How Chip Makers Are Circumventing Moore's Law to Build
Super-Fast CPUs of Tomorrow:
"But AMD achieved its biggest speed gains recently with its ridiculous-sounding
Threadripper CPUs. These are CPUs with a core count that starts as low as 8 and goes all
the way up to 32. A core is kind of like the engine of the CPU. In modern computing,
multiple cores can function in parallel, allowing certain processes that take advantage of
multiple cores to go even faster. Having 32 cores can take something like the rendering
of a 3D file in Blender from 10 minutes down to only a minute and a half, as seen in this
benchmark run by PCWorld."
"Infinity Fabric is a new system bus architecture based on the open source Hyper
Transport. A system bus does what you think it would—bus data from one point to
another. Infinity Fabric’s neat accomplishment is that it busses that data around really
fast and allows processors built with it to overcome one of the primary hurdles of
chiplet CPU design: latency.
Chiplet design isn’t new, but it’s often been difficult to accomplish because it’s hard to
make a whole bunch of transistor on separate die talk to each other as quickly as they
can on a single piece of silicon. But with AMD’s Threadrippers, you have a number of
its typical Ryzen CPUs laid out on the Infinity Fabric and communicating nearly as
quickly as if they were on a single die.
It works really well, and the results are a super-fast processor that is so cheap to
make that AMD can sell it for a fraction of the price of something comparable from
Intel—which continues to use monolithic design in its high-core-count CPUs. In a way,
Infinity Fabric is a way to cheat Moore’s Law because it’s not a single fast CPU—it’s a
whole bunch attached via the Infinity Fabric. So it’s not AMD overcoming the
limitations of Moore’s Law, but circumventing it."
See:
https://gizmodo.com/how-chip-makers-...d-s-1831268322