No hi-K (or MG) for AMD's 45nm

[Idontcare]

AMD believes that it can be "competitive" without high-k at this time, but Silcott noted that high-k is an option to be introduced at a later stage of 45 nm production - and "definitely at 32 nm".

http://www.tomshardware.com/20...mos_45_nm_processors_/

This is really not good news because IBM previously reported the power consumption characteristics of their 45nm SOI process without hi-k and without MG for their CELL processor...and power consumption at same processor speed was not reduced as much as one would expect or like.

Power consumption of the 45 nm CELL processor is less than forty-percent that of the 90 nm CELL processor ? now less than 20 watts.

http://www.realworldtech.com/p...icleID=RWT022508002434

When I read the power reduction info on IBM's 45nm CELL I was really really really hoping that AMD would be spared because they'd at least have hi-k gate oxide (or so I thought they had previously claimed).

But seeing that they now plan to move to 45nm without hi-k I expect their power consumption reduction over 65nm is not going to be anything to write home about (just as 45nm CELL's isn't).

 

[Nemesis 1]

http://ca.news.finance.yahoo.c...processors-server.html

 

[imported_wired247]

Unless intel licenses this technology to AMD, I strongly doubt AMD's mid-high end CPUs will be competitive with Intel's for quite a few years. IT is very nontrivial to incorporate the high-K annd metal gate

 

[KingstonU]

Disappointing, but hopefully they will have it implemented by the time Fusion on 32nm rolls out in 2009 and they can show some impressive improvements.

 

[Extelleron]

I don't see how 45nm = 40% of 90nm is that bad in terms of power consumption.

QX9650 consumes about 100W vs the 130W of the QX6850, for about a 30% efficiency increase going from 65nm->45nm. So 45nm power consumption would be ~75% of 65nm.

So if going from one node to another is 75%, then 40% for two nodes doesn't seem too bad.

 

[heyheybooboo]

Originally posted by: Nemesis 1
http://ca.news.finance.yahoo.c...processors-server.html

""AMD Demos 45nm Native Quad-Core Processors for Server, Desktop""

I read somewhere (maybe XS?) that cores had already been delivered to developers ...

I don't know what that means for AMD production but I assume manufacturers can now validate systems & BIOSs

 

[Nathelion]

Well I suppose we can expect shanghai for christmas... right on time for Nehalem

 

[Idontcare]

Originally posted by: Extelleron
I don't see how 45nm = 40% of 90nm is that bad in terms of power consumption.

QX9650 consumes about 100W vs the 130W of the QX6850, for about a 30% efficiency increase going from 65nm->45nm. So 45nm power consumption would be ~75% of 65nm.

So if going from one node to another is 75%, then 40% for two nodes doesn't seem too bad.

Intel reduced power consumption by 50% in going from 65nm->45nm when normalized to the number of transistors...

 

[Extelleron]

Originally posted by: Idontcare

Originally posted by: Extelleron
I don't see how 45nm = 40% of 90nm is that bad in terms of power consumption.

QX9650 consumes about 100W vs the 130W of the QX6850, for about a 30% efficiency increase going from 65nm->45nm. So 45nm power consumption would be ~75% of 65nm.

So if going from one node to another is 75%, then 40% for two nodes doesn't seem too bad.

Intel reduced power consumption by 50% in going from 65nm->45nm when normalized to the number of transistors...

Yes, but that's not a fair statement. Yorkfield's 41% more transistors are virtually all cache transistors, which don't consume nearly as much power as logic transistors. You can more than double the amount of transistors on a Penryn if you go from 6MB L2-> 16MB L2, but you're not going to see anywhere close to 2X the power consumption.

I don't think AMD is going to be able to compete with Penryn/Yorkfield on power consumption, the only reason they ever had an advantage was Intel had Netburst, which didn't even try to compete in terms of performance per watt. But I think AMD will see a decent lowering in power consumption going from 65nm->45nm.

 

[nullpointerus]

Heat is the wasted energy, so if power consumption is not cut by much, heat output probably will not change much either. So this news means that the heat reduction won't be great?

From what I've read, these chips run cool at stock speeds but really heat up near 3 GHz. But, AFAIK, those could be people comparing X2's to X4's, and we all know quad-core chips have a much higher heat output than dual-core chips (all else being equal). Are current Phenoms really o/c-limited by heat?

Does this foretell anything for Phenom B3 overclocking?

I'm getting desperate for some positive news...

The stock Phenoms are fast enough for most users. I think AMD's real problems are with enthusiasts, such as overclockers and those who multitask heavily. Athlon chips were such hits because they were inexpensive and still reasonably overclockable, the 64's and X2's hitting some fairly amazing OC's for their day (33% to 50% in some cases).

Not to go too far off-topic, but I did some reading in that 50-page Phenom overclocking thread, and motherboard/BIOS limitations played a big role. I hope the newer boards previewed in Anandtech's CeBIT article, coupled with the B3 chips, will provide some much needed and (relatively) reliable overclocking and make Phenom popular enough to demistify the "black art" of Phenom overclocking.

Core 2 is pretty stiff competition for overclocking, especially at 45nm.

 

[Idontcare]

Originally posted by: Extelleron

Originally posted by: Idontcare

Originally posted by: Extelleron
I don't see how 45nm = 40% of 90nm is that bad in terms of power consumption.

QX9650 consumes about 100W vs the 130W of the QX6850, for about a 30% efficiency increase going from 65nm->45nm. So 45nm power consumption would be ~75% of 65nm.

So if going from one node to another is 75%, then 40% for two nodes doesn't seem too bad.

Intel reduced power consumption by 50% in going from 65nm->45nm when normalized to the number of transistors...

Yes, but that's not a fair statement. Yorkfield's 41% more transistors are virtually all cache transistors, which don't consume nearly as much power as logic transistors. You can more than double the amount of transistors on a Penryn if you go from 6MB L2-> 16MB L2, but you're not going to see anywhere close to 2X the power consumption.

I don't think AMD is going to be able to compete with Penryn/Yorkfield on power consumption, the only reason they ever had an advantage was Intel had Netburst, which didn't even try to compete in terms of performance per watt. But I think AMD will see a decent lowering in power consumption going from 65nm->45nm.

Sigh, now we get graded on how "fair" our facts are? Color me unimpressed, I worked on 45nm technology and I'm telling you this is not a good showing. Such results are expected when you don't implement hi-K or MG at 45nm, but having excuses doesn't pay the bills too well.

Per AMD's press release today, they expect 20% power consumption decrease with 45nm over 65nm. This is in-line with IBM's power consumption reduction in 65nm -> 45nm CELL.
http://www.amd.com/us-en/asset...Press_Presentation.pdf

 

[Extelleron]

Originally posted by: Idontcare

Originally posted by: Extelleron

Originally posted by: Idontcare

Originally posted by: Extelleron
I don't see how 45nm = 40% of 90nm is that bad in terms of power consumption.

QX9650 consumes about 100W vs the 130W of the QX6850, for about a 30% efficiency increase going from 65nm->45nm. So 45nm power consumption would be ~75% of 65nm.

So if going from one node to another is 75%, then 40% for two nodes doesn't seem too bad.

Intel reduced power consumption by 50% in going from 65nm->45nm when normalized to the number of transistors...

Yes, but that's not a fair statement. Yorkfield's 41% more transistors are virtually all cache transistors, which don't consume nearly as much power as logic transistors. You can more than double the amount of transistors on a Penryn if you go from 6MB L2-> 16MB L2, but you're not going to see anywhere close to 2X the power consumption.

I don't think AMD is going to be able to compete with Penryn/Yorkfield on power consumption, the only reason they ever had an advantage was Intel had Netburst, which didn't even try to compete in terms of performance per watt. But I think AMD will see a decent lowering in power consumption going from 65nm->45nm.

Sigh, now we get graded on how "fair" our facts are? Color me unimpressed, I worked on 45nm technology and I'm telling you this is not a good showing. Such results are expected when you don't implement hi-K or MG at 45nm, but having excuses doesn't pay the bills too well.

Per AMD's press release today, they expect 20% power consumption decrease with 45nm over 65nm. This is in-line with IBM's power consumption reduction in 65nm -> 45nm CELL.
http://www.amd.com/us-en/asset...Press_Presentation.pdf

I was just saying that you can't factor cache transistors into a statement like that, because they consume so much less power.

If AMD only expects 20%, then I'm just as unimpressed as you are. With 20% less power, Phenom 45nm won't consume much less (if at all, in certain situations) than 65nm Kentsfield. Still, I would rather AMD release 45nm CPUs now w/o High-K/MG than wait till 2009.

 

[Martimus]

From what I've read, these chips run cool at stock speeds but really heat up near 3 GHz. But, AFAIK, those could be people comparing X2's to X4's, and we all know quad-core chips have a much higher heat output than dual-core chips (all else being equal). Are current Phenoms really o/c-limited by heat?

From that same 50 page document, I remember KTE saying that temps were fine (often staying below 50C) but other things were limiting him.

edit: I tried looking through the forum posts to find a quote, but there are 57 pages of posts and I gave up quickly. I already read through them once, I don't have enough interest to do it a second time.

 

[nullpointerus]

Originally posted by: Martimus

From what I've read, these chips run cool at stock speeds but really heat up near 3 GHz. But, AFAIK, those could be people comparing X2's to X4's, and we all know quad-core chips have a much higher heat output than dual-core chips (all else being equal). Are current Phenoms really o/c-limited by heat?

From that same 50 page document, I remember KTE saying that temps were fine (often staying below 50C) but other things were limiting him.

edit: I tried looking through the forum posts to find a quote, but there are 57 pages of posts and I gave up quickly. I already read through them once, I don't have enough interest to do it a second time.

A gallant effort, to be sure!

Thanks for the recollection, BTW.

 

[Dainas]

All I'm praying for is that 45nm will scale with clocks quickly and watts the way process shrinks are supposed to, just not failing like K8 65nm did with the former. Hi-k or not AMD getting back to atleast neck and neck with Intel with this architecture, seem about as far flung as them introducing a socket 939 phenom variant to sop up as much untapped market as possible.

 

[Kuzi]

Originally posted by: Extelleron
I was just saying that you can't factor cache transistors into a statement like that, because they consume so much less power.

If AMD only expects 20%, then I'm just as unimpressed as you are. With 20% less power, Phenom 45nm won't consume much less (if at all, in certain situations) than 65nm Kentsfield. Still, I would rather AMD release 45nm CPUs now w/o High-K/MG than wait till 2009.

If we assume that 45nm AMD CPUs will have larger caches + faster IMC clocks, then my guess you guys are right about their new CPUs still being less effecient than Intel's 45nm ones.

But I also believe AMD are being a little modest about the power reduction numbers, and it is a little better than 20%.

 

[Sylvanas]

AMD to use High K metal gates

 

[Idontcare]

Originally posted by: Sylvanas
AMD to use High K metal gates

No one is disputing what AMD claims they will eventually use at 45nm.

Currently they claim they will eventually use (1) high-k gate oxide, (2) metal gates, and (3) ultra low-k dielectric for interconnects.

But my post was specifically in regards to what their 45nm is today, what is in the 45nm shanghai samples sent to their partners and what we will get the opportunity to buy this year.

Powerpoint slides detailing what your node needs to compete is not the same as shipping product from a competitive node.