HKMG overated?

[JAYDEEJOHN]

". 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. "
Thats not theory. Already its proven without HKMG, soi is doing quite well, much better than any outlook Ive heard on these forums

 

[NMDante]

Fine. Want to see what advantages HKMG could bring? Okay.
SpecFP_2006rate

ASUSTeK Computer Inc. Asus P6T Deluxe (Intel Core i7-940) 4 1 4 8 82.3 79.2 Nov-2008
ASUSTeK Computer Inc. Asus P6T Deluxe (Intel Core i7-965) 4 1 4 8 86.1 82.9 Nov-2008
Supermicro A+ Server 1021M-UR+B, AMD Opteron 2384 8 2 4 8 118 105 Nov-2008
Dell Inc. PowerEdge M805 (AMD Opteron 2384, 2.70 GHz) 8 2 4 8 117 104 Nov-2008
Dell Inc. PowerEdge R805 (AMD Opteron 2384, 2.70 GHz) 8 2 4 8 117 104 Nov-2008

These are single socket Core i7 systems, using the overrated HKMG, vs. dual socket Shanghai CPUs. Hmm...Not that big an advantage for a dual socket system, which also has 32GB memory, and running Linux compared to the Core i7 systems, running Vista Ultimate, with 12GB of memory.

So, yeah...HKMG is overrated and not needed. You win. AMD doesn't need it at all. I guess having a dual socket, using SOI, that is around 28.3% faster than a single socket system, using the overrated HKMG, is good enough. (28.3% comes from avg of Core i7 and Opteron 2384 ratings - 84.2 and 117.3)

SpecOrg


Also, you have yet to show any post claiming that anyone has said that HKMG is needed for 45nm.

 

[JAYDEEJOHN]

If I were talking about an Intel cpu, then I could make the same comparison, one with, one without. Not apples to apples. Compare a Barcy to Shanghai, thats how its done, as 45nm on the same arch minus die shrink. You see, Im sure people werent thinking AMD would be making Intel cpus that wouldnt work that well when they all said it wouldnt work that well without HKMG, Im thinking they were talking about AMD cpus, and how they wouldnt perform well.

 

[JAYDEEJOHN]

I have one saying itd be a burner, or something to that effect, and theres been others, but Im not going there. Ill stick with the accepted not work that well, since you and yomama have both agreed so



Look, heres what we do know. At 45nm without HKMG, we see AMD coming in at over 15% faster on the clocks. Better than any C2D jump, from one tick or tock to another. Ive given you the power results, now it all comes down to my if. If it ocees as well, Im 3 for 3 and theres none for thee

 

[NMDante]

Okay. Barcelona vs. Shanghai, courtesy of Anandtech.
Anandtech Shanghai Database testing

■In the first four load points, the performance of Shanghai @ 2.7 GHz vs. Barcelona @ 2.3 GHz is within 2% even though there is a 17% bump in clock. At load point five, Shanghai is able to outpace Barcelona by 11% but still less than the clock bump.
■Shanghai uses approximately 12% less power than Barcelona.
■ For the first four load points Shanghai is again ~12% more efficient than Barcelona, and as much as 28% or 47% more efficient than the Intel systems depending upon the DIMM configuration.
■Again, the performance of Shanghai @ 2.7 GHz is comparable to Barcelona @ 2.3 GHz at the lower load points but is able to lead Barcelona by as much as 25% at load point five.
■Again, the performance of Shanghai @ 2.7 GHz at the lower load points is not significantly better than Barcelona @ 2.3 GHz. At load points four and five the faster Shanghai core is able to easily outpace Barcelona.
■Shanghai is able to lead Barcelona by as much as 37%.

What about how Shanghai fairs against its older brother Barcelona? Well, in some cases, the gain is clearly just the increased clock speed. However, in others Shanghai achieves an increase of anywhere from 10-15% over and above the clock speed difference. It's obvious that Shanghai is what Intel would call a "Tock" of the clock for the Barcelona architecture, and it is a nice little bump for turning a few knobs and a die shrink. 2009 will be a very interesting year for AMD and Intel. Whether or not Shanghai can hold down the fort until Istanbul comes out remains to be seen.

Still huge gains at some, not so much at others. So, 17% speed increase equals spotty performance gains, and that is good? Okay.

Oh, and please...use your "burner" statement. You seem to include me in your discussion, even though I was never in any of the previous HKMG threads that Just_An_Engineer, or you, have linked. So, please...share your great burner find.

 

[JAYDEEJOHN]

It wasnt you. My points are made, unless you want to continue. My points are, at 17% clock increases, at lower power draw, that doesnt equate to not work that well, and wont without HKMG. If it ocees as well, then the it wont work well wont fit there either. So just where does HKMG seem so misinterpreted? Maybe by those that espouse it? Or, on a better theme, by those that would benefit Intel, their choice Im sure, and all without true knowledge. As these words, wont work that well are used with the knowledge of the lack of HKMG, which was used prior, and in context with all this, it was wrong thinking, it wasnt on the spot, and the usage of HKMG along with working good vs not work good without it is overated. Its wrong, and Ive shown you already thats the case.

If you happen to think otherwise, just wait til Denebs numbers

 

[JAYDEEJOHN]

When Intel went from 65 to 45nm, did they produce a 17% clock increase? They had HKMG. I know they didnt. I dont know the power numbers, but Im asking you, what was the power dropoff/usage at 45nm using HKMG? Was it vastly more than 12%? Id guess the ocing went up around 10%? That fair?

 

[JDocs]

Yes, its an improvement however even AMD is hoping for a larger improvement with HKMG. Yes, 45nm without HKMG works and works decently but AMD admits it would work even better with HKMG. So in context 45nm with HKMG 45nm without vs HKMG; without does not work well. Lets not forget Intel's 6 core 3.2ghz to 3.6ghz 45nm HKMG chip. That could be done without HKMG just not as easily.

 

[cjl]

When Intel went from 65 to 45nm, did they produce a 17% clock increase? They had HKMG. I know they didnt. I dont know the power numbers, but Im asking you, what was the power dropoff/usage at 45nm using HKMG? Was it vastly more than 12%? Id guess the ocing went up around 10%? That fair?

The power dropoff was rather impressive actually.

Note that the E6850, a 3GHz 65nm part, has 1.94 times the idle power requirement and 1.55 times the load power of the E8500, despite the fact that the E8500 is clocked roughly 5% higher and has 50% more cache, as well as a 40% increase in transistor count. I'd call that a bit more than a 12% power dropoff, yes.

That's what HKMG did for Intel.

 

[JAYDEEJOHN]

I agree 100% with that. But if youve been around here long enough, and even going back to when Intel first announced HKMG, the response, and then, as up til now knowledge of HKMG vs SOI has been majorly skewed, and everyone was left with the idea @45nm without HKMG = Fail. Itll help. And could be needed down the road, but many here were led to belive itd be needed for 45nm, or it wouldnt work. Thats overating it, and underating SOI. Ive gotten this response only because of certain diehards that believe it still. No where will I ever say HKMG isnt good, but putting it in perspective of whats gone on HERE is what Im after. Itll be a great addition to AMD,IBM,TSMC etc

 

[JAYDEEJOHN]

That is impressive. I know you need to add in the node shrink. But even so, the power draw does impress.
I wont lie to make myself look good. My point is, like I said, I got the impression, as well as others did here, that without it, itd be failure. I concede the power numbers, without a doubt. But as for the ocing and the clocks themselves, that needs to be addressed, or has been actually, and found not to be so true, regarding the usage, or non usage of HKMG.


@cjl, whered you get those numbers? If this is the case for anyone going to HKMG for power usage, then I can sorta make some projections

 

[cjl]

Those numbers are from xbit labs. They're one of the only review sites that I've found where they actually measure the power the CPU itself is drawing, rather than the overall system power.

 

[NMDante]

When Intel went from 65 to 45nm, did they produce a 17% clock increase? They had HKMG. I know they didnt. I dont know the power numbers, but Im asking you, what was the power dropoff/usage at 45nm using HKMG? Was it vastly more than 12%? Id guess the ocing went up around 10%? That fair?

■Results: The Intel QX9650 actually consumed less energy than the QX6850 at both idle and load, which was a bit shock. It just goes to show that Intel has spent a lot of time with Yorkfield making sure it is energy efficient and that the core doesn't leak. From this testing it shows that 45nm is much better for power consumption as it was 20% more energy efficient at load over the Intel QX6850!
■Results: Crysis came out just days before this article was published, but we included it as the community is following this game like no other. The CPU scaling found on this game was much more defined than with BioShock and that makes for a nice chart. The Intel QX9650 was 8.5% faster than the equally clocked Intel QX6850 and almost 25% faster than the AMD Athlon 64 X2 6400+ Black Edition processor. It's clear that CPU choice will matter with games like Crysis!
■Results: World in Conflict had a new patch come out in the past week and we were able to benchmark the latest patch (002) on the latest and greatest processors from both AMD and Intel to show you what performance will be like. The low, average and high frame rates in World in Conflict scaled nicely and the Intel QX9650 once again lead the pack. It was 16% faster on average than the Intel QX6850, with the high frame rate being just shy of a 22% improvement! The extra cache is no joke when it comes to current PC games!

Legit Reviews

Since you like gaming so much, 8.5% faster in Crysis, 16% in World in Conflict. All while being 20% more efficient than 65nm.
Seems like an improvement overall, even without the 17% clock increase.

Also, you keep claiming that people (or diehards, as you claim) have stated that 45nm without HKMG would fail, yet you have linked a thread that had 0 reference to HKMG and 45nm. You just keep adding HKMG to anything that says 45nm and SOI won't work. That does not equal 45nm with HKMG only. YOU keep thinking that. Unless you can actually provide a statement with someone claiming that 45nm will only work with HKMG, then you are just crying over nothing. PERIOD.

 

[JAYDEEJOHN]

Actually, the link was justaengineers, and yes, HKMG was mentioned. Thats how he found it on search. I did as well, but decided it wouldnt be right to quote, but the cat was out of the bag already, and I was being told something other than what Id read. Or rather, it was to the point of this post.

I concede that HKMG is really nice for Intels cpus, and hopefully translates to anyone using it regarding power usage. At least there, its not overated. But the things thats been said about not much higher than 3Ghz, thats what I remember most, and those were when it first came out, and AMD was stuck at 2.6, and the feeling was, even at 45nm, without HKMG, theyd never make it higher than that. Im sure you were there and read them. It was JJ, DSN, the rest, and yes, thats how it went down. Later posts had tuned it down some, but there were varying comments about AMD trying without it, itd burn, things like that, and thats simply not the case.

Like going from 3.4 at 1.6v on a C0 revision to 4.0 at 1.6v on C1, no one ever would have thought this possible, and some still dont. All Ive got to say is, dont try and sell that to coolaler

Its been an ongoing bashing, misrepresentation of something thats very useful, but has been aimed carelessly, and abused for fanboism. As for the clocks, its been totally misunderstood, as has SOI and AMD and their abilities

Im not a cpu fanboi at all, really. Im betting its hard to accept that, but if someone starts to thump Intel, and continues it for years, and a whole bunch jump on that wagon, guess what kinda threads youll see me making?

 

[JDocs]

JaydeeJohn but the catch is averages not the extremes. 4ghz @ 1.6volts. AMD demoed 3ghz 65nm X4s and how many did they sell? HKMG has allowed Intel to do better averages which is what matters.

45nm without HKMG won't fail in itself. It will however be more expensive, produce lower yeilds and have other issues that HKMG helps to overcome making 45nm without HKMG a failure by comparison in corporate terms. The current 65nm X4s are not bad chips but by comparison to the Intel 65m they are a failure.

 

[JAYDEEJOHN]

I agree. Im talking about 45nm tho. The monies are already spent, check my link at the beginning. AMD has focused on 45nm, not 65, they never really did alot there, and never intended to. I know that sounds bad, and it is, but its true.
So going to 45nm, we will see improvements thatll change attitudes about AMD, as here is where their true fpcus has been all along. Here amd forwards.I dont have the quote, but Hector said so himself.
Im sure with AMDs limited monies, its been hard to stay and conquer any node, and they sorta blew over 65nm, trying to catch up, or keep pace. Im guessing that at 45nm, we'll see an improvement and part of that is dispelling a lil of the SOI inadequices, as well as AMD. Im hoping so, right along with everyone else, except the haters

 

[JAYDEEJOHN]

I dont think HKMG helps yields per se, possibly in that using it could raise the overall clocks more, thus making for a lil better yields, which, more is more, just not that much more. And using their new liquid immersion is the money thats already been spent, and more than likely, youll see Intel using it, as was in my link, after AMD runs the kinks out first. Thats part of the reason theres alternatives to having to go to HKMG at 45nm a need as being not so, and yes, the yields from AMD may be good for all we know, and possibly really good, as this new process is just that, and so far, looks promising, seeing whats been previously thought not likely, if not possible

 

[Mathos]

Neh, you know what, I agree with Jaydee on people overhyping the need for AMD to transition to 45nm. I can remember many long debated thread, saying that there is no way that AMD could do 45nm without using HKMG, otherwise they would have massive current leakage and all sorts of other bs.

The article he linked to isn't by AMD it's by an independent company that tests IC characteristics for all companies. The article points out that the current leakage is much lower than they had expected, even lower than that of HKMG gates on the 45nm Intel procs. Only thing that seems to be a problem is the current drive strength on the gates is a lot lower, but still higher than the previous 65nm gen.

 

[JAYDEEJOHN]

As I understand it, the current drive strength, is a sort of resistance factor? In other words, if it is high, it has less resistance, if low, then more

 

[Amiga500]

No. It's an opinion about 45nm using SOI. Nothing about HKMG.
What does that tell you?

Are you actually being stupid, or just pretending to?



I don't believe there is truly anyone on this forum stupid enough to believe HKMG is not inferred when people were saying '45nm SOI would not work well for AMD'.



You are arguing on the basis that you have not actually seen these exact words in this exact order "AMD need HKMG for 45nm", while half the posts on the linked thread allude to that.


But hey, that seems to be a similar line of argument to the overclocking thread.

 

[Homeboy2]

while you guys were arguing i was doing your girlfriends

 

[NMDante]

Are you actually being stupid, or just pretending to?



I don't believe there is truly anyone on this forum stupid enough to believe HKMG is not inferred when people were saying '45nm SOI would not work well for AMD'.



You are arguing on the basis that you have not actually seen these exact words in this exact order "AMD need HKMG for 45nm", while half the posts on the linked thread allude to that.


But hey, that seems to be a similar line of argument to the overclocking thread.

Yeah, I'm being stupid, since the comment was a reply to another posters comment about how the first generation will be on SOI. Yup. I guess since someone says "45nm SOI simply won't work well for them. " that means "45nm needs HKMG". Holy crap. What was I thinking? So, from now on, everyone should just be able to read what anyone says, and add or subtract what they want, to fit what they believe.

Like the OP, you keep claiming that there was many statements about AMD needing HKMG for 45nm, yet it's okay to assume what someone meant (even thought the comment, and the comment it was being replied to, have no reference to HKMG).

Yeah, I'm just soooo stupid. Half the posts on the linked thread TALK about HKMG, never about how AMD will NEED it. Learn to read what is written, without adding your own ideas.

Oh, wait, if it has HKMG in it, it must refer to how AMD needs it, right?

 

[Amiga500]

I guess since someone says "45nm SOI simply won't work well for them. " that means "45nm needs HKMG".

So, you actually are really being stupid then?


So... since 45nm SOI won't work well for them, what was the changes they would have to make to make it work well?


[Sarcasm]Outside of HKMG - since that is soooo obviously not was inferred[/Sarcasm]

 

[NMDante]

Neh, you know what, I agree with Jaydee on people overhyping the need for AMD to transition to 45nm. I can remember many long debated thread, saying that there is no way that AMD could do 45nm without using HKMG, otherwise they would have massive current leakage and all sorts of other bs.

The article he linked to isn't by AMD it's by an independent company that tests IC characteristics for all companies. The article points out that the current leakage is much lower than they had expected, even lower than that of HKMG gates on the 45nm Intel procs. Only thing that seems to be a problem is the current drive strength on the gates is a lot lower, but still higher than the previous 65nm gen.

Yes, but the lower leakage is due, as you even mentioned, to lower drive current, which Intel uses a much higher current.

Intel has a long history of pushing processor-speed performance ratings, and did so by designing transistors with high on currents to drive fan out gates quickly. In fact, Intel's 45-nm high-K metal gate transistors have the best peak drive currents on the market with 1.36µA/µm for NFETs and just over a milliamp for PFETs. Compared with Intel's speed-burners, a typical 45-nm transistor on AMD's Shanghai is a lot less powerful. I know very little about CPU benchmarks let alone server tests, but I know the speed performance for a Shanghai-based server would be limited at some point by the lower drive current available from the AMD 45-nm transistors.

That still does not take away the fact that AMD has transitioned to 45nm without the use of HKMG, but would the processor speeds be faster with HKMG? Would AMD be able to push a higher drive current if they had access to HKMG process? That is unknown, but could be seen later, for AMD's next generation 45nm or 32nm node.

AMD plans to introduce its version of an HKMG process developed in partnership with IBM midway through 45-nm production.

Again, based on the OPs original question of whether HKMG is over rated, it's not. The easiest answer is that AMD did not have the capability to use HKMG on it's first generation of 45nm CPUs, and had to use what was available. If AMD could have used HKMG @ 45nm, who believes they wouldn't?

 

[piesquared]

JaydeeJohn but the catch is averages not the extremes. 4ghz @ 1.6volts. AMD demoed 3ghz 65nm X4s and how many did they sell? HKMG has allowed Intel to do better averages which is what matters.

45nm without HKMG won't fail in itself. It will however be more expensive, produce lower yeilds and have other issues that HKMG helps to overcome making 45nm without HKMG a failure by comparison in corporate terms. The current 65nm X4s are not bad chips but by comparison to the Intel 65m they are a failure.

I wonder if we should use some common sense and realize that cpuz isn't reading that voltage right. All it takes is a little and it's free.

If not, I guess i'll have to show the cpuz screen from another post showing 4.0ghz @ 1.168v.