HKMG overated?

Page 5 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar


I guess you now agree that 45nm SOI won't work well for them.

Funny, how I was not one of the people supposedly to claim that AMD needed HKMG (in any of those linked threads), but yet I'm seen as one of them now.
 
Yep.

CPU-Z often misreads AMD CPU voltage. It would read 1.6 volts on my 9850 no matter what voltage it was set at... so it's probably just reading the wrong sensor.
 
Also, on the original topic, it makes sense for AMD to continue developing new SOI processes, at least in the near term. They have been trying to transition their GPU's to an SOI process, and they are looking to put those "cores" on CPU dies with the whole Fusion architecture drive. So, continuing with SOI now will allow Fusion to debut sooner rather than later.. and with more current processes.

I have also read that IBM's involvement in the process R&D is a double edged sword. From what I understand, IBM comes up with rather crude processes for low volume production, and AMD spends most of the time perfecting the processes they get from IBM... hence the delays in transitioning to 45 nm (which they have had for a while). Right now it appears 32nm and 22nm are in the works at various stages. Hopefully AMD has learned something about their struggle with IBM's 45nm design and can roll out the next processes quicker.
 


No. I previously had doubts 45nm SOI would work well for them. Those doubts appear to have been misplaced.


Unlike some [hint hint], I can admit when I may have been wrong.



Funny, how I was not one of the people supposedly to claim that AMD needed HKMG (in any of those linked threads), but yet I'm seen as one of them now.
Click to expand...

Quit acting like a 5 year old. You know damn rightly why all the critics were of the view AMD's process would not be up to the quality of Intel's.



I also notice you refuse to answer my question, so again:

Since 45nm SOI wasn't going to cut the mustard in eyes of the THG forums "self-appointed experts"... what changes were AMD going to have to make to get their process up to the standard of Intel?

Go on... answer that without mentioning HKMG.
 


Since you need to know, and I guess you did not read jaydeejohn's article, which clearly explains how AMD achieved 45nm without HKMG, I guess I will have to simplify it for you.

They (AMD) had to use SOI in an optimized method, to achieve 45nm. Now, because of the optimized way they created their gates, they are still not as powerful or as fast as Intel's gates, using HKMG (sorry, I had to mention it). So, because they HAD to use what was available to them, AMD created gates that can accept more power than previous Barcelonas, but could not push the voltages as much as Intel's gates, due to possible (I said, possible) leakage. Did it reach the "standard" of Intel? Who knows. Until we see server CPUs based on Nehalem, we won't know. Now, did this process make Deneb better than existing Agena CPUs? We don't know that either.

■Compared with Intel's speed-burners, a typical 45-nm transistor on AMD's Shanghai is a lot less powerful.
■How'd they do that? The answer, in a word, is optimization. Although there are no new materials or techniques like adding an additional stressor for strained silicon engineering, AMD improved transistor performance by squeezing every last drop out of the performance-enhancing structures already in use at 65 nm.
Click to expand...
Wow. That was hard to figure out. (Thanks, jaydeejohn, for the link).

Now, you still haven't shown any such post claiming that AMD needed HKMG. Not assumptions, not YOUR opinion of what a poster wrote. Go search...I did. And all I found was talk about HKMG, never anything about how AMD would need it at 45nm, which you cannot get a grip on. In fact, I think there was more question about AMD going to 45nm, with SOI, than there was about needing HKMG at 45nm. BUT in your world, any mention of HKMG = "AMD needs HKMG".

Selective comprehension is quite a skill.

**edit**
I concede, just like jaydeejohn did earlier.

It doesn't matter what I post, because there will always be those that believe that "45nm SOI simply won't work well for them." = "AMD needs HKMG".
Enjoy making other posts mean something else, too.

Peace.
 


I don't question your "ifs" if you just simply use it hypothetically. However, as opposed to hypothesis, you generate this entire thread based on your "ifs", and put it as if its "fact".

No, its not fact. Its "if". IF AMD's 45nm SOI achieve similar power consumption and clockspeed like Intel's 45nm parts, then yes, HKMG is overrated. However we have not seen anything that supports this hypothesis.
 
not read through all the post, so just say sorry in advance if its been coverd.

the question you should be asking jaydee is, what kind of processor would Intel have at 45nm if they didn't go with HKMG, and to my mind i don't think it would overclock as far as it has.
 
Well, the slide weve all seen appears to be real. Now, IF Deneb comes in at able oc to 4Ghz, then this puts it all to bed. In server, I could argue and you could argue about power as being a prime concern, and effect its overall success to a much greater degree than on DT, if it matters at all on DT, because only the top end chips show a 125 TDP, and who buys those? And will those people worry about it?

I love the fact history is being bent here for some peoples arguement. We all know how it went down, and what it all was about. Ill use the word if until Deneb comes out and shows 4Ghz as being here, then therell be no point of it

As for the gates not having enough ability, thats most likely true to a point, but the channels/pipes have better ability than Intels, as has been shown by my link, and HKMG is for gates only, and its possible either that AMD is keeping the voltage down from hitting to hard on the gates, or the pipes are too resistive and limit it, as this could be the case as well

I think you guys need to work for AMD, as youve proven youre good at revisions, like whats happened here in the past, and what youre portraying as how it really went, at least in your eyes. Its sad when everyone knows what went down, and you are here denying it, just for???? Sticking up for Intel, and in the mean time, trying to hold onto the idea AMD cant do good, period. My bet is, yes, theyll do good at 45nm without HKMG, as well as TSMC will too, but thats something you ignore, and thats the very heart of this thread, and you try to revision the past, and whats been said about AMD, HKMG and the NEEDS of its usage at 45nm. It doesnt matter what you say about the past, as it cant be changed, and we all know how it went, so all your revisionist theories go out the window

Is it soooo hard to except that maybe AMD is going to have something decent? Thatll oc nicely? Have better IPC? And be as good as C2s? And, all without HKMG?

No where in this thread, or anywheres else have I ever said HKMG isnt an improvement, but that again isnt going to the heart of this thread, nor pointing to what others have said, knowingly, about its use, and how itd effect AMD at 45nm without it. Sorry guys, you just cant change history
 
rangers, it truly doesn't even have to be as fast as the core2's. i think that is the disconnect that exists here. for consumer and commercial OEM lines, they just need to be better than they were and AMD needs to be able to deliver in high volumes. if anything hurts them it would be failure to deliver large quantities to OEM manufacturers. having to change over lines because your supplier didnt deliver will make a high volume pc manufacturer laugh in your face when you expect a third chance.
 



have to disagree

AMD need a fast processor to compete in the enthusiast market, but OEMs will lap it up all they are concerned about is the price, so you have a point there
 
Heres a few snippets http://theovalich.wordpress.com/2008/11/18/tsmc-introduces-40nm-volume-production-advances-in-front-of-intel/

http://www.eetimes.com/rss/showArticle.jhtml?articleID=210604347&pgno=3

Now, looking at my first link, if HKMG was so important, surely TSMC would have used a lil of that 10 Billion dollars towards it. Seeing as they arent going to use it til 28nm, and theyll soon be taking the process node lead, it puts HKMG in perspective as to how important it REALLY is.
 
Basically this entire argument is based purely on speculation. Can AMD do without HKMG? We don't know yet because all we've seen is a slide that implies that their CPUs can go over 4Ghz.

There are plenty of other variables besides what materials are being used. Even if AMD did use HKMG it doesn't guarantee that there would be a significant performance boost or power drop if there is a problem with the design / process itself. We can all agree that HKMG is a good thing but it's not necessarily a deal breaker.

AMD could be holding off due to financial issues or they simply thought it wasn't necessary yet, only time will tell.
 


Theo is a brain dead idiot, and frankly linking to him says the same about you.

Word, Playa.
 
Awww, did I up set you? Linking to him has nothing to do with changing facts. These are the facts. Whats really sad is, spud, you were there as well, and saw what was said, and you refute what Im saying? That the common word/knowledge given at the time of how HKMG would have to be a najor player at 45nm? And without it, nothing would work properly? And so we see others, or everybody else going to smaller nodes, 45nm and down, below Intels current offerings, without HKMG, and because Theo has something to say about it, and I include him, its obviously wrong?

For some reason, people are having a hard time swallowing all this, but for the ones that were there and read all the threads know better, and here you are bashing me. Nice, this isnt a good post from you spud, or did you vote for Obabma? And association only applies when and where YOU want?
 
If having HKMG is so beneficial, theyd do that before aiming at smaller nodes. If HKMG was sooo needed, those smaller nodes wouldnt work without it, this is what we were told, and thats bull. Its overated, it isnt needed yet. It will help, itll bring nice power savings, but at this point, only server usage in cpus is it as desired anywheres remotely near whats been said, and Shanghai currently is looking pretty good without it.

I bring in the TSMC info because not only is someone else doing it without HKMG (going to 45nm and smaller), but it also shows another player , or maker of chips that CAN afford to make this decision, and have decided not to, which goes to the heart of the matter, and everything thats been previously said about it.
 


Here it goes again..

Let's break it down really quick.

First of all, Theo Valich is an idiot, period. He was the one who wrote on Reverse Hyperthreading, and we all know how well that turned out :sarcastic:. In short, I wouldn't go to him if I want accurate factual information.

Secondly, there is a reason why TSMC invests in smaller node before HKMG. For a foundry company like TSMC, it cares more about reducing the cost and increase production rather than performance oriented. But does that mean TSMC doesn't want HKMG? No, because they will have a hard time producing quality wafers if they stick to bulk CMOS beyond 45nm. This is why they decided to implement 28nm with HKMG. However, does that mean they have to get it now? No, because they can still produce decent wafers using 45nm bulk, and for them, it is good enough.

Another point I want to raise about Theo Valich is that, besides the fact that he's an idiot, he simply just looks at one side of things. He stated that by concentrating on smaller nodes, TSMC can massively improve economy of scale, thus saving both Nvidia and ATI a lot of money. This is true, yet he omitted a large part of equation: performance. Sure, Intel may not be able to produce as many dies (raw ones) as TSMC, but since everything is done in house, Intel is able to control the quality of the die, and tweak the process as it sees fit. For companies that rely on foundry to produce their chips, this is simply not possible. So while TSMC can produce more raw dies per wafer, I wouldn't be surprised if Intel can extract more functional dies per wafer.

Taking a quote from your second source:
For 32-nm, SiON is a safer bet and the vast majority of devices will not require a high-k/metal-gate duo, Wei said. SiON ''has a clear position in the market," he said. "It's a lower risk technology."
Click to expand...

In other words, while HKMG is a much better process technology, not every company is going to receive the same benefit like Intel does if they transition to HKMG. Hence the term, "resource allocation".
 


Intel BK 2009. /lol
 
This thread shouldn't have reached even page 2, since there is no link whatsoever to any article that has both 'HKMG' and 'needed' in the same page - which was the main point of the OP.

Keep in mind that the relevance of what engineer wannabes forum members say is quite different from that of an official response from Intel, AMD or any other company. Yet, I don't doubt 'someone' (from this forum) have said something like HKMG being 'needed' - and I couldn't care less, since it's probably not part of any design teams inside Intel or AMD.

The main problem is that people seem enthusiastic in 'odd ways' about any possible good news about AMD, which has been miserably failing over the past 2 years and, suddenly, is about to deliver the next super duper thing just because they got their GPU line allright last time.

I'm sorry, Jaydee, but, although I have a lot of respect for all the years you seem to have been around (and many very good posts I have read from you), things start to get 'questionable' when titles like "Watch out, Intel, here comes AMDhabi!!!" show up and then we have discussions about a 'die-shrink' that is yet to beat a 2-years old 65nm double-cheeseburguer.

BTW, there was a post in XtremeSystems comparing an underclocked Shanghai vs Barcelona, at the same clock speed. Shanghai was 11% faster in *one* benchmark and something in between 1-4% in *every* other, so, unless you are talking virtualization or server performance here, I can't see any evidence of Deneb's forthcoming unimaginable gains in the desktop environment.

I'll try to find the link again, since I don't wanna spread more BS than this thread has been having since its beginning.
 
http://www.xtremesystems.org/forums/showthread.php?t=207460

attachment.php


Here it goes: amazing performance per watt improvement.

2.6% average performance improvement.

Unimaginable [again] [for daydreamers AMD fanboys].
 


If you upset me I would have been talking with turpit, and since I could give a **** there is no issue. Linking to that wanna be idiot is what all the previous clowns that have come and gone have done. Baron, Nine Inch, Pope, and the rest of the tools that have a issue with their imaginations.

As for facts I don’t have a clue what your talking about I don’t read these **** smear threads with intent to be somehow educated. I read them to gauge posters. I'm interested in the type of people I may or may not have to engage in not the topics or the criteria. I do my research myself; I don’t need some half wit clown with a PP slide telling me the entire industry is somehow going to be changed from an apparent node change.

For the record Intel’s switching speeds at all nodes are noticeably faster than the entire industry and that’s all that matters.

I never said anything about 45nm and don’t need to HK allows for a more robust gate switching speed with reduced leakage, and that’s all it does; it doesn’t make 45nm work. All it allowed was Intel to push their own switching records to new heights’ and it works.

Can’t vote for Obama I R Cannuck.

Again Theo = ass clown so linking to Theo = ass clown.

Word, Playa.

 
You must log in or register to reply here.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom