Plug GPUBoost™ into PCIe and build outstanding GPU over RDMA systems with extreme scalability. Maintain thousands of concurrent Queue Pairs (QPs) without performance degradation.
AI/ML processing at Scale
Deploy GPUBoost™ RDMA across GPU nodes to unlock the power of state-of-the-art GPUs for AI training systems with multi-GPUs. Experience linear scaling without
bottle necks over standard RoCEv2, with customizable hardware-based in-line congestion control and custom headers. Seriously scale AI GPU servers with direct
communication across GPUs. Ready to ship today. And already preparing Ultra Ethernet versions for tomorrow.
200Gbps, <2us latency
Up to 4K maintained active QP
Custom congestion control and metadata
Compatible with NVIDIA/AMD GPUs
Internal Half Height Half Length PCIe Card
Unleash the potential of your AI servers
GPUBoost™ is a GPU over RDMA accelerator card or DPU (Data Processing Unit). It accelerates data movement by allowing peripheral PCIe devices to directly access your GPUs.
Experience enhanced data throughput and reduced latency as GPU over RDMA facilitates direct and efficient data transfer pathways.
Access 200Gbps line-rate bandwidth and <2us end-to-end latency. Fully compatible with the Infiniband stack. RoCEv2 with multipath extension.
GPUBoost PCIe cards hook into the GPU host system via rdma-core, the Linux InfiniBand module and the MangoBoost device driver and export the GPU directly to Ethernet.
This ensures easy system integration of GPUBoost RDMA without requiring any application-level modifications.
Designed to support line-rate bandwidth for each network port, Mango GPUBoost RDMA is well-suited for high-performance scenarios such as large-scale AI/ML workloads or HPC applications.
GPUBoost RNICs use the opportunity of the RoCEv2 protocol and allow for more in-depth customization via their FPGA core, providing improved flexibility in comparison with other RNICs.
Micropolis is the first accredited reseller for MangoBoost DPU cards in the EMEA region.
We use a cookie to enable basic features of our website.
By clicking on "accept", and by your continued use of this website, you agree to our practice of using and storing an essential cookie in your browser.