| Scoobes said:
This may be a stupid question, but won't we be able to design smaller chips with higher yields if we can multi-stack? For instance, could you design the equivalent of a Titan GPU in transistor count but on two different layers, theoretically halving the transistors per wafer and improving yield (per layer)? I guess each layer/wafer would need to be QA tested seperately for that to work. Or am I pretty far off base? This isn't a topic I've read much on.
Also, saw this earlier. It sounds like AMD PR with a couple of silly statements from the writer thrown in, but it seems relevant to the thread: http://www.tomshardware.com/news/amd-apu-efficient-goal-moore,27099.html Wow. AMD is ambitious.
Rather than announcing a new product, AMD has announced a new target called the 25X20. The idea behind it is APU efficiency.
So what exactly does the target entail? Well, AMD wants its APUs to be 25 times more efficient by the year 2020. The company indicated that over the last six years (an equal time period from today till 2020), its products became about ten times more efficient. The company also recognized that if such improvements take place, the company will be outpacing Moore's law by about 70 percent.
"Creating differentiated low-power products is a key element of our business strategy, with an attending relentless focus on energy efficiency," said Papermaster. "Through APU architectural enhancements and intelligent power efficient techniques, our customers can expect to see us dramatically improve the energy efficiency of our processors during the next several years. Setting a goal to improve the energy efficiency of our processors 25 times by 2020 is a measure of our commitment and confidence in our approach."
Of course, energy efficiency should not be confused with power consumption. It is very unreasonable to expect the APUs to use 25 times less power six years from now. Instead, they will become more powerful and consume less energy while doing so, improving the overall performance per watt. That is AMD's target -- improving performance per watt by a multiple of 25.
Now, we're not exactly sure about how AMD intends to outpace science, though the company does mention three core foci: heterogeneous computing and power optimization, where the GPU and CPU aboard the APU work more closely together; intelligent real-time power management; and future innovation in power efficiency. In short, not only will the company shrink the transistors to make for more efficient APUs, it'll also work on newer architectures and better power management, and maybe even more.
There has to be something that gives AMD the confidence to make an announcement as ambitious as this… We're curious what it is that AMD has up its sleeve. |
You should know that there is a limit as to how many chips you can stack!
The reason why we're able to stack memory chips so easily compared to processing units has to do with the fact that they consume less power and thus in turn makes heat dissipation orders of mangnitude more manageable for stacking DRAM instead of CPUs. The problem with die stacking high performance processors is that surface area becomes a problem since it ends up being harder to cool so clocks need to be lowered to maintain an acceptable level of heat output but I see this as a liability since it's simply more profitable to sell the chips at full potential than to waste more silicon on trying to improve on performance per watt. It might even be more expensive if you consider a scenario where chip A whose die area is 50% larger than chip B which has two stacked dies is able to deliver the same performance despite chip B costing more to manufacture due to the fact that it needs more silicon wafers and that chip A also has the advantage to going higher clocks since it has more surface area for the heat to dissipate compared to chip B. Yes the smaller chips may have better yields but that doesn't translate to any advantages for high performance markets like high end GPUs and CPUs.
About AMD ... LOL I wouldn't exactly be putting any confidence in their words since they haven't delivered for the past 8 years aside from the graphics division.
