eRacks/AISLING

The eRacks/AISLING is a 4U rackmount AI inference server, eRacks-Certified and burned-in in our shop, sized for teams that want to run large language models (LLMs, the AI models behind chat assistants) on hardware they own rather than through a cloud API. Your data never leaves the building, and there are no per-seat or per-token fees: one purchase, unlimited use.

Platform

AISLING is powered by the AMD Ryzen Threadripper 9960X (24 cores, 48 threads) on a workstation-class board with the PCIe lanes (the high-speed connections between the processor and the graphics cards) to drive multiple GPUs at full bandwidth. It ships in a heavy-duty 4U rackmount chassis with a single 1600W power supply. A server-class AMD EPYC platform upgrade is available for buyers who want more memory channels or a dual-processor option.

AI GPUs and memory

The base configuration includes two Intel Arc Pro B70 32GB GPUs (64GB of total VRAM, the fast on-card memory that holds the model), expandable to three for 96GB of VRAM, enough to run 70B-class models (roughly 70 billion parameters) at 4-bit quantization with headroom for context. Other GPU tiers are available, from low-profile cards up to NVIDIA RTX PRO 6000 Blackwell. System memory starts at 128GB of DDR5 ECC (error-correcting memory, which detects and fixes memory errors), configurable higher.

Storage and software

AISLING includes two mirrored NVMe boot drives and enterprise SATA storage, all configurable. Every unit ships with Ubuntu LTS (long-term-support Linux) and the full open-source AI stack (Ollama, Open WebUI, vLLM, llama.cpp, PyTorch) pre-installed, burned in (run under sustained load before shipping), and tested. Your staff reach the AI from a browser on day one, and the system is air-gap ready (able to run with no internet connection at all).

Where it fits

AISLING is the value 4U in the eRacks AI line: a real Threadripper platform and 128GB of ECC memory at an accessible price, right for a team standing up private chat, document Q&A (RAG, retrieval-augmented generation, where the AI answers from your own documents), and everyday inference. Need four GPUs, redundant power, or a server-class processor? Step up to the eRacks/AILEEN. Configure your exact build below, or ask us to size one to your models and user count at no charge.