The chip giant has renewed its CPUS Intel Xeon 6 catalog with new models specialized in Vran networks (vitual access networks) and computing on the edge, equipped with high -performance nuclei. This new family is divided into two large series, 6700 and 6500. The first is a step above the second, which positions at a more intermediate level.
The Intel Xeon 6 maintain the same bet and structure that we have seen in the Xeon 6 with high -performance nuclei, which means that we have the same design based on blocks or tiles, which allows grouping Up to 128 high -performance nuclei on a single CPU. This leaves us a high density of nuclei, and enables a high level of scalability in multisocket configurations.
In this case, the big difference we find in series 6700 and 6500 is that We have a maximum of 86 cores per CPUalthough with a high degree of scalability, thanks to the configuration support with up to 8 sockets. These new processors will be available with a TDP between 150 and 350 watts, support DDR5 MRDIMM memory in eight channels, and offer up to 88 PCIE GEN5 lines and up to 64 CXL 2.0 lines.
Intel Xeon 6: Specialization in its purest form
These new processors use Redwood Cove architecture at the CPU level, and They are manufactured in the Intel 3 node (CPU block), A state -of -the -art manufacturing process that improves performance and efficiency values, two keys that are fundamental when we talk about things as important as the cost of ownership on servers and data centers.
Intel is also aware of the importance of specialization and security, and that is why these new Xeon processors They maintain all the keys we saw in the Xeon 6 6900 series In this sense, which means that the Xeon 6700 and 6500 have:
- Intel AMX matric accelerators, which improve inference performance up to 10 times.
- AVX 512 instructions.
- Intel Software Guard Extensions e Intel Trusted Domain Extensions.
- At the encryption level they have a Vector AES, SHA2-256 and VPMADD52 extensions.
- También destacan los aceleradores Intel QuickAssist Technology, Intel Dynamic Load Balancer, Intel Data Streaming Accelerator e Intel «in-memory» Analytics Accelerator.
The MRDIMM (Multiplexed Rank Dimm) memory is capable of working at a speed of up to 8.800 MT/s, Multiply by two the transfer capacity of the DDR5 memory and is ideal for work scenarios that have a deep dependence on memory.
A server configured with an Xeon 6700 can count on the a whopping 136 PCIE lines, It will have an improved performance in entry and output operations, will offer a greater bandwidth with its MRDIMM memory, and will also have all the advantages offered by its integrated accelerators. At the cost level you will also keep An interesting balance.
According to Intel, your new XEON 6 processors can offer up to 58% more performance per watt with 40% load that an XEON 8592+. Less consumption improves efficiency and reduces the carbon footprint, but also the high density of nuclei and the high performance offered by the Intel Xeon 6 will also help us reduce the amount of necessary servers and the occupied space to carry out even workloads Very demanding.
Xeon 6 Soc: Prepared for Vran and Edge
This soc represents another important step forward by Intel in favor of specialization, since it maintains all the keys to an Xeon 6 and combines them with specific accelerators to improve their performance in very specific work loads, which allows it to offer a High performance in multipurpose tasks and also in very specific scenarios.
The XEON 6 SOCs are configured with up to 72 P nuclei, integrate the VRAN accelerators, half transcode, Ethernet, and maintain compatibility with AVX 512 and the acceleration of AMX matrices. According to Intel, these socs are able to offer a Performance improvement per watt of up to 70% compared to the fourth generation XEON. In the performance improvement is up to 430%, thanks to AMX matrices accelerators.
The multimedia accelerator also offers interesting value, because it can improve the performance per watt up to 14 times against previous generations. These socs also offer greater inference capacity with lower consumption, and are able to work with up to 38 cameras with vision for AI and deep learning.
The novelties do not end here, since Intel also presented new Ethernet solutions and Redn adapters among which we can highlight the Intel Ethernet E830 200GbEwhich allows flexible port configurations and precision time measurement (PTM), and the Intel Ethernet E610, with 10GBASE-T connectivity optimized for control plane operations.
