AMD pushes CXL, chiplets with latest data centre processor
AMD has launched its latest data centre processor with up to 96 cores and support for the CXL1.1 memory interface.
The 4.4GHz 9004 chip, codenamed Genoa, uses the latest Zen4 x86 cores with 192 threads and 384MB of L3 cache using high bandwidth memory on a silicon interposer. It includes support for 12 channel DDR5-4800 and 160 lanes of PCIe 5.0 as well as CXC 1.1+ in 64 IO lanes. There are also up to 4 links of Gen3 AMD’s proprietary Infinity Fabric with speeds up to 32Gbit/s.
The chip also supports software-defined memory with the CXL specification from MemVerge for the first time for application-aware memory tools. This can allow applications to use DDR and CXL memory on new servers.
AMD’s CXL1.1+ adds level 3 memory support from the CXL2.0 specification.
- €16m project to secure the chiplet supply chain in Europe
- Leti reports 96-core chiplet computer
- Software defined CXL memory fabric drives IP maker into chips
The Memory Machine Cloud Edition software from MemVerge adds the ability to provide transparent access to a pool of DDR and CXL memory, dynamically placing the most used data in the fastest tier and guaranteeing quality of service.
“4th Gen AMD EPYC processors and MemVerge software form a solid CXL platform capable of providing memory-intensive applications with transparent access to tiered memory fabrics and advanced memory services,” said Charles Fan, CEO and co-founder of MemVerge.
Gigabyte has launched a servers and boards with the new SPS socket specification for the chip. The M-Series Motherboards have single and dual socket options supporting Gen4 NVMe drives and PCIe 4.0 and/or 5.0 lanes while the R-series 1U and 2U rackmount servers are aimed at general-purpose systems. The E-Series edge servers are 1U and 2U short depth cards with options for IO on front or rear, moving data collection away from data centres and improving latency.
- RISC-V security IP for chiplet die-to-die communication
- Chip makers back CXL 3.0 for data centre memory
The G-Series 2U or 4U systems support GPUs for HPC, AI, and high-performance parallel processing, while the H-Series puts four hot-swappable nodes in a 2U chassis for HPC, HCI, and edge computing.
AMD has also added additional security layers with its Infinity Guardvi which has physical and virtual layers of protection with double the number of encryption keys compared to previous generations.
Supermicro and Dell have also announced boards and servers. HPE also announced support for 4thGen AMD EPYC processors across its supercomputing portfolio with the new HPE Cray EX2500 and HPE Cray XD2000 supercomputers. Lenovo introduced 21 new ThinkSystem servers and ThinkAgile hyperconverged (HCI) solutions, powered by 4thGen AMD EPYC processors, including ThinkAgile VX and ThinkAgile HX to enable fast hybrid multi-cloud deployment and simplify infrastructure management.
www.amd.com; www.gigabyte.com; www.memverge.com.
Other articles on eeNews Europe
- Cryogenic CMOS design for quantum chips on 22FDX
- SES signs deal to launch European quantum crypto satellite
- IBM sees 2023 as quantum inflexion point
- Codasip in Bristol IoT security buy for RISC-V
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
