Intel Corporation released the third-generation Intel Scalable processor on Tuesday, that is, the “Ice Lake” data center microprocessor, hoping that its internal production business will help solve the problem of chip shortages.
Intel claims that the new processor is designed for cloud computing vendors and other companies that run large-scale data centers, and has shipped about 200,000 test units.
As reported, the Ice Lake-SP platform has up to 40 “Sunny Cove” cores per processor, with built-in acceleration functions and new instructions that can significantly improve performance for AI, HPC, network, and cloud workloads.
Intel said that Ice Lake has given a 20% increase in IPC (28 cores, ISO frequency, ISO compiler) and improved single-core performance in a series of works. Additionally, Intel has expanded the number of cores from 28 to 40 in the previous generation of Cascade Lake and also provides 8 channels of DDR4-3200 memory, and each slot supports up to 64 Lane PCIe Gen4.
Intel’s proud AVX-512 instructions are also retained. Officially, the top 8380 Platinum Xeon with AVX-512 enabled performs 62% higher on Linpack than AVX2. Intel stated that with these enhancements, as well as AVX-512 for computing acceleration and DL Boost for AI acceleration, Ice Lake’s data center workload performance has increased by an average of 46%, while the average HPC performance has increased by 53 times.
In an early internal benchmark test, Intel also demonstrated that Ice Lake outperforms the recently released AMD third-generation Epyc processor on key HPC, AI, and cloud applications.
The new Ice Lake processor combines Intel Optane memory with traditional DRAM, and each socket supports up to 6 TB of system memory.
Optane PMem 200 is part of Intel’s data center product portfolio for the new third-generation Xeon platform, including Optane P5800X SSD, SSD D5-P5316 NAND, Intel Ethernet 800 series network adapters company’s Agilex FPGA.
In addition to moving to PCIe Gen4 that gives a 2x bandwidth increase compared to Gen3, Ice Lake’s slot-to-slot interconnect rate has increased by nearly 7.7% to increase the bandwidth between processors.
(Via)
