Taipei, Thursday, Oct 29, 2020, 15:26

Technology Front

COM Express Module Supports PCIe Gen4 and AI Optimization

Published: Sep 23,2020

717 Read

ADLINK Technology launches its next generation COM Express Type 6 module, the cExpress-TL, based on the 11th Gen Intel Core i7/i5/i3 and Celeron processor (formerly Tiger Lake UP3). Featuring a brand new high performance graphics core called Intel Iris Xe and supporting AVX-512 VNNI (Vector Neural Network Instructions), the cExpress-TL provides AI inferencing performance nearly 3X that previously available from non-VNNI platforms.

More on This

AUO to Acquire Shares of ADLINK Through Tender Offer, Forging AIoT Strategic Partnership

HSINCHU, Taiwan – AU Optronics Corp. (AUO) announced that as part of the value transformation plan, AUO’s Board of D...

ADLINK’s MVP-5000 Fanless Embedded Computers with 6th Gen Intel Core CPU

ADLINK Technology has announced the release of the MVP-5000 series, the newest member of ADLINK’s value family of fanless embedded computing platforms...

11th Gen Intel Core processors have a configurable power range of 15 to 28 watts, which makes them ideally suited for fanless space-constrained embedded applications. Typical industries in need of such high-performance/low-power characteristics are transport, medical, industrial automation and control, edge controllers, robotics and multi-camera-based AI.

The cExpress-TL is the first COM Express module to support PCI Express Gen 4 at 16 GT/s, effectively doubling the bandwidth compared to prior generations COM Express modules. Two other important improvements are: support for 2.5 GbE Ethernet, that leverages existing CAT5e cable infrastructure being used today for 1GbE Ethernet, and support for four USB 3.2 Gen 2 ports with transfer rates of up to 10Gb/s. Both high speed interfaces are highly suited to serve as camera input ports in AI related applications.

The integrated Intel Iris Xe Graphics core provides 96 execution units (EUs) and can be configured to support two 8K independent displays or four 4K independent display (HDMI/DP/eDP). Legacy display interfaces such as LVDS and analog VGA can be supported as build options.

AVX-512 Vector Neural Net Instruction (VNNI) is an x86 extension designed to accelerate convolutional neural network-based algorithms. VNNI significantly improves the number of elements processed per cycle and, depending on the type of workload, can boost inferencing performance by up to 300%. The Intel OpenVINO toolkit makes sure that inferencing workloads are run across all CPU cores and Intel Iris Xe graphics execution units.

For mission critical or rugged applications, the cExpress-TL supports in-band ECC (IBECC) that provides ECC level error correction on standard DDR4 SODIMM memories. The module is optionally available in an Extreme Rugged version supporting ambient -40°C to +85°C operation and resistance against high vibration and shock.

comments powered by Disqus