As follows from the information that Jim Keller, Senior Vice President of Technology and System Architecture of Intel, told the world, his company is currently working on the creation of a fundamentally new microarchitecture, which should become "significantly larger and closer to a linear dependence of performance on the number of transistors", than the contemporary design of Sunny Cove. Apparently, this should be interpreted in such a way that in a few years we will get a processor that will be much more complex and much more productive than the CPUs that the microprocessor giant currently offers.
The fresh Sunny Cove microarchitecture, which Intel uses in the new Ice Lake generation processors, was a serious breakthrough, because after a rather long break, it noticeably raised the IPC (number of instructions executed per clock). But processor guru Jim Keller, who currently works at Intel, says that this is far from the final point. Now he is working on the next generation of microarchitecture, which will be able to fully exploit the multiple increase in the transistor budget, which is expected over the next few years.
According to Intel estimates, the advantage of Sunny Cove cores in terms of specific performance compared to Coffee Lake cores reaches 15-18% (at the same clock frequency). However, the transistor budget of Sunny Cove exceeds the budget of its predecessor by a more serious amount - in the region of 38%. According to Keller, the Sunny Cove microarchitecture core consists of about 300 million 10nm transistors, while the Coffee Lake core contains about 217 million 14nm transistors. It turns out that the increase in performance in Sunny Cove does not reach a linear dependence on the size of the transistor budget: the progress in performance turned out to be about half the increase in the complexity of the semiconductor chip. According to Keller, this should not be the case.
Speaking at a lecture at the University of Berkeley, a leading specialist from Intel raised the issue of the evolution of microarchitectures of Intel processors and in the story did not stop at Sunny Cove, but mentioned a possible successor to this microarchitecture: βSunny Cove works with 800 instructions simultaneously, executing from 3 to 6 x86- instructions per clockβ¦ It has massive data predictors, massive branch predictors. But we are working on a microarchitecture generation that is much larger, and the law of performance growth is closer to linear. It's a really big change in thinking."
The position of the star engineer is that the processor industry is still far from the moment when some limit will be reached. According to Keller, Intel has ambitious plans for the future, which include a 50-fold increase in the number of transistors in processors and major improvements in almost every functional node. And there is nothing impossible in this. As Keller explains: βComputers are built by a huge number of people, but in reality it is a large number of small teams. You can improve branch prediction, instruction set, architecture, optimization, use better design tools and better libraries. The number of different application points where there is room for innovation is actually very, very large.β
Intel's current public plans include two iterations of microarchitecture improvements after Sunny Cove. The next design of Willow Cove, as promised, should have changes in the cache memory subsystem and a transition to a new semiconductor technology (probably 7 nm). Then, Golden Cove will increase single-threaded performance and focus on working with artificial intelligence tasks, along with optimizations necessary for better performance when working in networks of the fifth generation. Perhaps, in his report, Jim Keller had Golden Cove in mind, although nothing was specifically said about this.
Source: 3dnews.ru