Intel continues to pump more horsepower in its chips, by releasing a 24-core processor for high-performance computers.
The Xeon processor E7-8890 v4 chip is part of the Xeon E7-8800 v4 family of chips that Intel announced on Monday. It was announced just a week after Intel made headlines with a new 10-core Core chip for gaming code-named Broadwell-E.
The 24-core chip will go, in many cases, into monster performer four- to eight-socket servers. An eight-socket system could have up to 192 cores, with support for up to 24TB of memory.
In a data sheet, Intel estimated a 192-core system with 2TB of memory and two hard drives to be priced at about US$165,000. Add 24TB of memory, and the server price could skyrocket.
About 18 server makers are coming out with 51 systems running the new processors, said Pat Buddenbaum, general manager of enterprise IT solutions in Intel’s Data Center Group.
Lenovo and Dell will come out with integrated servers with storage and networking components. Some servers from other vendors — likely SGI and Fujitsu — could have up to 32 sockets, meaning systems could support up to 768 CPU cores.
Seven new chips with between four to 24 cores are being introduced in the Xeon E7-8800 v4 family. The chips will succeed processors from the Xeon E7 v3 family, which were announced last year and had up to 18 cores.
Except for Intel’s Xeon Phi supercomputer chips, the Xeon E7 chips are Intel’s fastest performing processors.
Intel is placing more emphasis on Xeon chips as it reduces its reliance on PCs. The company had a 99.2 percent market share in servers in 2015, according to IDC, but that share could fall next year if ARM servers are adopted in wider numbers and as AMD releases new x86 server chips.
The Xeon E7 chips are meant for high-uptime systems used by banks, financial organizations, and other industries.
For example, banks use existing Xeon E7 v4 servers for real-time fraud detection and transaction analysis, Buddenbaum said.
A number of servers with the new chips could be pooled to create a powerful deep-learning cluster, which could more precise analysis from oceans of data. It could be used to help autonomous cars recognize signs and images, or to run complex algorithms to make correlations in genomics.
Intel’s biggest focus for Xeon E7 v4 is in-memory processing, used for applications like databases. SAP, Oracle, and Microsoft, which sell analytics and database software, are placing more emphasis on in-memory processing because it’s faster than continuously shuffling data between the CPU, memory, and storage. The chips support DDR4 memory.
The Xeon E7-8800 v4 family plays into Intel’s other areas of focus — the Internet of Things, memory, and silicon photonics — because powerful chips could help advance those technologies. Servers could be toolboxes where data collected from IoT clients could be analyzed, Buddenbaum said.
Intel is putting emphasis on FPGAs (field-programmable gate arrays), but Buddenbaum said E7 v4 chips aren’t designed for use with them. FPGAs are a better match for the company’s Xeon E5 v4 servers, which have between one to four sockets.
The new chips are based on the Broadwell architecture and are socket compatible with the Xeon E7 v3 chips, meaning they can be plugged into existing servers using those sockets. The chip has similar error correction features to the predecessor chips.
The 24-core Xeon E7-8890 v4 has a frequency of 2.2GHz, 60MB of cache, and draws 165 watts of power. The other chips in the E7 8800 line have clock speeds from 2.1GHz to 3.2GHz, cache between 45MB to 60MB, and draw between 140 watts and 165 watts of power.
Intel also announced Xeon E7-4800 v4 series chips, which have between eight and 16 cores, draw 115 watts of power, and have 20MB to 40MB of cache. The chips have slower internal bandwidth but are designed for lower priced high-performance computers.