Phenom II info leaks out, AMD hints at something

[RiverRicer]

Originally posted by: BTRY B 529th FA BN
It will get better at 32nm when they introduce High-K

I was interested in how AMD managed to avoid HK/MG at 45nm and, in fact, postpone it until late 45nm or even 32nm. I found this article that describes the transistor improvements AMD has made from the 65nm tech. Interesting read.........

LINK

 

[BTRY B 529th FA BN]

Originally posted by: RiverRicer

Originally posted by: BTRY B 529th FA BN
It will get better at 32nm when they introduce High-K

I was interested in how AMD managed to avoid HK/MG at 45nm and, in fact, postpone it until late 45nm or even 32nm. I found this article that describes the transistor improvements AMD has made from the 65nm tech. Interesting read.........

LINK

Yes, very interesting read, thanks!

 

[Martimus]

Originally posted by: BTRY B 529th FA BN

Originally posted by: RiverRicer

Originally posted by: BTRY B 529th FA BN
It will get better at 32nm when they introduce High-K

I was interested in how AMD managed to avoid HK/MG at 45nm and, in fact, postpone it until late 45nm or even 32nm. I found this article that describes the transistor improvements AMD has made from the 65nm tech. Interesting read.........

LINK

Yes, very interesting read, thanks!

I agree. That was very nice, and it does make sense now why you would want to push HKMG to a smaller process - to lower power consumption (which was a major problem with Barcelona).

 

[Sureshot324]

Originally posted by: Ocguy31

Originally posted by: nyker96
they seem to hit above 4 ghz OC!

You are looking at a marketing slide. I would reserve my judgement.

It's still impressive considering there's no way you could hit 4ghz on air on a current Phenom, even with the best possible golden sample.

 

[myocardia]

Originally posted by: SlowSpyder
I think AMD's business model wasn't as bad as it sounds here. AMD obviously used future upgradeability as a reason to buy their hardware. It was a great selling point, to buy an AM2 socket A64 knowing that you'd be able to upgrade to a Phenom quad core down the road. But, you didn't get all of the functionality/performance without a full platform upgrade. This would prompt the enthusiast crowd to upgrade their boards with the CPU's making everyone happy. And having that upgrade path may have been a deciding factor for some people making the initial purchase... Asus I'm sure would love to sell you two boards, but would rather sell you one board then zero.

I agree with this. For Joe sixpack, it was a huge selling point. And don't we all realize that Joe sixpack is who these companies are marketing towards?

Of course the average AM2 board wasn't built with 125-140 watt CPU's in mind. Hell, most early AM2+ boards weren't built with that in mind.

This wasn't the motherboard manufacturers fault. The fault here falls squarely into AMD's lap, although admittedly, I don't blame them for doing it. I'm sure I would have done the same thing, had I been in their position. And honestly, they weren't lying. If a motherboard only supports half of the Phenoms made, it's still Phenom-compatible. AMD never made the claim that all AM2 boards would be compatible with all Phenoms, they merely said that the Phenoms would be socket-compatible, which they were.

 

[Idontcare]

Originally posted by: myocardia

Originally posted by: SlowSpyder
Of course the average AM2 board wasn't built with 125-140 watt CPU's in mind. Hell, most early AM2+ boards weren't built with that in mind.

This wasn't the motherboard manufacturers fault. The fault here falls squarely into AMD's lap, although admittedly, I don't blame them for doing it. I'm sure I would have done the same thing, had I been in their position. And honestly, they weren't lying. If a motherboard only supports half of the Phenoms made, it's still Phenom-compatible. AMD never made the claim that all AM2 boards would be compatible with all Phenoms, they merely said that the Phenoms would be socket-compatible, which they were.

With that in mind, and the idea floating here that Phenom II's can clock to 4Ghz on air, why didn't AMD create a higher clocked higher-ASP SKU with an allowed higher TDP of say 140 or 150W?

After all if people are going to overclock to 4GHz anyways (as AMD is touting you can) then AMD must acknowledge these folks are using existing mobo's certified at-best to use 140W chips, yes?

 

[Negronpope]

Looks pretty close to what AMD's hinted at. Phenom II X4 with 8MB of cache is probably the Deneb core while the Athlon X4 with a a 2MB cache is probably the Propos core. It makes sense to label the processors with lower amounts of cache as Athlons. Why give up a well recognized name?

 

[Cookie Monster]

Its not probably. It is.

 

[Martimus]

Originally posted by: Idontcare

Originally posted by: myocardia

Originally posted by: SlowSpyder
Of course the average AM2 board wasn't built with 125-140 watt CPU's in mind. Hell, most early AM2+ boards weren't built with that in mind.

This wasn't the motherboard manufacturers fault. The fault here falls squarely into AMD's lap, although admittedly, I don't blame them for doing it. I'm sure I would have done the same thing, had I been in their position. And honestly, they weren't lying. If a motherboard only supports half of the Phenoms made, it's still Phenom-compatible. AMD never made the claim that all AM2 boards would be compatible with all Phenoms, they merely said that the Phenoms would be socket-compatible, which they were.

With that in mind, and the idea floating here that Phenom II's can clock to 4Ghz on air, why didn't AMD create a higher clocked higher-ASP SKU with an allowed higher TDP of say 140 or 150W?

After all if people are going to overclock to 4GHz anyways (as AMD is touting you can) then AMD must acknowledge these folks are using existing mobo's certified at-best to use 140W chips, yes?

You keep saying this, but it isn't like Intel is doing it either. There has to be a reason both companies are being conservative with their processor binning. I would guess that it is to keep yields high, and avoid pushing down the bottom binning prices by creating higher binned processors (since the maximum price is pretty much set based on what individuals will pay for a processor.)

EDIT: Another reason is likely that it costs more money to test for higher binned processors, and it also requires that the chip maker gets MB support for the higher power consumption. It is one tihng when MB makers get RMA's for people overclocking their processors and killing the board; it is another thing entirely when they are killing it at stock settings (That won't just cost you a whole lot more replacing boards, it will give you a huge PR problem.)

Considering that you worked in this environment, I am surprised to see you asking this question. AMD had to run their processors at the bleeding edge during their last generation to even try to compete with Intel, but if they don't have to, I am sure they won't to maximize their profits. To be honest, the real reason I posted this is to get you to explain why you feel this is so strange from AMD, when it seems to be modus operandi from Intel? You have much more insight into how this stuff works than I will ever have, so I value your opinion on this.

 

[daveybrat]

I'm not too needy.....I just want a processor that plays my games, makes me dinner, and brings my porn to life!

 

[cubeless]

the oc market is a very small piece of the processor world... i'm sure they might think a little like the old racing 'win on sunday sell on monday' as far as advertising goes, but joe sixpack in general isn't worried about the headroom a chip has...

and if you get 'lucky', like it seems intel did with architecture/yield on c2, then you release faster chips on schedule and don't mess up your roadmap... it's a win-win for them: enthusiasts get to have fun oc'ing and joe sixpack has more power then he ever needs... and the r&d department has all the time they need to get the next thing out...

 

[BTRY B 529th FA BN]

but what about Joe the plumber

 

[cubeless]

f**k him... he's rich and buys highend prebuilts...

edited for niceness...

 

[Idontcare]

Originally posted by: Martimus
You keep saying this, but it isn't like Intel is doing it either. There has to be a reason both companies are being conservative with their processor binning. I would guess that it is to keep yields high, and avoid pushing down the bottom binning prices by creating higher binned processors (since the maximum price is pretty much set based on what individuals will pay for a processor.)

Intel is in a position of strength, they get to dictate the performance necessary for gaining entry to any given ASP tier. You need a part that performs on par to a 3.2GHz i7 to gain access to the portion of the ASP supply/demand curve that exists above $1000.

Intel's actions are entirely expected and have guided the pricing/binning model as it has existed for more than 3 decades in this industry segment. At times it was AMD playing this strong hand and Intel was clamoring to bin higher parts to access the higher ASP territory.

What doesn't ring true is AMD's actions in this regard. If AMD really could bin out 4GHz Phenom II's then they would. It really is as simple as that. They are not in a position with the luxury to decide to sell 4GHz $1000 chips as 3Ghz $300 chips.

The arguments that it isn't worth the efforts, or that the mobo makers aren't inclined to create higher TDP mobos, just doesn't ring true either. No one passes on the opportunity to make $2 and instead just opts to make $1. Businesses don't successfully operate that way. The quad-father and FX-74 is an example.

Given the historic approach that has guided binning and ASP decisions in the past 20 years, I am compelled to come to one of two conclusions - (1) the pretense that a healthy supply of Phenom II's binnable at 3.5+GHz is false and just BS, or (2) a forthcoming higher-binned higher-ASP part is imminent and perhaps such a flagship/premiere CPU is intentionally being held out for the debut of AM3 platform (FX only on AM3, some such marketing situation).

I just don't believe that AMD is sitting on highly binnable silicon but for lack of effort just simply don't want to go to the trouble of selling the silicon at higher points.

Consider this - what sense would it make for Intel to limit the top speedbin for i7 to 2GHz (even though it can clock to 4GHz) in a world that already has 3.2GHz Yorkfields? It would make no sense at all if Intel did it, and it makes no sense at all for AMD to do it...so I am waiting for the other shoe to drop (AM3/FX chips at 3.4GHz or higher...or proof from consumers that the 3GHz chip struggles to get above 3.4GHz, meaning the supply of such parts would be too small to create its own SKU)

 

[Martimus]

Originally posted by: Idontcare

Originally posted by: Martimus
You keep saying this, but it isn't like Intel is doing it either. There has to be a reason both companies are being conservative with their processor binning. I would guess that it is to keep yields high, and avoid pushing down the bottom binning prices by creating higher binned processors (since the maximum price is pretty much set based on what individuals will pay for a processor.)

Intel is in a position of strength, they get to dictate the performance necessary for gaining entry to any given ASP tier. You need a part that performs on par to a 3.2GHz i7 to gain access to the portion of the ASP supply/demand curve that exists above $1000.

Intel's actions are entirely expected and have guided the pricing/binning model as it has existed for more than 3 decades in this industry segment. At times it was AMD playing this strong hand and Intel was clamoring to bin higher parts to access the higher ASP territory.

What doesn't ring true is AMD's actions in this regard. If AMD really could bin out 4GHz Phenom II's then they would. It really is as simple as that. They are not in a position with the luxury to decide to sell 4GHz $1000 chips as 3Ghz $300 chips.

The arguments that it isn't worth the efforts, or that the mobo makers aren't inclined to create higher TDP mobos, just doesn't ring true either. No one passes on the opportunity to make $2 and instead just opts to make $1. Businesses don't successfully operate that way. The quad-father and FX-74 is an example.

Given the historic approach that has guided binning and ASP decisions in the past 20 years, I am compelled to come to one of two conclusions - (1) the pretense that a healthy supply of Phenom II's binnable at 3.5+GHz is false and just BS, or (2) a forthcoming higher-binned higher-ASP part is imminent and perhaps such a flagship/premiere CPU is intentionally being held out for the debut of AM3 platform (FX only on AM3, some such marketing situation).

I just don't believe that AMD is sitting on highly binnable silicon but for lack of effort just simply don't want to go to the trouble of selling the silicon at higher points.

Consider this - what sense would it make for Intel to limit the top speedbin for i7 to 2GHz (even though it can clock to 4GHz) in a world that already has 3.2GHz Yorkfields? It would make no sense at all if Intel did it, and it makes no sense at all for AMD to do it...so I am waiting for the other shoe to drop (AM3/FX chips at 3.4GHz or higher...or proof from consumers that the 3GHz chip struggles to get above 3.4GHz, meaning the supply of such parts would be too small to create its own SKU)

I agree with your points completely that AMD would have a 3.2GHz part if they could make one that fit within their power envelope. But I could see why they would stay away from an FX set-up right now, because I pretty much doubt they can divert resources from other areas to truly design that platform at launch. They have to focus on the most important portions of their business to succeed: which I believe to be their manufacturing process, and the integrated performance of the actual hardware. So even if they can bin a 150W 3.6GHz quad core, the extra engineering required to put together such a platform in January may be either too expensive for a proper ROI (especially if it Intel also has plenty of headroom - so they could just do the same thing to make your effort null and void) or they may not have the resources available to do that period due to all the effort put just getting the 45nm chips out. I will compare this to something I am intimately familiar with. A program that I worked on had a requirement that a volley of 5 shots would have a 100% kill percentage on a particular type of target. The gun was able to be even more accurate than that, so the contractor decided to put an arbitrary requirement of a 100% kill percentage on a single shot. The problem with this is that they spent resources on this (testing, software rewrite, engineering) and were able to do it at a major cost to the government, but they fell behind in many other areas and the program almost was canceled. My point is that when you have issues in other areas, you don't continue to work on your position of strength after you have already met your goal; unless you have the resources to work on both you problem areas and your position of strength. It is possible that AMD believes that they meet the initial goal set for the Deneb speed binning to compete with Mainstream Intel components, and that they will come back with an FX platform (as was rumored after the first spin of Deneb) later with the introduction of AM3. I seem to remember a member of AMD even saying that they were considering doing just that a few months back.

Wow, that was a rather large block of text. We have about 6 1/2 weeks to find out how well the Phenom II platform actually overclocks. It wouldn't surprise me at all to find out that they can only overclock a small amount, or if they overclock very well. I do doubt that they will compete much with the top speed i7 processors though.

 

[Idontcare]

Originally posted by: Martimus
It wouldn't surprise me at all to find out that they can only overclock a small amount, or if they overclock very well. I do doubt that they will compete much with the top speed i7 processors though.

I very much see your points. They are valid, no disagreement there. Just unprecedented IMO, which makes me pause for thought when I consider the situation.

I also can't discount the point made above by piesquared that binning out a 150W+ TDP SKU would be marketing fodder for Intel. That makes sense to me too.

But riddle me this - when was the last time a flagship (highest binned) CPU from an underdog turned out to overclock well? That's kinda a new novelty for this industry if it turns out to be true.

 

[Extelleron]

Originally posted by: Idontcare

Originally posted by: Martimus
It wouldn't surprise me at all to find out that they can only overclock a small amount, or if they overclock very well. I do doubt that they will compete much with the top speed i7 processors though.

I very much see your points. They are valid, no disagreement there. Just unprecedented IMO, which makes me pause for thought when I consider the situation.

I also can't discount the point made above by piesquared that binning out a 150W+ TDP SKU would be marketing fodder for Intel. That makes sense to me too.

But riddle me this - when was the last time a flagship (highest binned) CPU from an underdog turned out to overclock well? That's kinda a new novelty for this industry if it turns out to be true.

Well first of all there is such a big difference between what we can call 24/7 and what AMD or Intel can sell, and it is not just about power consumption. Pretty much the industry expects and demands that when they buy a processor, it can be expected to work error free at the rated speed for 10 years or more. What does this mean? Well for one, AMD/Intel must guarantee stability in such a way that there is 0% chance of error in the absolute worst of conditions (Linpack 64-bit as the benchmark I would think) over a period of 100% load over years....versus what we consider stable, Prime95 for 24 hours at the most.

Not only that, but AMD/Intel are limited by voltages that will last long-term; while we can pump 1.4V in a 45nm CPU to reach that OC and be OK for the most part, AMD can maybe go up to 1.25V to reach the clock targets it needs and still be able to guarantee that the Deneb chip you buy will not be damaged or degraded.

With AMD, there is a third factor; the 4GHz clock likely includes the boost from ACC, something that could not be a factor in a retail chip.

I'm sure you know all of this already but it puts a pretty big dent into your argument that AMD should be able to extend retail clocks beyond 3GHz if a 4GHz overclock is possible. Also consider one last point that I think is very important to consider; the point of market position.... if AMD wants to launch a 3.2GHz Deneb, they want it to sell in a higher slot in the market than the 3.0GHz 940 (obviously). Yet neither the 3.0/3.2GHz Denebs will be able to command a significantly higher price than the Core i7 920 given its far superior performance..... so if AMD prices the 940 at the level of the i7 920..... where does the hypothetical 3.2GHz 960 fit? AMD wouldn't be able to command a higher price enough for the selective binning to be worth it and thus concentrates on selling processors that are competitive and high yield.

 

[myocardia]

Originally posted by: Idontcare
I also can't discount the point made above by piesquared that binning out a 150W+ TDP SKU would be marketing fodder for Intel. That makes sense to me too.

Agreed.

But riddle me this - when was the last time a flagship (highest binned) CPU from an underdog turned out to overclock well? That's kinda a new novelty for this industry if it turns out to be true.

Hah, I love the way you word things sometimes. Anyway, the last time that happened was at the introduction of the Core 2 Duo, which wasn't all that long ago, in retrospect.


edit: Oh yeah, while I don't think that a large percentage of Denebs will be clocking to 4Ghz and beyond, what if AMD is just waiting on the board manufacturers to catch up, as far as the amount of power the boards can supply? AMD took quite a bit of flak over non-overclocked Phenoms killing motherboards.

 

[Raqia]

http://en.expreview.com/2008/11/18/a...html#more-1307

These are platform benches, so I don't think they're even using the same graphics cards in those gaming scores.

On the CPU tests, we're seeing what a 15% boost in clocks speed from 2.6 Ghz Phenom to the 3.0 Ghz Phenom 2 does: 40% scaling in POVRay, Quicktime encoding, and OfficeFX (whatever that is) scores seem pretty promising.

 

[BTRY B 529th FA BN]

And I thought the i7 was the most interesting thing for 08

 

[ArchAngel777]

Originally posted by: myocardia
Hah, I love the way you word things sometimes. Anyway, the last time that happened was at the introduction of the Core 2 Duo, which wasn't all that long ago, in retrospect.

Notice the bolded in his statement

Originally posted by: Idontcare
But riddle me this - when was the last time a flagship (highest binned) CPU from an underdog turned out to overclock well? That's kinda a new novelty for this industry if it turns out to be true.

While the Core 2 Extreme did overclock well (2.93 to 3.73), it wasn't an underdog. So he is totally right... The underdog from past history always pushes the envelope. Remember the PIII 1.13Ghz? AMD was destroying or at least on par during that time so they released a flaky chip in attempt to stay competitive. As I recall, the P4 1.4Ghz came out at that time and was a real stinker in performance. Most of the X2's highest bin (3.4Ghz? Or is it 3.2?) are pretty much at their limit.

 

[Idontcare]

Originally posted by: Raqia
http://en.expreview.com/2008/11/18/a...html#more-1307

These are platform benches, so I don't think they're even using the same graphics cards in those gaming scores.

On the CPU tests, we're seeing what a 15% boost in clocks speed from 2.6 Ghz Phenom to the 3.0 Ghz Phenom 2 does: 40% scaling in POVRay, Quicktime encoding, and OfficeFX (whatever that is) scores seem pretty promising.

Welcome to the AT forums!

What is OfficeFX?

OfficeFX is a seamless software addition to Microsoft® PowerPoint® that lets you transform your traditional presentations into engaging rich media events. Whether novice or expert, if you know PowerPoint, you can use OfficeFX effectively in minutes. A few quick selections will instantly energize your presentations with dynamic 3D backgrounds, realistic lighting, shading, and a wide range of animations and transitions from subtle and sophisticated to eye catching and high impact. Using OfficeFX from directly within PowerPoint, your products, brands, and messages take on the polish of a professional broadcast or a high end corporate event, while you control the content in a tool you already know.

Text

I'm guessing the pov-ray test bench they run is not disk I/O intensive which is why they get the nice speedup above and beyond the clockspeed increase.

Looking at a fairly I/O bound bench like Nero Recode and things are a little more in-line with what you expect from the sheer clockspeed boost of 2.6GHz -> 3GHz. A 20% gain is nice, the extra 5% from IPC (faster L3$?) boost and all.

 

[Martimus]

Originally posted by: Idontcare

Originally posted by: Martimus
It wouldn't surprise me at all to find out that they can only overclock a small amount, or if they overclock very well. I do doubt that they will compete much with the top speed i7 processors though.

I very much see your points. They are valid, no disagreement there. Just unprecedented IMO, which makes me pause for thought when I consider the situation.

I also can't discount the point made above by piesquared that binning out a 150W+ TDP SKU would be marketing fodder for Intel. That makes sense to me too.

But riddle me this - when was the last time a flagship (highest binned) CPU from an underdog turned out to overclock well? That's kinda a new novelty for this industry if it turns out to be true.

I agree, although one thing that leaves me to believe that there is still some headroom is the naming convention for the processor. They left a lot of upward options for more speed bins. They left almost no options for extra speed bins with the original Phenom processors, but they didn't have much headroom on those at all. That being said, we will find out the truth soon enough.

 

[SlowSpyder]

Originally posted by: Raqia
http://en.expreview.com/2008/11/18/a...html#more-1307

These are platform benches, so I don't think they're even using the same graphics cards in those gaming scores.

On the CPU tests, we're seeing what a 15% boost in clocks speed from 2.6 Ghz Phenom to the 3.0 Ghz Phenom 2 does: 40% scaling in POVRay, Quicktime encoding, and OfficeFX (whatever that is) scores seem pretty promising.

Pretty interesting, but unfortunately we don't know if the comparrison there is a Phenom 9950 with a 780G integrated gpu vs. a Denab @ 3GHz with a 4870 or apples to apples.

I am looking forward to getting one of these Denabs to tinker with when they are available. I'm pretty happy with my current Phenom as a gamer, but the hardware enthisast side of me likes to upgrade to new stuff for no good reason. This annoys my wife.

 

[Negronpope]

So, a really foolish question. If these processors are supposed to be out in the first quarter of '09, shouldn't we have already seen early spins of the silicon out in the wild by now? After all it's mid-November. So, why hasn't anyone seen these yet?