Aivres

Overview

Aivres is an AI server vendor offering OCP-compliant full-stack AI infrastructure solutions built around NVIDIA GPUs. The company segments its portfolio into three tiers by model parameter scale: the OCP Full Stack Solution on NVIDIA GB200/GB300 NVL72 (KRS8000/KRS8500) targeting trillion-parameter models (>1T, e.g. GPT-4 at 1.8T parameters), the OCP OAM Module Solution on NVIDIA HGX B200/B300 (KR9288 10U / KR5288 5U DLC) for very large models (10B–1T, e.g. LLaMA 70B/405B), and a PCIe Card GPU Solution on NVIDIA H200/RTX PRO 6000/B300A (KR4268/KR6268) for large models under 10B parameters.

The KRS8500 OCP Full Stack rack integrates 72 NVIDIA Blackwell GPUs with networking, storage, and a liquid cooling system in a single rack, including Aivres Full Mode Monitoring and Aivres Diagnosis Automation. The secondary liquid cooling loop employs comprehensive three-tier leakage detection, redundant temperature sensors, and real-time monitoring of conductivity, turbidity, and pH. Liquid cooling quality assurance is structured across four stages: design assurance (BIM collision prevention, flow resistance performance), manufacturing assurance (production process, quality inspection, factory pre-installation), service assurance (service training, annual inspection), and construction assurance (project management, professional debugging).

Aivres leverages Open BMC with Redfish to unify server diagnosis automation across pre-boot, booting, and post-boot stages, replacing disparate bus protocols (I2C, USB, BIOS SSIF) with a single API for state registers, telemetry, and firmware updates. The supporting software stack includes CUDA, CUDA-X, Magnum IO, NCCL, DCGM, MPI, NVML drivers; DL frameworks (TF, PyTorch, MXNet, JAX, NeMo Megatron); Ubuntu 22.04 OS; BMC with PLDM Type 5/Redfish/IPMI; and an AI infrastructure management tool achieving 92% diagnostic accuracy. Aivres commits to delivering guaranteed cluster application performance at order-time agreed criteria.