New server chips have a multi-chip layout and are intended primarily for artificial intelligence systems. Intel announced the imminent release of the Xeon Scalable family of processors, which are not designed for the consumer market, but for artificial intelligence systems, SaaS (infrastructure as a service) and high-performance computing (HPC). The chips were named Cascade Lake AP (Advanced Performance), in comparison with existing Xeon Scalable, they have a higher level of performance.
Cascade Lake AP received a multi-chip layout, similar to AMD solutions, and will use 14nm manufacturing technology. The chips are equipped with 48 cores and a 12-channel DDR4 RAM controller with support for up to two sockets. So far this is all we know about the new processors. It is not entirely clear why the manufacturer mentioned only the number of cores without the usual “cores/threads”. Does this mean that new chips do not have hyperthreading, and the company deliberately avoids the security problems that may arise with this technology in some scenarios of their use? By the way, Cascade Lake will include hardware fixes for Specter and Meltdown vulnerabilities.
Overall, the company claims a performance improvement of 20% compared with existing Xeon Scalable processors and 240% compared with AMD Epic. New crystals contain a number of instructions AVX512, designed to operate neural networks. Of course, Cascade Lake AP should get a new connector, which will probably be LGA 5903.
In addition to the Cascade Lake AP server, Intel also announced that its Xeon E-2100 CPU (6 cores / 12 threads) is already commercially available. This processor is designed for small workstations and is essentially identical to the main Core models with the only difference that it supports ECC memory and uses the server version of the chipset.
For more stuff visit our site techverses.com and discover what you want.