Announced last NovemberThe fourth-generation EPYC server processor, codenamed "Genoa,", and the derivative version code-named "Bergamo" designed for cloud-native application service computing needs, and the derivative version code-named "Geona-X" designed for scientific computing needs, have now entered the market. The server processor code-named "Siena" designed for intelligent edge computing and telecommunications service needs is also ready to enter the market in the second half of this year. AMD said it will use this to fill the demand of the cloud computing market and can combine with the device side to achieve greater computing benefits.
Driving greater computing performance with the fourth-generation EPYC server processors
AMD emphasized that the fourth-generation EPYC 97X4 series server processors, code-named "Bergamo", will be equipped with up to 128 Zen 4c architecture cores, which will be able to correspond to 2.6 times the computing performance with the highest virtual CPU core density in cloud-native computing environments, as well as 2.7 times the power utilization efficiency, and can run more than three times the number of containerized applications on a single server.
In addition, by integrating the 3D V-Cache cache memory design, the fourth-generation EPYC server processor can use up to 96 Zen 4 architecture cores and more than 1GB of L3 cache memory to more than double computing efficiency and reduce the time and power consumption required for original work.
In this release, AMD also announced a partnership with AWS to enable its Amazon EC2 M7a server system, which is now available for preview, to deliver up to 50% better performance than the previous generation.
In addition, Oracle's Oracle Cloud Infrastructure will also introduce the Amazon E5 Instances server system based on the "Geona" design, and is expected to be officially launched in July this year.
Meta also announced that it will introduce a server system designed based on "Bergamo" to speed up the operation efficiency of its services including Facebook, Instagram and WhatsApp.
Microsoft will introduce a server system designed with "Geona-X" into its Azure cloud computing platform to boost the performance of its supercomputing designed for cloud architecture.
Driving the demand for generative AI computing with the Instinct MI300 series accelerators
As for the currently popular demand for automatically generated artificial intelligence computing, AMD also emphasized that through the Instinct MI300 series accelerators, it will be able to meet more computing needs.
This includes the Instinct MI3X, built with the CDNA 192 acceleration architecture and equipped with up to 3GB of HBM300 memory. This processor is capable of processing large-scale natural language models, such as the Falcon-400B model with over 40 billion parameters. AMD also announced that a total of eight Instinct MI8X processors will be used in industry-standard designs on the Infinity architecture platform to meet the needs of automated generative AI training and inference.
AMD expects to provide Instinct MI300X samples to specific companies for testing in the third quarter of this year. The Instinct MI2023A, which was previewed during CES 4 earlier this year and combines a Zen 3 architecture CPU and a CDNA 300 architecture GPU, is an APU accelerator designed for supercomputing and artificial intelligence computing needs, and has now begun providing samples to partners for testing.
In terms of software applications, AMD will continue to build a complete AI computing acceleration application model with the ROCm software ecosystem, and will also continue to closely integrate with major open source AI software ecosystems in the market, such as PyTorch.
Build a more complete cloud-to-endpoint computing environment
Through past acquisitionsXilinx's Pensando Systems technical resourcesAMD announced the launch of SmartNICs, designed for computing needs such as virtualization and security. These SmartNICs integrate with cloud solutions like VMware vSphere, making public cloud computing more reliable and meeting a wider range of computing needs. AMD's Smart Network Switches, designed with the P4 DPU and developed in partnership with Aruba, a subsidiary of HPE, will further enhance the benefits of hybrid cloud service applications for enterprises.
By expanding the EPYC server processor portfolio, driving accelerated computing with Instinct and Alveo, leveraging Alveo and Versal FPGA architectures to provide flexible computing scheduling, ensuring data transmission security through intelligent network adapters and DPU components, and improving overall computing efficiency, AMD expects to meet the needs of optimizing the implementation of modern data center-to-endpoint computing processes.
This time AMD also announced that it will launch the next code-named "Giglio" (Editor’s note:The name should come from the new DPU design of Giglio Island in Italy. It is expected to launch application products before the end of this year to meet more zero-trust computing architecture configuration needs.
AMD also announced the Pensando Software-in-Silicon processor-level software development kit (SSDK), which allows customers to quickly develop or migrate services and easily deploy them on the Pensando P4 DPU design. It can also coexist with existing Pensando platform features, including customized virtualization and security application services.
