AMD has launched the XBB U200 and U250 accelerator cards that support PCIe Gen3x16 interface. These cards are designed to accelerate data center workloads and are based on Virtex 16 nm UltraScale™ architecture capable of delivering 18.6 to 33.3 Tera Operations Per second (TOPs) INT8 peak. The accelerator cards support up to 64 GB of off-chip memory and feature a 100GbE network interface, which provides high-speed connectivity for data center applications. With these cards, organizations can significantly improve application performance and reduce latency for their data-intensive workloads.
The XBB U200 and U250 accelerator cards offer a range of features including support for OpenCL™ and CUDA® programming models, native FP16 and INT8 precision, and high-bandwidth memory interfaces. They are also optimized for machine learning and deep learning applications and can be used in conjunction with AMD’s ROCm software platform. Additionally, the accelerator cards provide enterprise-grade security features such as secure boot and firmware validation.
In summary, the XBB U200 and U250 accelerator cards from AMD provide high-performance computing capabilities for data center applications. With their PCIe Gen3x16 interface, support for up to 64 GB of off-chip memory, and 100GbE network interface, these cards are well-suited for data-intensive workloads and can significantly improve application performance.
High performance fast/low latency, such as speech recognition
Adaptable reconfigurable acceleration; offering changing algorithms to optimize for any workload
Deployable in the cloud or onpremise; AWS, Hwawei, and Nimbix accessible
2x QSFP28 (100GbE) network interface
PCIe Gen3x16
892K to 1341K LUTs
INT8 TOPs (peak) 18.6 to 33.3
225 W maximum total power
Passive or active cooling
Tell us what you're after