Hardware AMD Reveals 2012 Roadmap

aussiebear is absolutely right, that's the reality of things and I'm pretty sure most people reading here are (mostly) aware of the situation. Regardless, this is an enthusiast site harboring that small percentage of users who DO build their own and are interested in power, so we're entitled to wish and hope

 

what about Ram DDR4 support
Chipset designing special for High-end Desktop that support 3D Softwares is best way that both intel and AMD follow it .
I request from both Company that support DDR4 in Next Chipset (ie for intel X79 and for AMD 990FX) to success agianst consumers

 

There is no such thing as DDR4 yet. JEDEC hasn't ratified the specification. There are test samples floating around, but you could say that about a lot of technology

The chipsets no longer contain the memory controller anyway, so the X79 or 990FX could already "support" DDR4 in future motherboard iterations if new CPUs are announced that use it.

aussiebear - Triple channel memory is a false economy. Unfortunately AMD is not going triple channel in 2012. Why? Efficiency and costs.

With every additional channel your efficiency goes down due to the limitation of real-world interleaving and where the data is stored. Intel gets away with it more for X58 due to the type of platform it was designed for, but it never gave 30% more performance. Mainstream this is not the case. AMD (well, ATI) has tried to compensate it by adding local memory buffers before - ever since about 2003. This might happen again.

The complexity of PCB and tracing goes up with each additional channel, which means motherboards and their development costs more. This is why Intel introduced FB-DIMMs and then LR-DIMMs for servers.

The overclocking capacity is also limited the more channels you add, but it's also a factor of how the memory controller works (frequency relationship between uncore and memory for example). (However, LGA2011 memory use will be different for reasons I'm not allowed to explain yet.)

No platform designer will use more than dual channel memory and four DIMM slots in the mainstream consumer realm. X58 was an exception because it was designed for workstations. Physical DIMMs cost a finite amount of money, and adding another costs more for OEMs. Frequency and capacity can drop in price though, and with 30nm class memory coming on the market 1866+ DDR3 is easier than ever to make.

It also means that AMD has to redevelop a memory controller to add another memory channel - that's physical silicon validation time and cost, as well as extra silicon and pin-count and redesigning of things like prefetchers and drivers. Remember AMD needs these chips on the market ASAFP. It's already 6 months behind Sandy Bridge and has had to resort to using old K10.5 cores to quicken and simplify development. It won't have the time or resources to develop another channel if it wants to move Bulldozer to mainstream and introduce VLIW4 Radeon cores in 2012.

In addition, even FM1 is based on the AM/AM2/AM3 socket size, which keeps costs lower on the manufacturing and development front. Socket design is actually extremely difficult to make sure every connection is made and the cooling is even (again look at the LGA2011 socket with TWIN clips!)

Simply getting your memory controller to scale in frequency is just easier and cheaper for all involved.

Ultimately though Llano APUs perform good enough for their intended purpose: they nail Intel = design win, but still leave the door open for improvement in future iterations (some form DDR4 maybe if it's late 2012) and the APUs don't impede too much on AMD's discrete GPU sales. Remember of all companies, AMD still needs to make money after several years in the red.