Each 2S server can have either 3 Infinity fabric lanes and 160 PCIe Lanes with another 12 more PCIe links for the platform, or 4 IFIS, 128 PCIe, and 12 PCIe for the platform. The platform can be configured with 3 or for Infinity fabric lanes, enabling scalable PCIe lane counts in a 2S configuration. In essence, these SerDes can have multiple personalities, making the options of what is connected very configurable. The larger package with more layers is a big deal for companies like UnimicronĪMD’s scalability of IO is very noteworthy. The dimensions grew from 58mm x 75mm to 72mm x 75mm. The mounting mechanism is more robust, and the pins have a much tighter 0.94 x 0.81mm pitch. When the slides have a ton of words, that means they are great because there is very little explanation needed you can read them!Īnother noteworthy point is that the socket is completely reworked. The IO die is now beefed up to communicate with 12 chiplets through a much larger package that has more layers. It is built on an N6 process node rather than N5 like the CPU chiplet. The IO Die is arguably a much larger and more important change with the 4th generation Epyc launch. It has even more families and variants, but it’s not fair to AMD to share some of those details today.

Lastly, AMD’s next generation is called Turin, which we expect to launch in the first half of 2024. Siena can be thought of as 1/2 of Genoa or Bergamo from memory to core counts. We would also say it’s for some enterprise deployments due to lower power and Capex demands. Genoa X will be Genoa but with 3D V-Cache, and potentially multiple variants.

Genoa will also have another variant called Genoa X is for “technical.” This is an odd definition, but it is for computational fluid dynamics, EDA, and other workloads requiring more cache. The cores for Zen 4C are laid out to maximize density at the cost of frequency. The IO die and platform are shared with Genoa, so many aspects are similar, but replacing the Zen 4 core for the Zen 4C core, which has the same core architecture and L2 cache with half the 元 cache per core. Genoa, the topic for today, is for general purpose and mainstream.īergamo is for cloud-native workloads. In the 4th generation, there are 4 variants. While CPUs are for general-purpose workloads, there is increasing customization for various end markets. AMD is coming with more and more variants of their CPUs.

Amd zen 4 genoa