD off, and they continue to fight back time and time again with amazing new tech. They refuse to be written off.
Imagine if AMD somehow managed to put Istanbul on 45nm HKMG w/sse5.

 

[zenmaster]

True, but it wasnt IBMs decision to forego HKMG, and still show a great cpu. IBM still isnt using HKMG either, and its beeen construed as a weakness, like they cant do it. Im saying, its possibly too early to use it, just like when Intel said going to a native quad at 65nm wasnt a good idea, maybe using HKMG at 45nm before youve gotten everything out of youre current arch shouldnt be used, but it looks like AMD has done just that finally, and even without it, its looking pretty good, and should look even better with it

You do realize that AMD will be shipping the first 45nm Quads at lower speeds than their current 64nm parts due to leakage and that the faster parts will not ship until AMD upgrades to HKMG?

But hey...............
What does AMD know about Phenoms.........

They should really read the rumor boards so they would now what speeds their chips run at............

 

[JAYDEEJOHN]

And why go over 3Ghz? How many quads at real peoples prices are over 3Ghz? That doesnt make sense. Theyd price themselves right out of contention. This is like the 4xxx series, remember? Compare the 4850 to the GT, remember the undersell, the low intro pricing? Maybe you dont, but I do, and its been referred to as a possibility, and going over 3Ghz would go against the grain of this style of approach, simple as that

 

[JAYDEEJOHN]

Obviously, you didnt read my link. Start from there, then go forwards

 

[NMDante]

A quick question, jaydeejohn.

If HKMG is overrated, why is IBM and AMD still planning on using it at 32nm?

 

[JAYDEEJOHN]

How many times have we heard that Intel is the best because it uses HKMG, and without it, itd be impossible to keep leakge down? To hit ANY high clocks at stock? Have ANY ocing abilities? How IBM was waaaay behind Intel in not having it? AMD as well?

TSMC isnt going to use it anytime soon as well, and they were what? Put down once again for? Not having HKMG. Now if that isnt overhyping something that appears to have a lessor usage that ALL THOSE CLAIMS weve been hearing the last 2 years, I dontr know what is?


Now, will it help? Yes. But to get alllll the acclaim that Intel got, and allll the putdowns IBM, TSMC and AMD got for not having it shows its been overhyped. Where did I say it wouldnt be used? Where did I say its useless? In another month or two, maybe longer, Intels going to lose the process lead to TSMC, and guess what? They arent using HKMG.

 

[Just_An_Engineer]

Simply Put, If the AMD Chips could run that fast that cool.....
They would be selling them at higher not lower speeds than the current chips.

Intel's chips are capable of running at much higher clock speeds without increasing voltage or requiring exotic cooling yet I don't see you complaining about them not releasing them with higher stock speeds. Why should we expect that it be any different with AMD? Just because they're not Intel?

 

[Just_An_Engineer]

You do realize that AMD will be shipping the first 45nm Quads at lower speeds than their current 64nm parts due to leakage and that the faster parts will not ship until AMD upgrades to HKMG?

But hey...............
What does AMD know about Phenoms.........

They should really read the rumor boards so they would now what speeds their chips run at............

You do realize that the 45nm Shanghai clocked at 2.7Ghz that was just released last week is the highest clocked AMD quad released to date right?

 

[NMDante]

How many times have we heard that Intel is the best because it uses HKMG, and without it, itd be impossible to keep leakge down? To hit ANY high clocks at stock? Have ANY ocing abilities? How IBM was waaaay behind Intel in not having it? AMD as well?

TSMC isnt going to use it anytime soon as well, and they were what? Put down once again for? Not having HKMG. Now if that isnt overhyping something that appears to have a lessor usage that ALL THOSE CLAIMS weve been hearing the last 2 years, I dontr know what is?


Now, will it help? Yes. But to get alllll the acclaim that Intel got, and allll the putdowns IBM, TSMC and AMD got for not having it shows its been overhyped. Where did I say it wouldnt be used? Where did I say its useless? In another month or two, maybe longer, Intels going to lose the process lead to TSMC, and guess what? They arent using HKMG.

Whoa, you getting way too emotional over this.
First, where did anyone claim that HKMG was the deciding factor for anything? I don't remember anyone claiming such a thing. In fact, the only time HKMG was mentioned, was when detailing how Intel used it to lower power leakage and have better transistor performance at 45nm.
Second, who claimed you said it wasn't going to be used? Or said it was useless? I look at all the replies, and no one said anything to the effect.
Third, you claim TSMC will take the process lead from Intel? In what? What process are you talking about? You know, TSMC and Intel use different processes, right?

As for your overhyping...it seems like you are the only one who is claiming any sort of hype for HKMG. It helped Intel at Intel's 45nm node, and will continue to 32nm. IBM was going to use it at 45nm, but pushed it to 32nm. Why? Who knows? Maybe IBM is still looking for a way to use it with SOI, along with bulk. And because of this push, AMD did not use it at it's 45nm node.

So, who exactly hyped HKMG the way you care claiming?

 

[JAYDEEJOHN]

I know you were here then, its been hyped that way even not so long ago. If you dont remember, you can look up all those posts, as back then I was assured, as many others, that HKMG was NEEDED, and ONLY Intel had it or was going to have it. Ill let my, and others own memories decide this, as I truly dontr feel the need to dig up allll those posts


As fas as TSMC, the process node will soon be ahead of Intels, and will stay there. Theyre planning on using HKMG at 28nm if I recall, and not til then, so again, overhyped as to its performance/usefullness

 

[NMDante]

Again, jaydeejohn, I don't remember anyone hyping HKMG, as much as you are.

It was mentioned that HKMG was going to be needed after 65nm, since power leakage would be substantial (theorized) at 45nm and lower, without it. That was a theory. IBM was going to use low-k MG at 45nm and go to high-k later. Again, if you believe high-k to be overrated, why would IBM consider using it, if it was not needed at all?

I kind of understand where you are going, that AMD's 45nm doesn't have high-k, and it seems to be doing fine. If you are basing it solely on the Shanghai release, that's fine. But Barcelona power consumption was not bad, even at 65nm. And looking at the "rumored" Phenom II 940's TDP of 125W, that is still pretty high, considering that the 9950 also ran at 125W.

If AMD had the opportunity to use HKMG at 45nm, I'm sure they would. The problem is, they are dependent on IBM's SOI process changes. So, when IBM pushed it out until 32nm, AMD couldn't do anything, and had to use what was available to them, which was SOI.

Even the article you linked mentioned that AMD was squeezing every last drop of performance from SOI (not a direct quote).

If Deneb shows good performance and power consumption, without HKMG, that is good. But until independent tests are ran, it's hard to tell if AMD's 45nm will live up to the "rumors".

 

[JAYDEEJOHN]

At the time, it wasnt espoused as theory, but fact only. It was rubbed in the face of anyone that didnt have it. It was being told as being NEEDED, not as thoery, but as fact, and AMD was laughed at for not having it, and itd never work without it. JJ, you, alot of others were saying this. You can go to a certain site, and find a whole troop of people who said these things. Now, when reality comes around we see its not a Godsend, but a nice addition thats NOT needed, and theres other alternatives out there. Liquid immersion shows this.

Id go as far as the link goes, and add that Intel had already squeeezed everything out of thier process, and thus went to HKMG, but thats not what this threads about. Nor is it about theories. Nor about HKMG being useless. The past portrayal of its usefullness has been scattered with total ineptness, and those purpoting it, and went along with all of it, now see that it was hype, and if they dont, well thats something I wont comment on.

It just seems, whatever wheel one can bring to the wagon is used, and HKMG was used in a big way. Im just exposing this, as it happens alot, and even tho what we hear and what we see may be helpful, it may not always be fully truthful. I say, it was bandwaggonning,overhyping and a true misunderstanding of whats REALLY going on that caused this. What caused that, I leave to someone else to say

 

[zenmaster]

Intel's chips are capable of running at much higher clock speeds without increasing voltage or requiring exotic cooling yet I don't see you complaining about them not releasing them with higher stock speeds. Why should we expect that it be any different with AMD? Just because they're not Intel?

That is because they don't need to do this.
They leave headroom so they can release new processors at higher speeds and then bump processors down the line.

And not all of Intel's processor are that low.

Example - The Extreme i7 Ships at 3.2 Ghz w/ guarenteed auto OC to 3.35 Ghz.....

AMD will be shipping 2.4-2.7ghz with a lower IPC

If their chips could hit those speeds on simple cooling, they would realease them and sell them at higher prices.
They would actually sell a combination of low margin and high margin chips.

This is what they did when they were actually competitive.
It's not that hard.

And as the article is talking about server chips, these generally can't be OC'd.
So if you wanted to go AMD, you would be hurting themselves more.
If somebody wanted the horse-power, they could not get it.

No way AMD would refuse to provide a processor that could meet the High Margin sector if it was simply a matter of giving it to customers.

You may think AMD is stupid, but I don't think they are that stupid.

 

[JAYDEEJOHN]

No way would a 299$ gpu come against a 700$ gpu....ooops

 

[Just_An_Engineer]

I know you were here then, its been hyped that way even not so long ago. If you dont remember, you can look up all those posts, as back then I was assured, as many others, that HKMG was NEEDED, and ONLY Intel had it or was going to have it. Ill let my, and others own memories decide this, as I truly dontr feel the need to dig up allll those posts

I remember several threads claiming that SOI @ 45nm would be a disaster. You have to go way back to around March to find them however. Unfortunately the search function on this forum sucks royally so I was only able to find a couple of threads.

Thread full of AMD and SOI bashing by the usual suspects: http://www.tomshardware.com/forum/248409-28-fuddo-45nm-scheduled-launch-2009

Thread dispelling some of the rumors: http://www.tomshardware.com/forum/249064-28-tomshardware

 

[JAYDEEJOHN]

TY, as I really do hate to point out the obvious when others are in denial, and even what I just typed sounds harsh, but it is true, and thats what this threads about

 

[Just_An_Engineer]

That is because they don't need to do this.

I suspected that you would go with the typical fanboy response....

They leave headroom so they can release new processors at higher speeds and then bump processors down the line.

Then why hasn't Intel ever released a Core 2 processor clocked higher than 3.33Ghz when several models have been widely reported to be able to OC to nearly 4Ghz using conventional cooling?

 

[someguy7]

Because those chips have no competition from AMD. Both Intel and AMD have done this. When Intel was getting its behind smashed with netburst they kept pushing the clock rates higher. Which in turn made those machines space heaters. At the time AMDs fastest X2s where not out. They didnt need to be. People where overlocking up to the same speeds back then. Once AMD lost the performance crown to conroe they had to release chips that pushed the cpus/process very limits(x2 6400)

If AMD had come out with K10 on time and it beat Intels cpus you best believe Intel would of jacked up the GHZ on the core2s.

These new 45nms from AMD better smash Intels if they plan only releasing 3ghz parts. While thoae same chips can OC to 4ghz on air without extreme effort/power/luck. Otherwise its just plain retarded. If you got the headroom on your cpus and your cpu is losing to its competitor and you dont clock it up than AMD should stop making computer parts and run a short bus company.