So what is the next step for AMD?

[phexac]

It's current CPU offerings, quad cores, triple core, and dual cores cannot compete with Intel. In a relatively short period of time, Intel will be rolling out its new architecture chip, taking its lead even further. What is AMD's response. Has there been word of new architecture, promise of faster clocked Phenoms, anything at all?

I am just curious what the next step for AMD is since we clearly know where Intel stands now and where they are going (Nehalem). We know where AMD stands now (where it does not want to be), but where is it going given the shitty situation?

 

[GFORCE100]

The next step for AMD is 45nm of what they already have + some more cache bringing the total up to 6MB Vs 4MB as of now.

Even at 45nm yields may be poor initially or poor/average for whenever 32nm comes along. AMD's 90nm process is actually better than its 65nm but they can't make their Phenoms using it even if they wanted to (which I'm sure they would if they could do magic). The die size at 90nm would be way high in the region of 400mm2 let alone power consumption.

It's worth noting that AMD can't pack transistors as tightly as Intel can, even at 65nm. Intel has a 30% or so lead in this respect alone. Take a look at say the Core 2 Quad 65nm Vs the Phenom. Intel has 572m transistors, AMD has 450m but Intel's dize size is only 1mm2 larger than the Phenom with 2x the amount of cache onboard. The fact it uses a lot less power as a result is also impressive.

AMD really thought they were God for a while in the CPU Market and now they're paying the prize. They concentrated on making fun of Intel with their amusing marketing games instead of concentrating on tomorrow as in business just because the sun shines today doesn't mean it will shine tomorrow. Their choice, they wanted this outcome as they didn't do anything to prevent it.

At the very least they should have stopped acting God and not gone for the monolithic quad core design given they have a poor track record in terms of engineering/manufacturing.

Had they done that, and had they stopped acting I'm the boss like and not bought ATI then things would be looking a lot better right now. I personally have a feeling Hector amongst others at AMD might be somewhat good at business but he totally lacks an understanding of the CPU market. Either that or the guy really has no power within AMD at board room meetings to knock sense into the other few who want to play God and think they can make a name for themselves having just left university.

Quite sad what's happend to AMD but there you go, they asked for it and quite frankly they got it.

Their credit standing is very low and here's them thinking they'll build a $2b fab in NY state. Geez guys, if you didn't spend on ATI then yes you would but now they're even having to lay of 3000 staff because money is do darn tight. The CEO of a company is supposed to protect the interest of the company in which they control and I don't see Hector doing much of this over recent years. Where's the long term financial and business thinking let alone product roadmap/engineering thinking?

I doubt AMD will stand back on its feet by itself, and their Phenoms won't be anything exciting for quite a while yet. It's not enough to bring in the bread and butter.

 

[Zenoth]

Basically AMD is screwed ever since C2D came out, but that ain't new of course, they've had their time.

 

[Idontcare]

Originally posted by: phexac
It's current CPU offerings, quad cores, triple core, and dual cores cannot compete with Intel. In a relatively short period of time, Intel will be rolling out its new architecture chip, taking its lead even further. What is AMD's response. Has there been word of new architecture, promise of faster clocked Phenoms, anything at all?

It's a big question. CPU line-up wise thye have Shanghai (servers) and Deneb/Propus (desktop) 45nm chips coming out in Q4 this year. Q4 is a long way away though.

In the meantime we have B3 stepping of Barcelona (servers) and Phenom (desktop) 65nm chips. No one expects another stepping of the 65nm SKU's. Another speedbin (or two) might come out of the fab as those transistors are tweaked and optimized in the learning cycles to come.

There is a platform (spyder) push from AMD right now. Batmang, Lopri, and CTho9305 among others here can hopefully fill you in on the deal there. I haven't paid it much attention to AMD's non-CPU business lately.

http://www.anandtech.com/cpuch...ts/showdoc.aspx?i=3288

Likewise you can't forget the other half of AMD, it's GPU business, which also includes its chipset business. I don't game much but AMD's 3870 X2 is not too shabby from the reviews.

http://www.anandtech.com/video/showdoc.aspx?i=3266

 

[Foxery]

Originally posted by: GFORCE100
The next step for AMD is 45nm of what they already have + some more cache bringing the total up to 6MB Vs 4MB as of now.

We already know they can't do this until Q4.

AMD really thought they were God for a while in the CPU Market and now they're paying the prize.
At the very least they should have stopped acting God and not gone for the monolithic quad core design given they have a poor track record in terms of engineering/manufacturing.

...

Their credit standing is very low and here's them thinking they'll build a $2b fab in NY state. Geez guys, if you didn't spend on ATI then yes you would but now they're even having to lay of 3000 staff because money is do darn tight.

You have some very amusing ideas, which seem to stem from a complete lack of foresight or ability to grasp the big picture.

Native quad-core design and buying ATI are both part of long-term plans for future architectures. (They've admitted to overpaying for ATI, but they still want it.) Phenom also isn't a bad product, as compared to the previous generation of P4s and Athlons; it only looks so next to Core2 because Intel pulled out all the stops on their new product.

The short answer for the OP is, "they need to not die in 2008," and make sure to match Intel in 2009 with the next generation. They have plenty of plans, but none of them can get here until at least this Christmas.

 

[hooflung]

Since most of the market that matters doesn't care about overclocking, reports of AMD's doom is highly exaggerated. They are still working strong in the server markets and on desktops from OEMs. The Phenom is a pretty popular consumer PC too. A lot of gamers living in Australia, the Mediteranean and Asia do not overclock as much and stock clocks matter more as well as price.

The biggest blunder AMD has to overcome right now is making sure 780G chipsets are being sold with 5 phase power so the 9850 can be used on them.

 

[bryanW1995]

I think that they're going from 4x500mb L2 and 2mb L3 to 4x500mb L2 and 6mb L3 for a total of 8 mb cache.

 

[v8envy]

Do exactly what they're doing.

1. Scale down and shrink in the short term, try to find niches for current products. Focus on selling fewer units of product at higher prices (the right side of the supply/demand graph) as opposed to slashing prices and losing money on each unit but trying to make it up in volume. A 10% work force reduction is a good start.
2. Create products which compete at the high end, using enthusiast 'halo effect' to move volume at the mid range and low end
3. Continue to support big OEMs in spite of #1
4. Resume growth

AMD is doing this now. Previously they tried to clog the channel with product and slash prices. That didn't get them anywhere. So, expect AMD's products to offer less bang for buck than Intel counterparts in the short term. That doesn't mean an AMD based system will be more expensive than an Intel based one -- quite the opposite. They can still provide relatively low performance OEM boxes with a much lower total pricetag than the cheapest Intel ones.

There's obvious risk with this strategy -- the perception that you're a 'generic brand' suitable only for people on an extreme budget. It's hard to charge a premium for your products when you're perceived that way. But it beats the alternative.

 

[GFORCE100]

Originally posted by: Foxery

Originally posted by: GFORCE100
The next step for AMD is 45nm of what they already have + some more cache bringing the total up to 6MB Vs 4MB as of now.

We already know they can't do this until Q4.

AMD really thought they were God for a while in the CPU Market and now they're paying the prize.
At the very least they should have stopped acting God and not gone for the monolithic quad core design given they have a poor track record in terms of engineering/manufacturing.

...

Their credit standing is very low and here's them thinking they'll build a $2b fab in NY state. Geez guys, if you didn't spend on ATI then yes you would but now they're even having to lay of 3000 staff because money is do darn tight.

You have some very amusing ideas, which seem to stem from a complete lack of foresight or ability to grasp the big picture.

Native quad-core design and buying ATI are both part of long-term plans for future architectures. (They've admitted to overpaying for ATI, but they still want it.) Phenom also isn't a bad product, as compared to the previous generation of P4s and Athlons; it only looks so next to Core2 because Intel pulled out all the stops on their new product.

The short answer for the OP is, "they need to not die in 2008," and make sure to match Intel in 2009 with the next generation. They have plenty of plans, but none of them can get here until at least this Christmas.

Who says I said everything I could possibly say about AMD in my post above? I'm very aware of the big picture so to speak but how they executed this plan and the timeframe they chose is one of the main reasons why they're in the mess as of now.

I didn't say Phenom is a bad product, it's a good product but their manufacturing process is subpar to what they expected. Intel on the other hand have excellent CPU's now and also an excellent manufacturing process. Who wins, it's clear whichever way you put it.

 

[jiffylube1024]

Originally posted by: GFORCE100
The next step for AMD is 45nm of what they already have + some more cache bringing the total up to 6MB Vs 4MB as of now.

I'm pretty sure AMD is going to 8MB of total cache on 45nm. They're keeping the L2 cache at 512KB * 4 and upping the L3 cache from 2MB to 6MB, meaning 8MB total. No word yet on whether the L3 cache will get a boost from its current fixed 2 GHz speed or not.

Originally posted by: GFORCE100
Even at 45nm yields may be poor initially or poor/average for whenever 32nm comes along. AMD's 90nm process is actually better than its 65nm but they can't make their Phenoms using it even if they wanted to (which I'm sure they would if they could do magic). The die size at 90nm would be way high in the region of 400mm2 let alone power consumption.

It's worth noting that AMD can't pack transistors as tightly as Intel can, even at 65nm. Intel has a 30% or so lead in this respect alone. Take a look at say the Core 2 Quad 65nm Vs the Phenom. Intel has 572m transistors, AMD has 450m but Intel's dize size is only 1mm2 larger than the Phenom with 2x the amount of cache onboard. The fact it uses a lot less power as a result is also impressive.

AMD's 90nm process is more mature than its 65nm, but there is no question that 45nm is needed to go forward. Initially 45nm may be unspectacular, but once they get the new refined strained silicon processes, low-k and high-k processes online, they should be able to get clockspeeds much higher, just like Intel has.

As for the second part, I didn't realize AMD couldn't shrink its transistors as much as Intel could on the same process. I wonder if this has anything to do with the on-die memory controller of the AMD chips.

Originally posted by: GFORCE100
AMD really thought they were God for a while in the CPU Market and now they're paying the prize. They concentrated on making fun of Intel with their amusing marketing games instead of concentrating on tomorrow as in business just because the sun shines today doesn't mean it will shine tomorrow. Their choice, they wanted this outcome as they didn't do anything to prevent it.

At the very least they should have stopped acting God and not gone for the monolithic quad core design given they have a poor track record in terms of engineering/manufacturing.

AMD just doesn't have the resources that Intel does. AMD needs to put all of its eggs in one basket whenever they make a new family of chips. While Intel was flopping with the P4 architecture, AMD's Athlon64 family reigned supreme.

However, Intel had its Israeli-designed Pentium M core to modify and reconfigure as the Core architecture. AMD has no such fallback plan if one of their designs flops.

Originally posted by: GFORCE100
Had they done that, and had they stopped acting I'm the boss like and not bought ATI then things would be looking a lot better right now. I personally have a feeling Hector amongst others at AMD might be somewhat good at business but he totally lacks an understanding of the CPU market. Either that or the guy really has no power within AMD at board room meetings to knock sense into the other few who want to play God and think they can make a name for themselves having just left university.

No question Hector Ruiz's leadership is in question right now. However, you can't say for certain that had they not bought ATI they would be doing better now (except financially). They'd still be way late to the table on 45nm, DDR3 support, etc.

Originally posted by: GFORCE100
Quite sad what's happend to AMD but there you go, they asked for it and quite frankly they got it.

Their credit standing is very low and here's them thinking they'll build a $2b fab in NY state. Geez guys, if you didn't spend on ATI then yes you would but now they're even having to lay of 3000 staff because money is do darn tight. The CEO of a company is supposed to protect the interest of the company in which they control and I don't see Hector doing much of this over recent years. Where's the long term financial and business thinking let alone product roadmap/engineering thinking?

The ATI purchase was to consolidate their chipset/motherboard market and ensure they had stable platforms available for the forseeable future. Getting ATI's graphics division was just a bonus of sorts (at least that's how I saw the purchase). With Intel's Core 2 architecture becoming so popular, designing chipsets for AMD took a back seat to Intel chipsets, so AMD had to make sure they had stable platforms to run their CPUs on in the future.

Plus, the notebook market is growing by leaps and bounds, and you need solid integrated graphics to do well in that market. Intel has its home-grown GMA series (which is among the most installed graphics cards in the world), while AMD had no such graphics platform. Now they do, and that is very important to be a player in the laptop market (though they are still getting their asses handed to them currently, as their power envelopes can't touch Intel's in the mobile world, while keeping competitive in terms of performance).

 

[Martimus]

After Deneb, AMD has Bulldozer scheduled for next year. It will be the new architecture with a 45nm process. They plan to update that with a 32nm process shrink in 2010.

Also, when comparing the transistor size, you are comparing apples to oranges. The cache takes up MUCH less space than the core logic, and the IMC seems to take up even more space per transistor than the core. Some comparisons have shown that the Barcelona cache has caught up with the 65nm Intel cache with respect to transistors per square mm, but again this is an apples to oranges comparison since the AMD L3 cache is different thatn the Intel L2 cache.

I am by no means an expert on these things, but this is what I have picked up listening to people who claim to be experts at least. Maybe one of them can chime in and explain it better than me.

 

[GFORCE100]

jiffylube1024

All valid points that compliment my own above. You're filling in the gaps I left out. as I wasn't meaning to write a report on AMD, rather just voice the things that pop to my head initially.

Your text prove you think rationally, unfortunately I can't say the same for AMD. While the general concept was good, and their intentions were probably best at heart, the fact they tried to do too much too quickly brought down the roof on them.

I'm sure we could debacle if AMD would be better off with or without the payload that is ATI on their back but my point above was about cash. If AMD had not bought ATI then they could now build a new fab in NY and actually try and remedy their production capacity.

I was joking with the 90nm quad core, sure it's not feasible, however technologically speaking AMD would get better yields, and to twist some sarcasm into it all, quite frankly it's what they are in desperate need currently. They should be at 3GHz now, all singing all dancing and not producing toasters at 2.5GHz like in the early K6 days.

 

[bryanW1995]

I'm sure we could debacle

that seems to describe amd's current plight quite succinctly

 

[MGMorden]

I think people are getting too wrapped up in performance.

Once upon a time, AMD made chips that were very much slower than their Intel counterparts. These were the days of the K5's, K6's, and K6-2's. They were RESPECTABLE chips, but you were dreaming if you thought that one could compete with a Pentium II at equal MHZ. Even with the Athlon vs Pentium III comparison they were in many ways tied. AMD only got their performance crown when Intel seriously dropped the ball with the Pentium 4 architecture.

Thing is, people bought AMD chips EVEN BACK THEN, despite everyone knowing they were slower, because they were still cheaper. It still appears that in quantity, AMD is still a bit cheaper than Intel, and they'll keep a decent ammount of sales just because of that. I know that personally, I'm not into latest and greatest. I buy upgrades in big leaps. My Windows machine is running a Sempron 3400 single core right now. My homebrew mac is running an Intel C2D 1.8ghz (I bought Intel there soley because Intels just work better on homebrew macs). Both do just fine, but I MAY upgrade the windows machine to a phenom soon. The fact that the Q6600 is "teh hawtness" or anything like that right now doesn't matter. I don't care if it's 10% faster on a benchmark. The bottom line is that any existing quad-core chip will be a huge upgrade for me, and the Phenom chips are abit cheaper (and will run in my current socket AM2+ mobo).

 

[Foxery]

Originally posted by: jiffylube1024
The ATI purchase was to consolidate their chipset/motherboard market and ensure they had stable platforms available for the forseeable future. Getting ATI's graphics division was just a bonus of sorts (at least that's how I saw the purchase).

At one point they were also planning a future architecture akin to Intel's Larabee concept: some kind of combination CPU/GPU multi-core chip in 2009 or later. (Codename Fusion I think? Haven't seen it mentioned in a while.)

 

[Martimus]

Originally posted by: MGMorden
I think people are getting too wrapped up in performance.

Once upon a time, AMD made chips that were very much slower than their Intel counterparts. These were the days of the K5's, K6's, and K6-2's. They were RESPECTABLE chips, but you were dreaming if you thought that one could compete with a Pentium II at equal MHZ. Even with the Athlon vs Pentium III comparison they were in many ways tied. AMD only got their performance crown when Intel seriously dropped the ball with the Pentium 4 architecture.

Thing is, people bought AMD chips EVEN BACK THEN, despite everyone knowing they were slower, because they were still cheaper. It still appears that in quantity, AMD is still a bit cheaper than Intel, and they'll keep a decent ammount of sales just because of that. I know that personally, I'm not into latest and greatest. I buy upgrades in big leaps. My Windows machine is running a Sempron 3400 single core right now. My homebrew mac is running an Intel C2D 1.8ghz (I bought Intel there soley because Intels just work better on homebrew macs). Both do just fine, but I MAY upgrade the windows machine to a phenom soon. The fact that the Q6600 is "teh hawtness" or anything like that right now doesn't matter. I don't care if it's 10% faster on a benchmark. The bottom line is that any existing quad-core chip will be a huge upgrade for me, and the Phenom chips are abit cheaper (and will run in my current socket AM2+ mobo).

Yeah but the difference was that back then, the AMD chips were about 1/3 the price of an equivalent Intel chip. Now the price difference is neglible. I never bought Intel before because it just seemed rather stupid, considering the price difference (at least when I went to buy processors). Now, Intel makes a whole lot of sense with their pricing structure. If I were to buy a new computer other than a HTPC I would almost certainly buy an Intel processor. Although for a HTPC, I would still buy AMD because of the 780G chipset.

 

[soonerproud]

Originally posted by: Foxery

Originally posted by: jiffylube1024
The ATI purchase was to consolidate their chipset/motherboard market and ensure they had stable platforms available for the forseeable future. Getting ATI's graphics division was just a bonus of sorts (at least that's how I saw the purchase).

At one point they were also planning a future architecture akin to Intel's Larabee concept: some kind of combination CPU/GPU multi-core chip in 2009 or later. (Codename Fusion I think? Haven't seen it mentioned in a while.)

Fusion is still on the roadmap and is still slated to debut on 09 on AMD's mobile platform.

 

[soonerproud]

Originally posted by: Martimus
Yeah but the difference was that back then, the AMD chips were about 1/3 the price of an equivalent Intel chip. Now the price difference is neglible. I never bought Intel before because it just seemed rather stupid, considering the price difference (at least when I went to buy processors). Now, Intel makes a whole lot of sense with their pricing structure. If I were to buy a new computer other than a HTPC I would almost certainly buy an Intel processor. Although for a HTPC, I would still buy AMD because of the 780G chipset.

AMD's gains in market share are primarily in developing countries. In these markets, even a difference of a few dollars means the difference between eating or not.

 

[v8envy]

Originally posted by: Martimus


Yeah but the difference was that back then, the AMD chips were about 1/3 the price of an equivalent Intel chip.

Bingo. Half price or less for 10-20% performance deficit was totally FTW. That's why I ran a K2 instead of a P3.

Now? $200 (or less) for a Q6600 or $235 for a 9850. Oh, and I need a $150 enthusiast motherboard (without the much touted integrated graphics) to not have the Phenom machine literaly burst into flames vs $100 board for Intel. But wait, once both are overclocked the Intel is no longer 10% faster, it's now 30-40% faster! So, end result, 30% more cost for a 30%+ performance deficit. Did I mention the Phenom also uses about 50 more watts at load?

Pass, thanks.

Oh, and good luck installing the Phenom on your AM2 motherboard. Make sure the BIOS supports one and you have enough power.

The value proposition is simply not there for the 'high end' AMD parts. And I'm not sure it can be -- AMD has grown and matured since the days of the K2. The corporate culture, egos and investors may not embrace a strategy of low buck value player again until there's no other option remaining.

As pointed out, the 780G crowns AMD the king of the bottom of the barrel when paired with x2 processors -- for now. They have until Intel solves their platform deficiencies to come out with either compelling value or a compelling product.

 

[phexac]

I was actually at Best Buy today and saw a lot of machines they were selling contained AMD Phenom quad and tri-cores. That was an interesting surprise, and a pleasant one. The first machine I ever built was an AMD X2 3800+ rig, so I definitely want AMD to do well. Hopefully they will. Also looking for their next gen ATI video cards.

 

[bryanW1995]

amd is much more competitive in dells/hps/etc or with anybody who doesn't oc for that matter. It's simply with us crazy enthusiast overclockers that intel is demolishing them. Of course, what this tells us is that intel COULD demolish amd by a lot more than they currently are. I hope that amd gets to be somewhat competitive before nehalem comes out...how cool would it be to have a QX9950...

 

[v8envy]

Yup, it's because the average consumer doesn't really care about performance or power usage that Gateway and HP are able to put 5 Phenom powered boxes on the shelves to each Intel box. Sam's Club, Costco -- they carry nothing but Phenoms. The average consumer doesn't care about performance loss to TLB fix, so they'll happily buy that 9600 powered box for $20 less than a Q6600 and as far as they're concerned they got a great deal.

Whether they needed a quad in the first place is a whole different topic.

Used to be PC magazines would do machine shootouts, and us drooling consumers would go there to see who's got the best bang for buck. Now nobody reads dead tree format publications, and ANY PC is fast enough for household chores. AMD will be fine so long as this doesn't change.

 

[CTho9305]

Originally posted by: jiffylube1024

Originally posted by: GFORCE100
It's worth noting that AMD can't pack transistors as tightly as Intel can, even at 65nm. Intel has a 30% or so lead in this respect alone. Take a look at say the Core 2 Quad 65nm Vs the Phenom. Intel has 572m transistors, AMD has 450m but Intel's dize size is only 1mm2 larger than the Phenom with 2x the amount of cache onboard. The fact it uses a lot less power as a result is also impressive.

As for the second part, I didn't realize AMD couldn't shrink its transistors as much as Intel could on the same process. I wonder if this has anything to do with the on-die memory controller of the AMD chips.

You can't draw that conclusion the way you did. 1MB of cache is roughly the same area as a core, and is going to be >=75M transistors. 1 core is 15-30M transistors. If one chip uses smaller cores, the overall density will appear higher, even if the individual transistors are less dense. I see plenty of white space in this Barcelona die photo, which makes the area bigger without increasing transistor count at all.

To get to the right answer properly, you'd want to compare the area of 1MB of cache on each processor, and the area vs. transistor counts of the cores.

 

[Idontcare]

Originally posted by: CTho9305

Originally posted by: jiffylube1024

Originally posted by: GFORCE100
It's worth noting that AMD can't pack transistors as tightly as Intel can, even at 65nm. Intel has a 30% or so lead in this respect alone. Take a look at say the Core 2 Quad 65nm Vs the Phenom. Intel has 572m transistors, AMD has 450m but Intel's dize size is only 1mm2 larger than the Phenom with 2x the amount of cache onboard. The fact it uses a lot less power as a result is also impressive.

As for the second part, I didn't realize AMD couldn't shrink its transistors as much as Intel could on the same process. I wonder if this has anything to do with the on-die memory controller of the AMD chips.

You can't draw that conclusion the way you did. 1MB of cache is roughly the same area as a core, and is going to be >=75M transistors. 1 core is 15-30M transistors. If one chip uses smaller cores, the overall density will appear higher, even if the individual transistors are less dense. I see plenty of white space in this Barcelona die photo, which makes the area bigger without increasing transistor count at all.

To get to the right answer properly, you'd want to compare the area of 1MB of cache on each processor, and the area vs. transistor counts of the cores.

For 45nm Nehalem vs. Shanghai info:
http://chip-architect.com/news/Shanghai_Nehalem.jpg

Nehalem die = 246mm^2 (1.2% bigger than AMD)
Shanghai die = 243mm^2

Nehalem Core = 24.4mm^2 (59.4% bigger than AMD)
Shanghai Core = 15.3mm^2

Nehalem L2$ = 7.1mm^2/MB
Shanghai L2$ = 7.5mm^2/MB (5.6% bigger than Intel)

Nehalem L3$ = 5.7mm^2/MB
Shanghai L3$ = 7.5mm^2/MB (31.6% bigger than Intel)

Kinda crazy how much of a disparity exists between the core's themselves. 24mm^2 vs 15mm^2

If Shanghai even comes close to the IPC of Nehalem it will be a major coupe for the AMD designers. Intel is just throwing tons more logic transistors at pumping up those ALUs/FPUs.

Edit: corrected die sizes, thanks Extelleron for the heads-up.

 

[Extelleron]

Originally posted by: Idontcare

Originally posted by: CTho9305

Originally posted by: jiffylube1024

Originally posted by: GFORCE100
It's worth noting that AMD can't pack transistors as tightly as Intel can, even at 65nm. Intel has a 30% or so lead in this respect alone. Take a look at say the Core 2 Quad 65nm Vs the Phenom. Intel has 572m transistors, AMD has 450m but Intel's dize size is only 1mm2 larger than the Phenom with 2x the amount of cache onboard. The fact it uses a lot less power as a result is also impressive.

As for the second part, I didn't realize AMD couldn't shrink its transistors as much as Intel could on the same process. I wonder if this has anything to do with the on-die memory controller of the AMD chips.

You can't draw that conclusion the way you did. 1MB of cache is roughly the same area as a core, and is going to be >=75M transistors. 1 core is 15-30M transistors. If one chip uses smaller cores, the overall density will appear higher, even if the individual transistors are less dense. I see plenty of white space in this Barcelona die photo, which makes the area bigger without increasing transistor count at all.

To get to the right answer properly, you'd want to compare the area of 1MB of cache on each processor, and the area vs. transistor counts of the cores.

For 45nm Nehalem vs. Shanghai info:
http://chip-architect.com/news/Shanghai_Nehalem.jpg

Nehalem die = 243mm^2
Shanghai die = 246mm^2 (1.2% bigger than Intel)

Nehalem Core = 24.4mm^2 (59.4% bigger than AMD)
Shanghai Core = 15.3mm^2

Nehalem L2$ = 7.1mm^2/MB
Shanghai L2$ = 7.5mm^2/MB (5.6% bigger than Intel)

Nehalem L3$ = 5.7mm^2/MB
Shanghai L3$ = 7.5mm^2/MB (31.6% bigger than Intel)

Kinda crazy how much of a disparity exists between the core's themselves. 24mm^2 vs 15mm^2

If Shanghai even comes close to the IPC of Nehalem it will be a major coupe for the AMD designers. Intel is just throwing tons more logic transistors at pumping up those ALUs/FPUs.

You have the die sizes mixed up; Nehalem is 246mm^2, Shanghai is 243mm^2.

AMD is clearly at a disadvantage in the L3 cache; they don't have the resources to design specific SRAM for L3, so they reuse the same SRAM they use for L2. Intel designs SRAM designed for higher density for the L3, so they can stick 8MB L3 in the same area as AMD fits 6MB L3.

It seems like AMD has a lot more non-core/cache area on Shanghai compared to Nehalem.... total cache sizes are pretty similar and Nehalem has ~36mm^2 more core area, but the two are virtually the same in total die size.