Intel Xeon Scalable Gen3 Ice Lake SP: Intel’s 10-nm processor for the Server is finally here
With Ice Lake SP, Intel is now also part of the Xeon for the smaller servers to a new processor architecture, and the 10-nm manufacturing process. The CPUs are faster, and they are also cheaper, with AMD’s new Zen3-Epyc CPUs to compete.
Companies
(Image: Intel)
The Notebook-processors, Intel has now also in the Xeon processors for the Server with one or two sockets on the market that are manufactured in a 10 nm process and also with Sunny Cove on a current and a faster micro-architecture based. You can go as Xeon Scalable processors of the third Generation of the Start-up and solve the Cascade Lake processors, the second-Xeon-Scalable-Generation. The run in June as a Xeon-Scalable-Gen3-processors for servers with four or eight Sockets featured Cooper Lake models, although they are still 14 nm CPUs. Support, however, remains as the only x86 processors, the BFloat16 data format (BF16).
In the case of Cooper Lake, and Cascade Lake, Intel was able to grab a maximum of 28 Cores on The Ice Lake SP, due to the 10-nm manufacturing process up to 40 at the Platinum model 8380. Of the 64 Cores of the AMD processors and Intel does not approach continues to be, but promises for the monolithic, The advantages in terms of Performance compared to the AMD Processor, with its structure of Chiplets and I/O-The.
In addition to the Platinum models with between 32 and 40 Cores Intel still Silver and Gold offers models that start with eight Cores and up to 32 Cores reach. With its Sunny Cove core Ice Lake-SP models are designed to handle up to 20 percent more instructions per clock than Cascade Lake. To achieve this, I have optimized the developers of the Frontend and the cache is increased. In addition, the CPU supports new commands for cryptographic and compression and decompression of data. In addition, some models offer significantly greater RAM enclaves with up to 512 GB, in which the data are encryption-protected. The Software Guard Extensions (SGX) bring Intel to Version 2.0. In addition, Intel promises a higher performance AI Workloads with Intel Deep Learning Boost, and AVX-512 instructions, as here, the processor can operate at a higher clock rate. AMD is also supported with the Zen 3-Epyc-processors of the Milan family still just AVX2.
The maximum clock frequency is lowered, Intel Tiger Lake SP in comparison to Cascade Lake, however, and competition for use on the maximum clock rate of the AMD. This is probably so that the 10-nm-SuperFin technique is used that Intel has introduced in Tiger Lake, but the processors belong to the Icel-Lake Generation. The highest clock frequency of 38-Kerner 8368Q reached up to 3.7 GHz, the Single-Core Turbo mode. However, it is designed for liquid cooling.
Eight memory channels and PCIe 4.0
The number of memory channels increases the Intel Ice Lake SP in up to eight and includes the Epyc processors of the Milan family. Thus, the memory increases, the throughput significantly. The maximum memory size per Socket is four TB DDR4-3200. With the use of Optane Persistent Memory, up to six TB are possible. In addition, the new Intel processor supports all of PCI-Express 4.0. In order for NVIDIA’s Ampere can be used on Intel-based servers-accelerators such as the A100 or PCIe 4.0-Based NVMe SSDs with full bandwidth. In addition, Intel has unveiled the E810 2CQDA2 fast Ethernet Adapter with PCIe 4.0 interface, which has two 100GbE Ports. Ice Lake SP is 64 PCI-Express 4.0 Lanes. Thus, Epyc, with 128 Lanes in a Single Socket configuration has the nose in front. In a Dual-CPU systems is the same, however, since AMD uses 64 channels for the connection of the two CPUs while Intel uses three UPI Links its own interface.
The prices for the new Ice Lake-SP models, Intel has reduced in comparison to the previous Cascade Lake Xeons significantly and offers so that sometimes even the Epcy processors from AMD. The model range is, again, very extensive and also includes special models, which are optimized for Networking, the Cloud-use, use for AI Workloads, or a long life.In addition, there are special Single-Base variants with lower price, and with such Speed Select Technology.
(ID:47329416)