This is how the old rumor PII@ 4.0 4.4 G can be true.

[MTd2]

Use HKMG. But that would require such chip already taped out for a long time to make the rumor true. Maybe AMD is conducting extensive tests so that it can compete against westemere?

IMPORTANT EDIT

I am NOT refering to 3.5. I am refing to the OLD rumor about that 4.0 GHz resurfaced.

here:

http://www.chw.net/noticias/242561-am3- ... bores.html

EDIT:

Another rumor:

[abinstein]

The german site claims Ph2 3.5GHz this year. If this is true then with HKMG Ph2 can be pushed toward 4GHz next year. Bulldozer is a late 2010 or 2011 thing on 32nm, so ... both can be true (or false).

[Das Tipitz]

I don't see a 3.5GHz Phenom II being too out of the realm of possibility. With RB-C2 revision, it is definitely out of the question, but with an improved revision, it should be possible. AMD's process tends to improve substantially as time goes on, so if later this year we see a C3 stepping it is likely that it will be much, much better than what we are seeing now. Phenom II @ 3GHz @ 1.35V isn't anywhere close to utilizing the full 125W TDP, so a future CPU @ 3.5GHz @ 1.3-1.35V should still fit within the 125W level.

4GHz is not happening on 45nm however. Deneb chips can't overclock to 4GHz on air and getting it stable requires 1.6V+ of voltage. AMD isn't suddenly going to turn that around and be hitting 4GHz at a factory-stable (aka 24/7 stable for years running the hardest type of code imaginable, far beyond what oc'ers call "stable") at 1.3-1.35V. HKMG isn't going to bring that big of an improvement either. Intel didn't get anywhere near that kind of an improvement going from 65nm -> 45nm HKMG - no way AMD improves that much going from 45nm -> 45nm HKMG. And HKMG doesn't look like it is on the table, certainly we haven't heard a peep about it from AMD.

[enumae-k]

...And HKMG doesn't look like it is on the table, certainly we haven't heard a peep about it from AMD.

This is from the 2008 Financial Analyst Day - The Foundry Company (Doug Grose - November 13, 2008)

Page 21


According to Doug Grose, HKMG won't come until 32nm.

[Game_boy]

AMD seems to be targeting HKMG at low power, i.e. low clockspeeds. CPUs tend to run on the high-speed/high-power process (so 2-3GHz is typical) and GPUs tend to run on the medium-speed/medium-power process (so 500-800MHz is typical). Neither are likely to use a low-power process; that will be used for non-AMD wafer orders to compete with similar capabilities at other foundries. Most silicon demand isn't for cutting-edge nodes or high power.

[MTd2]

The german site claims Ph2 3.5GHz this year. If this is true then with HKMG Ph2 can be pushed toward 4GHz next year. Bulldozer is a late 2010 or 2011 thing on 32nm, so ... both can be true (or false).

I was NOT refering to 3.5. I am refing to the rumor about 4.0 GHz resurfaced.

[MTd2]

Anyway, considering that 32nm is available in the beginning of 2010, the 1st chips will be available 2 or 3 quarters later, just like it happened with TSMC and 40nm. It became available in the the mid of Q4, last year, and the first chips will be released at mid Q2.

Expect, then, the 1st 32nm HKMG, from AMD, at 3Q 2010, at the earliest.

[abinstein]

This is from the 2008 Financial Analyst Day - The Foundry Company (Doug Grose - November 13, 2008)

Page 21

According to Doug Grose, HKMG won't come until 32nm.

The problem with this logic is that while you believe the picture saying HKMG occurring 32nm, you completely disregard the time axis.

Do you think it is more possible for AMD to have 32nm or 45nm HKMG ready for production in early 2010?
The answer should be easy.

[enumae-k]

Do you think it is more possible for AMD to have 32nm or 45nm HKMG ready for production in early 2010?
The answer should be easy.

You need to look a little closer at the image, or slide. Bottom left there is a note...

Note: Left edge denotes first test chip
Right edge denotes qualcomplete

[raceb4us]

Something I had read earlier; even posted a link around here somewhere, stated AMD would move to HKMG sometime before 32nm. It makes sense to start moving to it once 45nm matures so they are not waging wars on multiple fronts. Maybe more click/clack than tick/tock.

[abinstein]

Do you think it is more possible for AMD to have 32nm or 45nm HKMG ready for production in early 2010?
The answer should be easy.

You need to look a little closer at the image, or slide. Bottom left there is a note...

Note: Left edge denotes first test chip
Right edge denotes qualcomplete

And what does this have anything to do with my question?
Do you think it's easier for AMD to complete 32nm HKMG qual before middle 2010, than 45nm?

If AMD could produce 32nm HKMG by middle 2010, then they definitely can do 45nm earlier. The only reason that they wouldn't is because the need for higher clock rate does not justify the cost of 45nm HKMG investment, and applying HKMG to the new 32nm node directly is seen as low risk (if either were not the case, they'd make 45nm HKMG first).

Back to topic.. I don't think it is possible for AMD to produce ph2 at 4.0GHz this year. Next year, maybe, whether on 32nm or 45nm.

[maduroutmb]

Did IBM/AMD even develop a non-HKMG 32nm process? 45nm with polysilicon gates was supposed to be impossible for bulk silicon and very difficult with SOI, but 32nm was widely understood to require HKMG.

[abinstein]

Did IBM/AMD even develop a non-HKMG 32nm process? 45nm with polysilicon gates was supposed to be impossible for bulk silicon and very difficult with SOI, but 32nm was widely understood to require HKMG.

You are correct. What I mean is whether HKMG comes out at 32nm or 45nm depends on the "risk" involving 32nm HKMG. If by middle 2010 AMD is confident in the yield of its 32nm HKMG, then it will go straight to the new 32nm node (with HKMG). OTOH, if the yield is not as expected, and 32nm has to be delayed, then AMD still could apply HKMG to 45nm to come in 2010, and delay 32nm to 2011, for example.

[Nylonox]

Did IBM/AMD even develop a non-HKMG 32nm process? 45nm with polysilicon gates was supposed to be impossible for bulk silicon and very difficult with SOI, but 32nm was widely understood to require HKMG.

You are correct. What I mean is whether HKMG comes out at 32nm or 45nm depends on the "risk" involving 32nm HKMG. If by middle 2010 AMD is confident in the yield of its 32nm HKMG, then it will go straight to the new 32nm node (with HKMG). OTOH, if the yield is not as expected, and 32nm has to be delayed, then AMD still could apply HKMG to 45nm to come in 2010, and delay 32nm to 2011, for example.

There is no way AMD could make such a process decision so late (i.e.-decide in mid 2010 to delay 32nm HKMG and instead ship a part on 45nm HKMG in 2010). Ignoring the major process development effort, any change of plans would require a re-layout of the device in question. They would have to develop both processes in paralell and have designs in-flight targeting each to change direction that quickly.

[abinstein]

There is no way AMD could make such a process decision so late (i.e.-decide in mid 2010 to delay 32nm HKMG and instead ship a part on 45nm HKMG in 2010).

The fact that (if 32nm HKMG works well) AMD go straight to 32nm HKMG doesn't exclude the possibility that they've been also playing along with 45nm HKMG all the time. AMD have always said (paraphrase) "we have the option to apply HKMG to 45nm." That to me means they're not waiting until next year to make decision whether to try HKMG on 45nm or not.

Ignoring the major process development effort, any change of plans would require a re-layout of the device in question. They would have to develop both processes in paralell and have designs in-flight targeting each to change direction that quickly.

That's precisely the "risk" part, isn't it? If we find AMD going full force to 32nm HKMG only, that should mean they are confident enough with this process that it won't flop. Otherwise, as you said, it's pretty risky. What if it doesn't work well in volume production? They better maker sure they still have the option to apply HKMG to 45nm in short time (I believe they do).

[enumae-k]

Do you think it is more possible for AMD to have 32nm or 45nm HKMG ready for production in early 2010?
The answer should be easy.

You need to look a little closer at the image, or slide. Bottom left there is a note...

Note: Left edge denotes first test chip
Right edge denotes qualcomplete

And what does this have anything to do with my question?

While removing my comments from the quote I erased some of your comment and had not quoted you correctly, I meant to include your initial comment.

The problem with this logic is that while you believe the picture saying HKMG occurring 32nm, you completely disregard the time axis.

The note asnswers this question.

In regards to your other question, well AMD and the Foundary Company have this as their roadmap so I don't need to speculate. Though the note also answers this question. AMD does not plan on having 32nm HKMG ready for production in early 2010, "Left edge denotes first test chip", so production would start in Q3 (?), with shipments possible in Q4 (?).

[MTd2]

More rumor:

[DragonSlaya04]

I dont think 1 GHz clock increase in 2 years is enough. They need either some kind of transition product to go from Ph2 to Bulldozer. Basically I'm saying they need some kind of uarch tweak within the next year. Perhaps HT or significant cache latency reduction.

[Boundless]

DragonSlaya04: > I don't think 1 GHz clock increase in 2 years is enough.

If you are referring to MTd2's allegedly leaked slide, let me ask:
what GHz per year gain have we seen since the X2 6400+ intro?.
Hint: we have yet to get back UP to 3.2 GHz.

67 MHz per month is quite aggressive for the current epoch in
x86 history, and is precisely the sort of in-Intel's-face "treadmill"
program I proposed in another thread.
_______
Assuming: the leaked table is legit, such a schedule can be kept,
and the 125W TDP is realistic (or isn't, but can be managed by
<secret cooling initiative deleted by moderator> ).

[maduroutmb]

You are correct. What I mean is whether HKMG comes out at 32nm or 45nm depends on the "risk" involving 32nm HKMG. If by middle 2010 AMD is confident in the yield of its 32nm HKMG, then it will go straight to the new 32nm node (with HKMG). OTOH, if the yield is not as expected, and 32nm has to be delayed, then AMD still could apply HKMG to 45nm to come in 2010, and delay 32nm to 2011, for example.

I was primarily replying to the slide, which seems to say that AMD is using HKMG only for low power processes. If anyone can clear that up (i.e. why the slide says what it does) I would deeply appreicate that.

[Edyros]

4 GHz in Q4 2010 is kinda late if you ask me. I thought that we would see such processor this year (some time in Q4). It's nice to see AMD improve on clock speeds, but we also want to see some other improvements to the core itself if they want to compete against Intel's 32nm Nehalem shrink.

As I said before, it seems to me that this is just a "stop gap" solution until they can get Bulldozer ready. How would they survive till 2011 with this strategy goes beyond my imagination.

[Das Tipitz]

4 GHz in Q4 2010 is kinda late if you ask me. I thought that we would see such processor this year (some time in Q4). It's nice to see AMD improve on clock speeds, but we also want to see some other improvements to the core itself if they want to compete against Intel's 32nm Nehalem shrink.

As I said before, it seems to me that this is just a "stop gap" solution until they can get Bulldozer ready. How would they survive till 2011 with this strategy goes beyond my imagination.

You will not see 4GHz on a Deneb 45nm processor ever, I guarantee it. Think about this for a moment: Overclockers can't get Deneb samples to 4GHz stable on air cooling at all, even with up to 1.6V or so. People can't get it Prime95 stable even with that insane amount of voltage. For AMD to release a 4GHz SKU, they need to get 4GHz 24/7/365 stable with ~1.3-1.35V capable of running the most demanding types of code imaginable, on a tiny air cooler, without a single error for years on end. No way is AMD going to improve to the point where what now takes ~1.6V to get benchmark stable is going to be 24/7/365 stable @ 1.35V on the same process technology. Don't say HKMG either, not only is it not on the latest roadmaps it won't make enough of a difference anyway. Intel isn't anywhere near a 4GHz quad SKU on their 45nm HKMG process and that is recognized as among/perhaps the best in the industry. You also have to consider whether it would be worth it for AMD to launch these SKUs, even if they can. I would bet you that AMD is more concerned about a SKU having good yields at this point than having the highest frequency possible. It is one thing to have a highly binned SKU if it is going to sell for a lot of money - i.e. Intel Extreme editions and AMD FX processors. But with Deneb, a 3.5GHz Deneb still can't really sell for much more than the price of the Q9650 ($316). Even a 4GHz model, how much could they sell it for? Maybe $450-500 or so. A Core i7 940 is going to be just as fast or sometimes even faster than a 4GHz Deneb.

If you're going to start talking about overclocking, then remember Intel can do the same thing. Good E0 E8xxx CPUs can do 4GHz undervolted with reasonable power levels. Core i7 can easily hit 3.6-3.8GHz on stock voltage. Especially with a future 45nm revision of Bloomfield imminent there is no doubt that Intel can push clocks just as well as AMD can. With the Core 2 family, certainly a higher SKU than the Q9650 is very possible.

AMD might launch 6 core Istanbul processors for the desktop market. A 3GHz or so Istanbul CPU sounds very reasonable for the 2010 timeframe. A 6 core Istanbul should be able to compete against quad-core Nehalems in multithreaded applications.

You also have to remember Intel's lineup is slowing down a bit, mainstream Nehalem on 45nm looks like it might not even happen at this point. The current rumors point towards the dual-core mobile Nehalem on a MCM package with a GPU has been canceled and Intel will wait until mid 2010 to release the 32nm part. I would not be surprised with the economy and success of the Core 2 family that you see LGA 1156 not even materialize until 2010, perhaps not until 32nm processors are ready. Now once that happens then AMD is in trouble, but at least then Orochi is not far away. Until then, AMD might actually not be doing too bad competing against Core 2 processors in the low-end/midrange market.

Personally I think the key for AMD is to move the rest of their product line to 45nm and focus on the low-end parts. Rana dual cores with 2x1MB L2 should come in and do very well in the <$150 market where you are competing against stripped down Core 2 CPUs. The X2 7750 is quite a good part at the ~$70-80 price point, release a 3.2GHz 45nm chip with 1MB L2 per core for $100 and you've captured that market. Propus with its very small die will allow AMD to have cheaper quads than Intel, and at those low end price points all Intel has are very stripped down C2Qs w/ 4MB L2. And now that frequencies are so much higher with 45nm, tri-cores should render the E8xxx dual cores rather useless. I can foresee AMD having a iron grip on the <$200 market if they play their cards right. >$200, it's not going to be a good 2 years or so but if AMD launches a 6 core CPU in that timeframe then they can actually compete against Nehalem CPUs in some regard.

[abinstein]

You will not see 4GHz on a Deneb 45nm processor ever, I guarantee it. Think about this for a moment: Overclockers can't get Deneb samples to 4GHz stable on air cooling at all, even with up to 1.6V or so.

Uh?? Don't you know there is already a company selling ph2 @4.0GHz OC?
http://www.rainrecording.com/products/element-x/

[Boundless]

abinstein: > Don't you know there is already a company selling ph2 @4.0GHz OC?

It's even Best Buy SKU: 9196602

"Element Digital Audio - Now available in ludicrous speed"

Love it.
I might have called it "Next Level Cooling (TM)"
_______
Does it include an oversize black Dark Helmet?

[AussieFX]

You will not see 4GHz on a Deneb 45nm processor ever, I guarantee it. Think about this for a moment: Overclockers can't get Deneb samples to 4GHz stable on air cooling at all, even with up to 1.6V or so.

Uh?? Don't you know there is already a company selling ph2 @4.0GHz OC?
http://www.rainrecording.com/products/element-x/

He doesn't get out much.

http://www.ocxtreme.org/forumenus/showt ... 375&page=2