AMD Extends Infinity Fabric Bus Technology CPU and GPU Finally Share Memory

In AMD’s Ryzen processor architecture, the Infinity Fabric (IF) bus is a core technology that allows many CCX modules to connect. Previously, if buses were primarily used for connections between CPU cores, AMD is finally starting to be used for EPYC CPUs and Radeon graphics cards.

At the OGHPC conference, AMD described the memory consistency between the CPU and the GPU connected via the IF bus, and now they can work together with the Radeon graphics card, with 4 sets of Radeon Instinct acceleration cards.

AMD Extends Infinity Fabric Bus Technology CPU and GPU Finally Share Memory

AMD did this without incident, even saying that they had been slow edgy in this regard, which is a necessary heterogeneous high-performance computing, IBM and NVIDIA jointly developed NVLink 3.0 has achieved 300GB/s bandwidth, Cray’s own SlingShot bus bandwidth has reached 200GB/s.

Intel is also developing a CXL bus, which is based on PCIe 5.0 bus technology and has easily more than 128GB/s of bandwidth.

AMD’s IF bus has grown to a second generation on the 7nm Zen2, with bit width expanded from 256bit to 512bit and bandwidth from 42GB/s to 92GB/s.

In early January, AMD also tapped Joshua Friedrich, a former IBM Power 9 processor development engineer, to the position of vice president, who joined IBM in 1999 and has more than 20 years of advanced processor development experience before becoming Power Processor Technical Director at IBM. The processor Power 9 used by the current TOP500 First Superpacing Is the Processor Power 9 that he led during his tenure.

Joshua Friedrich’s entry into AMD will undoubtedly play an important role in the close integration of EPYC processors and Radeon Instinct acceleration cards. For now, AMD does have that plan, and that’s the most important thing they did to buy ATI 14 years ago.