Aivres Drives AI Innovation with AMD Instinct™ MI350 Series GPUs Unmatched Performance and Efficiency
SAN JOSE, Calif.—June 12, 2025— Aivres, an AI, data center, and cloud computing solutions provider, announced support for AMD Instinct™ MI350 series GPUs in their AI server offerings.
The AMD Instinct MI350 series GPUs redefine AI and HPC performance with breakthrough compute power, memory, and efficiency to accelerate AI innovation. Built on the 4th Gen AMD CDNA™ architecture, the AMD Instinct MI350 series GPUs deliver industry leading 288 GB HBM3E memory, 8 TB/s high bandwidth, and expanded datatype support to meet diverse, evolving demands of AI and HPC. They are designed to maximize computational throughput, memory bandwidth utilization, and energy efficiency to empower AI workloads from LLM training and inference to high-performance computing tasks like scientific simulations and computational modeling. The AMD Instinct™ MI350 series also come with both air-cooled and direct liquid-cooled (DLC) configurations, supporting high density compute with advanced cooling infrastructure at scale.
The Aivres KR9298 is a 10U 8-GPU high-performance heterogeneous accelerated AI server supporting two AMD EPYC™ 9005 processors and the AMD Instinct MI350X platform, integrating 8 GPU OAM modules fully connected via 4th-Gen AMD Infinity Fabric links onto an industry-standard OCP design, providing a ready-to-deploy platform to accelerate time-to-market. Powered with AMD Instinct MI350X, KR9298 is poised to deliver up to 2.3 TB total HBM3E memory, 48 TB/s peak memory bandwidth, and up to 36.8 PFLOPs and 72 PFLOPs performance for FP16 and BF16 respectively.
Besides setting new standards in AI compute performance, Aivres KR9298 boasts additional design features and benefits that make it ideal for demanding AI workloads and scalable compute environments in the data center. It maximizes thermal efficiency with 20 fans and ample headroom for powerful air cooling. Its front-cabling configuration lowers I/O component temperatures by 20%, significantly reducing failure rates, and fewer rear cables improve overall system cooling, ensuring stable operation for large-scale models. With up to 12 PCIe slots, KR9298 supports flexible I/O expansion for high scalability and diverse workloads. The server also employs a stable and reliable power design, with N+N redundancy on partitioned PSU to CPU nodes and N+1 redundant power supplies to ensure data integrity, consistent performance, and operational security. Its fully modular chassis design provides ease of operation and maintenance as well as limitless scalability in the data center.