Этот веб-сайт использует cookie-файлы. Продолжая просмотр, вы соглашаетесь на их использование. Ознакомьтесь с нашей политикой соблюдения конфиденциальности.>


    FX-Series FPGA Accelerator Cards

    Huawei’s FX-Series FPGA accelerator cards unlock the ultimate performance with PCIe 3.0 x16 interconnect lanes and two 100 GE network ports. The cards use the Xilinx UltraScale+ 16-nm VU9P and VU5P chips and support up to 32 GB of external memory. FX-Series cards are optimized for workloads such as Artificial Intelligence (AI), gene sequencing, video encoding, image processing, data compression, and network processing. They deliver compelling compute power and large-bandwidth data connections for various types of hardware-accelerated services and help turbocharge data centers with optimal energy efficiency.

    Application Scenarios

    • Video Processing

      Video applications, such as automatic image classification and recognition, image search, transcoding, real-time rendering, livecasting, plus augmented and virtual reality, require high real-time computing performance that can be provided by FPGA servers. Offering cost-effective video solutions and ideal for video scenarios, they reduce H.265 encoding deployment costs by 40%, improve H.265 encoding compression rate by 30% to 50% over H.264, and reduce HD livecast expenditure by 30% to 50% for Content Delivery Networks.

    • Image Transcoding

      Expanding image content continues with the rapid development of technologies such as big data, the IoT, mobile interconnection, and cloud computing and consumes a large amount of server resources. The FPGA server features advantages such as efficient parallel computing, high throughput, and low latency, and is ideal for solving issues such as low CPU processing concurrency, slow image processing, and heavy computing resource consumption. TCO is down 1/7 of the original, performance is  improved by 15x, and latency is slashed to 1/3.
    • Genome Research

      Precision medicine can be achieved through gene sequencing and analysis as well as rapid analysis of mass biological and medical data. Many fields, such as pharmaceutical development and molecular breeding, also require mass data processing and hardware acceleration to resolve performance bottlenecks for biological computation. FPGA servers meet such requirements with their outstanding programmable hardware computing performance that is up by 6x. CAPEX is down by 40%, deployment space is reduced by 80%, and power consumption is down by 70%.

    • Financial Analytics

      The financial industry has stringent requirements on computing capabilities, and demands rapid response achieved through ultra-low latency and high throughput. Service scenarios include financial computing based on the pricing tree model, high-frequency financial transactions, fund and securities transaction algorithms, financial risk analysis and decision-making, plus transaction security assurance. The FPGA server provides an optimal hardware acceleration solution for these scenarios by using programmable hardware acceleration technology. In certain scenarios, the FPGA server delivers 100x performance improvement above that achieved by software. FPGA servers also boosted computing performance and analysis accuracy, as well as achieved ultra-low latency through customized hardware circuits.

    • Deep Learning

      Multi-layer neural networks in machine learning require a large volume of computing resources. The training process involves mass data management, while the inference process requires ultra-low latency. In addition, machine learning algorithms are continuously improving. FPGA accelerator cards meet these requirements due to their parallel computing, programmable hardware, low power, and low latency. FPGA accelerator cards dynamically provide the optimal hardware circuit design for various machine learning algorithms, meeting the strict requirements on mass computing and ultra-low latency. Therefore, FPGA accelerator cards serve as a compelling solution to meet the hardware requirements of machine learning. With FPGA cards throughput is improved by 1.5x and latency is reduced by half.


    Model FX600 FX300
    Form Factor Full-height, 3/4-length Half-height, half-length
    FPGA Xilinx UltraScale+ 16-nm VU9P
    2,586K LE, 6,800 DSP
    345.9 Mbit/s
    Xilinx UltraScale+ 16-nm VU5P
    1,314K LE, 3,474 DSP
    168.2 Mbit/s
    PCIe PCIe 3.0 x16/PCIe 4.0 x8 PCIe 3.0 x16
    Memory 4 DDR4 DIMMs, 16 GB/32 GB 2,133 MHz 64-bit SDRAM ECC 2 DDR4 DIMMs, 16 GB 2,133 MHz 64-bit SDRAM ECC
    Network 2 QSFP28 100 GE ports, compatible with 40 GE 2 QSFP28 100 GE ports
    Power Supply Passive cooling
    Up to 200W
    Passive cooling
    Up to 75W
    Indicators/Buttons Status indicator
    Optical port status indicator
    Golden self-healing button
    Status indicator
    Optical port status indicator
    Golden self-healing button
    Configuration FPGA configuration via JTAG
    1 Gbit/s QSPI flash memory
    FPGA configuration via JTAG
    1 Gbit/s QSPI flash memory
    Temperature 0°C to 45°C (32°F to 113°F) 0°C to 45°C (32°F to 113°F)

    Для партнёров

    Дополнительные маркетинговые материалы для партнёров можно найти здесь.

    Перейдите по ссылке в раздел для партнёров, чтобы проверить статус запроса, управлять заказами, связаться со службой поддержки или ознакомиться с информацией для партнёров Huawei.