Rumor: AMD "Piledriver" FX CPU production to begin Q3 2012

[AtenRa]

But, people that buy $150-200 CPUs do use the IGP. Why should they make separate chips, which will inevitably eat into their profit margins more than not offering a core or two more per chip will?

Intel have created a new die for the Quad core Socket 2011 Core i7 3820, do you actually believe they couldnt manufacture and make money from a 6-core ~160-200mm2 IvyBridge die selling it at the same price as Core i7 3770K ??

 

[KompuKare]

Wait a second there: I'm fairly sure that if Intel used the IGP space to add another 2 cores they would not sell it at the same price as a 2600K/3770K unless it ran at a very slow clock speed and was locked. This is Intel who totally segment their market.

Remembering which Intel CPUs support visualization is beyond my memory for instance. If people want workstation class performance Intel wants them to buy S2011 and they don't want to cannibalize that market or the Xeon market.

So and so million transistors and a certain die space has some bearing on prices but it's hardly the only factor.

 

[AtenRa]

Intel could have one more Desktop die. Ill explain,

They could keep the low end Dual Cores for up to $100-130. Keep the same Quad cores they currently using from $130 to $199 market. This way you have covered more than the 60-70 percent of the Desktop market and all of your CPUs have iGPUs.
Intruduce a new 6-core 160-200mm2 die for the $200 to $300 Desktop/workstation market (perhaps even high end Laptop paired with discrete GPUs).
Finaly they would keep the 8-Core behemoth Socket 2011 for CPUs starting at $499 and above. With the 22nm process they will be able to have a high frequency 130W 8-core CPU for the high end Desktop/Workstation market.

 

[Edrick]

Intruduce a new 6-core 160-200mm2 die for the $200 to $300 Desktop/workstation market (perhaps even high end Laptop paired with discrete GPUs).

Eventually they will. When the market demands it and they can profit the most from it. Not before.

 

[pelov]

I believe you should read again Anand's SandyBridge E review. The 3960X at 3.3GHz humiliates Sandy 2600K that is clocked at 3.4GHz even in lower resolution gaming

I'm the one who doesn't understand? You're aware the 3960X has 15MB L3 cache and you're surprised it beat it in gaming? L3 cache helps a significantly in apps like gaming that are very L2/L3 heavy. Furthermore, adding 2 more cores and ignoring the cache means they have to redesign the entire die. You can't just slap a core with X amount of cache as you see fit. It doesn't work that way.

Again we are talking about desktops here but you have to bring Laptops in to the conversation. You know im right but you trying to divert the conversation to a different domain.
Users that buy $200-300 Desktop CPUs dont need the iGPU. It is a dead die space for them. All im saying is that it is better to have two more cores sitting idle in that space and use them when you would need them than have the iGPU that they will never use it.

Laptops are important. The 2600K/3770K is actually a laptop chip with an unlocked multiplier. We have't had a true blue desktop CPU since the Q6600 (and even that's debatable). Intel uses their mobile chips for their desktop platforms and they use their server chips for their workstation platforms. Why? Because those 2 segments make them far more money!!! This not only explains why it has an IGP but also the reduction in power consumption, the stalling in thread/core count and the smaller dies. None of those make sense for the desktop enthusiast. They don't care. Get with it, buddy. Desktops are an afterthought in CPU design and it's been this way for years. Intel aren't the only ones, either. I'd suggest you take a look at Johan's overview of BD's pitfalls

Desktop Performance Was Not the Priority

No matter how rough the current implementation of Bulldozer is, if you look a bit deeper, this is not the architecture that is made for high-IPC, branch intensive, lightly-threaded applications. Higher clock speeds and Turbo Core should have made Zambezi a decent chip for enthusiasts. The CPU was supposed to offer 20 to 30% higher clock speeds at roughly the same power consumption, but in the end it could only offer a 10% boost at slightly higher power consumption.

http://www.anandtech.com/show/5057/the-bulldozer-aftermath-delving-even-deeper/1

Intel nor AMD will ever make a chip for the enthusiast ever again. Let's make that point clear. We don't bring them enough money compared to mobile/server so there's no point in creating an entirely new chip specifically for the desktop enthusiast. That will never happen. Intel sells far more locked multiplier chips than they do unlocked and OEMs make far more cash with on-die graphics then giving their money over to nVidia or AMD. Your hypothetical 3770K with 2 more cores and 4 more threads would cost millions and millions of dollars (what, close to 10mil alone for the new mask?) and it would make them exactly how much cash? Why? For who?

I know you don't like it but tough luck. If Intel or AMD thought we'd make them enough money for a true blue desktop enthusiast chip to appear then they'd have made one. For a number of years we haven't seen a chip like it. In fact, Before Bulldozer AMD has offered chips more closely to what you're describing and where has it gotten them? Barely floating above water trying to appease the enthusiast while still losing out on those same benchmarks to Intel laptop/server-based alternatives.

 

[IntelUser2000]

Intel have created a new die for the Quad core Socket 2011 Core i7 3820, do you actually believe they couldnt manufacture and make money from a 6-core ~160-200mm2 IvyBridge die selling it at the same price as Core i7 3770K ??

In the case of the 3820, its still a derivative of the top die, which is a 8 core Sandy Bridge EP. The "modular" design means you take the biggest part, and cut as needed to get the lesser core variants. That's also why the largest SKU has the largest graphics.

But for mainstream, the 3770K is the top die. That means making a 6 core chip would have to be entirely new.

 

[pelov]

In the case of the 3820, its still a derivative of the top die, which is a 8 core Sandy Bridge EP. The "modular" design means you take the biggest part, and cut as needed to get the lesser core variants. That's also why the largest SKU has the largest graphics.

But for mainstream, the 3770K is the top die. That means making a 6 core chip would have to be entirely new.

I don't think he understands this. To make this hypothetical chip Intel would have to create a third platform... Yes! A third one! Not to mention redesign the entire die layout and all of the fab work that goes along with making an entirely new chip.

For those that don't know, 2011 workstation platform chips are essentially the server chips and the 1155 desktop chips are essentially laptop/all-in-one/non-enthusiast chips. Both platforms and chips tweaked very slightly to fit their respective platforms.

A graphics-less 3770K + 2 cores/4threads would be a chip specifically for the desktop enthusiast and nobody else and wouldn't fit in either platform. Good luck with that.

 

[Ferzerp]

Xeon most certainly doesn't need an igpu.

It's bad enough having to buy one on each system, try to justify sticking me with 2-8 per system depending upon config.

 

[AtenRa]

Both SB and SB-E are modular designs (IB as well). You can add or remove cores, L3 cache etc.

Intel could remove the iGPU and add two more Cores in to the IvyBridge socket 1155 CPUs without two much effort.

Same for a 6-core derivative die taken from the 8-core SB-E. Just remove two core blocks (core + L3) and you have a 6 core die.

 

[pelov]

Intel could remove the iGPU and add two more Cores in to the IvyBridge socket 1155 CPUs without two much effort.

And what about the chipset and socket compatibility? Neither of those would work.

Same for a 6-core derivative die taken from the 8-core SB-E. Just remove two core blocks (core + L3) and you have a 6 core die.

They have this already. It's called the 3930K available for ~$600.

 

[Edrick]

Intel could remove the iGPU and add two more Cores in to the IvyBridge socket 1155 CPUs without two much effort. Same for a 6-core derivative die taken from the 8-core SB-E. Just remove two core blocks (core + L3) and you have a 6 core die.

Adding cores to an existing design is much more complex than disabling cores that are already on the die. Also, as pelov stated, they already have a 6 core version of the SB-E. so what is your point exactly?

 

[AtenRa]

And what about the chipset and socket compatibility? Neither of those would work.

You do know that 2550K is socket 1155 without the iGPU ?? it works perfectly. Same with AMDs Athlon II X4 641 and 651 both FM1 CPUs with no iGPUs. I dont see what is the problem having 2 more cores and no iGPU with the socket compatibility ??

They have this already. It's called the 3930K available for ~$600.

The Core i7 3930K is an 8-core 435mm2 die with two CPU cores disabled costing $594. Im talking about a smaller 6-core die.


http://www.anandtech.com/show/5091/...dy-bridge-e-review-keeping-the-high-end-alive.

 

[wlee15]

In the case of the 3820, its still a derivative of the top die, which is a 8 core Sandy Bridge EP. The "modular" design means you take the biggest part, and cut as needed to get the lesser core variants. That's also why the largest SKU has the largest graphics.

But for mainstream, the 3770K is the top die. That means making a 6 core chip would have to be entirely new.

The 3820 is an entirely distinct die from the rest of the 8-core Sandy Bridge EP line.

 

[Cerb]

Intel have created a new die for the Quad core Socket 2011 Core i7 3820, do you actually believe they couldnt manufacture and make money from a 6-core ~160-200mm2 IvyBridge die selling it at the same price as Core i7 3770K ??

I believe that if they made it, they would not change current i7 prices at all, but instead make the 6-core even more expensive. However, it would also be even more expensive to them, as a 6-core would have a very small market, that simply isn't worth it. It's not like it would b free to design and roll out a new CPU.

Intel could remove the iGPU and add two more Cores in to the IvyBridge socket 1155 CPUs without two much effort.

They can reduce time to market and per-chip design cost compared to previous design paradigms, but it's not like they can plug some numbers into a computer and get a new good chip design spat out in a flash. It's more like a car platform than it is like Lego blocks.

 

[IntelUser2000]

Both SB and SB-E are modular designs (IB as well). You can add or remove cores, L3 cache etc.

Intel could remove the iGPU and add two more Cores in to the IvyBridge socket 1155 CPUs without two much effort.

Same for a 6-core derivative die taken from the 8-core SB-E. Just remove two core blocks (core + L3) and you have a 6 core die.

I don't think its a simple matter of adding cores. The modularity they talk about is clearly based on taking the biggest base chip and cutting down as necessary. Notice its not as flexible as it sounds like. The 4 core Sandy Bridge only have 3 variants: 4 core + GT2/8MB L3, 2 core + GT2/4MB L3, 2 core + GT1/3MB L3. You could think of few more, but they chose not to and disabled it. Ivy Bridge adds one more, a 4 core + GT1.

I think Sandy Bridge E/EP counts as a entirely different layout. Never in their history have they made a die specifically for enthusiasts, it was always a derivative/cut down chip from the server/workstation side. There has to be a very good reason for not having more derivatives. It doesn't make having a 6 core one impossible, but highly unlikely when they have not done it with higher volume 4 core parts.

The chips that classify as a different layout and not a derivative(rather than disabled as in case of 2550K), takes quite a big longer to arrive. It's probably not a coincidence that Lynnfield took longer than Bloomfield, MP chips compared to others, Sandy Bridge E/EP compared to mainstream one.