48-core Xeon processors offers significant benefits for Deep Learning
The search for more cores soldiers on as the new Intel® Xeon® processor (code named “Cascade Lake”) launched today. With up to 48 cores, the new Xeons look to be a significant step forward—particularly for deep learning.
The advancements include a boost to deep learning by introducing a new vector neural network instruction (VNNI)—part of the AVX-512 extension—that’s designed to accelerate deep learning algorithms. Specifically, it aims to increase performance in areas such as speech recognition, language acquisition, and object detection, as well as optimizing frameworks like Caffe, MXNET, and TensorFlow.
For enterprise users looking to max out memory capacities, the new theoretical ceiling will be 6TB for dual-socket boards across 12 memory channels, promising higher bandwidth and throughput than AMD’s latest EPYC™ line. Additionally, Intel is adding support for 3D XPoint-based memory modules (i.e., Intel® Optane™ DC Persistent Memory), which increases data resiliency for multimedia delivery and real-time data analytics.
This time around, Intel has also made significant strides with advancements in security. “Cascade Lake” will include hardware mitigations for three distinct speculative execution vulnerabilities.[1] This will help prevent potential information leakage via a side-channel attack.
[1] Spectre (Variant 2), Meltdown (Variant 3), and L1 Terminal Fault.