Amd infinity fabric now allows memory sharing between cpu and gpu

In AMD's Ryzen processor architecture, the Infinity Fabric (IF) bus is a core technology. Only with this technology can many CCX modules be interconnected. Previously, the IF bus was mainly used for connection between CPU cores. Now AMD has finally started using it for EPYC CPUs and Radeon graphics cards.

Third-generation AMD Infinity Fabric enables shared memory between CPU + GPU

At the OGHPC conference, AMD introduced memory consistency between the CPU and GPU via the IF bus. They can now work in conjunction with the EPYC CPU and Radeon graphics cards, and there is no problem combining it with 4 Radeon Instinct acceleration cards.

It is not surprising that AMD has done this, and even said that they have already taken some slower steps in this regard. This is necessary for high performance heterogeneous computing. NVLink 3.0 jointly developed by IBM and NVIDIA has achieved a bandwidth of 300 GB / s. The bandwidth of the SlingShot bus developed by the company also reached 200 GB / s.

Visit our guide on the best processors on the market

Intel is also developing the CXL bus. It is planned to be based on PCIe 5.0 bus technology and the bandwidth will easily exceed 128 GB / s.

The AMD IF bus has been developed for the second generation 7nm Zen2. The bit width has been expanded from 256 bits to 512 bits, and the bandwidth has been increased from 42 GB / to 92 GB / s.

In early January of this year, AMD also hired Joshua Friedrich, a former IBM Power 9 processor development engineer, as vice president. The latter joined IBM in 1999 and has more than 20 years of experience in advanced processor development, being the director of the first Summit Power 9 supercomputer.

After joining AMD, Joshua Friedrich will undoubtedly play an important role in the investigation of the tight integration of EPYC processors and Radeon Instinct acceleration cards. We will keep you informed.

Mydrivers font