Heterogeneous Computing | 異質運算

  • What is it?
    Traditionally, heterogeneous computing refers to a system that uses more than one type of computing cores, such as CPU, GPU, DPU, VPU, FPGA, or ASIC. By assigning different workloads to specialized processors suited for diverse purposes, performance and energy efficiency can be vastly improved. The different elements are interconnected via high-throughput, low-latency channels so they can operate as a single unit.

    In recent years, the definition of heterogeneous computing has expanded to encompass processors based on different computer architectures. For example, processors based on the Arm architecture may be a better choice for some tasks, due to their higher number of cores, better power-efficiency, and compatibility with Arm-based mobile devices. The adoption of an alternative architecture may reveal smarter ways to handle existing workloads and computing tasks.

  • Why do you need it?
    If you are using coprocessors besides CPUs in your server solutions, such as GPUs or DPUs, then you are already reaping the benefits of heterogeneous computing. Coprocessors can greatly accelerate computing speed and reduce the time it takes to complete a task. Especially in the development of artificial intelligence through machine vision and deep learning, which requires the server to process a vast amount of data converted to graphical form, the use of GPUs that operate at lower frequencies but have more cores than traditional CPUs is vital.

    By the same logic, installing processors based on an entirely different architecture may be another way to realize the full potential of heterogeneous computing. In recent years, the trend has been to explore alternatives to the conventional x86 architecture, such as Arm processors. Industry experts are looking for ways to unleash the potential of these new products and develop new computing solutions. If your current configuration of processors and coprocessors is performing below expectations, it may be that heterogeneous computing based on a different architecture will have a better chance of fulfilling your needs.

  • How is GIGABYTE helpful?
    Nearly all of GIGABYTE's server solutions support the conventional sense of heterogeneous computing, where CPUs are paired with coprocessors such as GPGPUs to handle workloads related to parallel computingmachine learning, etc. Whether the main processor is the latest Intel® Xeon® Scalable processors or AMD EPYC™ processors, they can transfer data to peripheral devices such as graphics cards, storage devices, and high-speed network cards with PCIe Gen 4.0, which has a maximum bandwidth of 64GB/s and is twice as fast as PCIe Gen 3.0.

    What's more, GIGABYTE has worked for many years with Arm-based systems. GIGABYTE offers a full range of servers designed for Ampere® Altra® processors, based on the Arm architecture. These products are highly recommended for high performance computing (HPC)cloud computing, and edge computing because they can address diverse deployment scenarios with scalability and flexibility. NVIDIA® has also unveiled the NVIDIA® Arm HPC Developer Kit, which includes an Ampere® Altra® processor, two NVIDIA® A100 Tensor Core GPUs, and two NVIDIA® BlueField®-2 DPUs, all contained inside a G242 server by GIGABYTE. This shining example of heterogeneous computing will pave the way for industry leaders to find new ways of tackling computing workloads with tools suited to the task.

  • WE RECOMMEND
    G242-P31

    Product

    G242-P31
    H261-Z60

    Product

    H261-Z60
    R182-M80

    Product

    R182-M80
    Back to top