August 6th, 2019 by Adam Armstrong
Xilinx Expands Its Alveo Portfolio At FMS 2019
Today at FMS 2019, Xilinx, Inc. announced that it is expanding its Alveo data center accelerator card portfolio with the launch of the Alveo U50. Xilinx states that the U50 is the first low profile adaptable accelerator with PCIe Gen 4 support. The company goes on to state that the card can supercharge a broad range of critical compute, network and storage workloads on any server or any cloud. Xilinx will be showcasing the Alveo U50 and other product demonstrations in booth 313 at Flash Memory Summit (FMS) 2019, taking place August 6-8 at the Santa Clara Convention Center in Santa Clara, Calif.
Xilinx’s Alveo U50 is powered by the company’s UltraScale+ architecture. The U50 is stated to deliver between 10-20x improvements in throughput, latency and power efficiency to meet the emerging challenges of dynamic workloads such as cloud microservices. Since the U50 card can move compute closer to the data, users can eliminate latency and data movement bottlenecks.
The Alveo U50 is in a half-height, half-length form factor and low 75-Watt power envelope. For performance the card utilized HBM2 for 100Gbps network connectivity and supports PCIe Gen 4 and CCIX interconnects. The 8GB of HBM2 delvers over 400 Gbps data transfer speeds and the QSFP ports provide up to 100 Gbps network connectivity. The card also supports NVMe-oF, disaggregated computational storage and specialized financial services applications.
Customer benefits and use cases include:
- Deep learning inference acceleration (speech translation): delivers up to 25x lower latency, 10x higher throughput and significantly improved power efficiency per node compared to GPU-only for speech translation performance
- Data analytics acceleration (database query): running the TPC-H Query benchmark, Alveo U50 delivers 4x higher throughput per hour and reduced operational costs by 3x compared to in-memory CPU
- Computational storage acceleration (compression): delivers 20x more compression/decompression throughput, faster Hadoop and big data analytics, and over 30 percent lower cost per node compared to CPU-only nodes
- Network acceleration (electronic trading): delivers 20x lower latency and sub-500ns trading time compared to CPU-only latency of 10us
- Financial modeling (grid computing): running the Monte Carlo simulation, Alveo U50 delivers 7x greater power efficiency compared to GPU-only performance for a faster time to insight, deterministic latency and reduced operational costs
The Alveo U50 is sampling now with OEM system qualifications in process. General availability is expected for fall 2019.