MangoBoost, a leader in system solutions for AI data center efficiency, has achieved record-breaking performance in the MLPerf Storage v2.0 benchmark with its Mango StorageBoost™ solution. This milestone highlights the power of DPU-accelerated NVMe/TCP storage systems, delivering unmatched performance, scalability, and cost-efficiency for AI training workloads.
MangoBoost’s StorageBoost™ sets record in MLPerf Storage v2.0.
Achieves 6.2x GPU scalability for 3D-UNet on NVIDIA A100 GPUs.
Delivers up to 2.05x higher throughput per 400G bandwidth on H100.
Outperforms NVIDIA BlueField-3 with lower total cost of ownership.
NVMe/TCP Initiator and Target enable zero CPU overhead.
GPU Storage Boost ensures direct, efficient data transfers.
MangoBoost’s Mango StorageBoost™ solution demonstrated exceptional results in the MLPerf Storage v2.0 Fabric-attached Block Storage category. The submission, utilizing NVMe/TCP Initiator (NTI) and Target (NTT), achieved line-rate throughput over a 400G Ethernet fabric, providing near-local SSD performance for distributed AI workloads like 3D-UNet on NVIDIA A100 and H100 GPUs. It delivered up to 6.2x GPU scalability on A100 and 1.25x to 7.5x on H100, alongside 1.57x higher throughput per 400G bandwidth on A100 and up to 2.05x on H100. “The MLPerf Storage benchmark has set new records for an MLPerf benchmark, both for the number of organizations participating and the total number of submissions,” said David Kanter, Head of MLPerf at MLCommons.
The Mango StorageBoost™ architecture outperformed NVIDIA’s BlueField-3 DPU in both NVMe/TCP and NVMe/RDMA under equivalent conditions, offering superior performance-to-cost and significantly lower total cost of ownership. Deployed with NTI on the host and NTT on the storage server, connected via a 400G Ethernet switch, the system emulated demanding AI workloads with near-zero CPU overhead. This setup ensures maximum bandwidth utilization, making it ideal for large-scale AI training environments requiring high throughput and scalability.
Mango StorageBoost™ comprises three core components. The NVMe/TCP Initiator (NTI) offloads the entire NVMe/TCP stack to hardware, achieving full-duplex line-rate performance without CPU consumption. The NVMe/TCP Target (NTT) accelerates TCP/IP and NVMe-oF processing, enabling storage disaggregation over standard Ethernet with zero CPU involvement. The GPU Storage Boost (GSB) facilitates direct DMA transfers between GPU memory and storage, bypassing the CPU to enhance I/O efficiency. These components collectively deliver near-local SSD performance, optimizing AI training workflows.
Designed for immediate deployment, Mango StorageBoost™ integrates seamlessly with standard server platforms and GPUs, requiring no modifications to existing hardware or software stacks. This flexibility, combined with reduced CPU utilization and infrastructure costs, positions MangoBoost as a leader in AI storage solutions. The solution’s ability to handle complex AI workloads while maintaining cost-efficiency makes it a compelling choice for data centers aiming to maximize performance.
MangoBoost’s record-breaking MLPerf Storage v2.0 results solidify its position as a pioneer in DPU-accelerated storage for AI training. By delivering high performance, scalability, and cost savings, Mango StorageBoost™ enables enterprises to build efficient, future-ready AI infrastructure.
MangoBoost is a provider of cutting-edge, full-stack system solutions for maximizing compute efficiency and scalability. At the heart of the solutions is the MangoBoost Data Processing Unit (DPU), which ensures full compatibility with general-purpose GPUs, accelerators, and storage devices, enabling cost-efficient, standardized AI infrastructure. Founded in 2022 on a decade of research, MangoBoost is rapidly expanding its operations in the U.S., Canada, and Korea.
TechIntelPro empowers B2B technology decision-makers with the latest industry trends, deep insights, leadership perspectives and actionable intelligence to drive both organizational and professional milestones—and thrive on the tech pulse.