RFQ -Requirements Specification AI Workstation for Offline LLM.pdf
PDF 125 KB Posted
- Attached to
- NVIDIA GB300 Tower Workstation
- Solicitation number
- N00173-27-Q-1301388970
About this file
This document is a Requirements Specification for an AI Workstation designed for offline large language model development, training, and execution.
The workstation requires an NVIDIA GB300 Superchip integrating a 72-core NVIDIA Grace CPU and B300 Tensor Core GPU with 1400W maximum power draw, paired with 496 GB of LPDDR5X system RAM and 252 GB of HBM3e GPU memory. An additional NVIDIA RTX 2000 Ada Generation GPU (16GB GDDR7 ECC) must be included with PCIe 5.0 x8 interface connectivity. Storage consists of redundant high-speed NVMe in RAID 1 configuration (2x 2TB PCIe 5.0 M.2 SSDs) for the operating system and critical applications, plus two additional M.2 NVMe PCIe 6.0 x4 slots for high-throughput data storage. The system must feature a 1600W 80+ Titanium certified power supply, 2x 400Gb/s QSFP112 ports, 1x 10GBase-T Ethernet port, and 1x dedicated Gigabit Ethernet port for system management. Expansion capabilities include 1x PCIe 5.0 x16 Full-Height, Full-Length slot and 2x PCIe 5.0 x16 Single-Wide slots operating at x8 bandwidth. The chassis must be a tower workstation form factor with maximum depth of 595mm (23.4 inches).
Software requirements specify Ubuntu 24.04 LTS (or more recent stable version) as the operating system and mandatory pre-installation of the NVIDIA AI Development Toolkit, including necessary drivers, CUDA libraries, and AI frameworks.
View the file
On GovTribe
Work with this file
- Download the original file
- Contacts named in this file
- Similar government files
- Ask GovTribe AI about this file
Text version
Requirements Specification: AI Workstation for Offline LLM
1. Overview This document outlines the minimum hardware and software specifications for a workstation capable of developing, training, and running large language models in an offline environment.
The system must provide substantial computational power, high-speed memory and storage, and a pre-configured software environment for AI development.
2. Component Specifications
Component Requirement
CPU/GPU Module NVIDIA GB300 Superchip, integrating a 72-core NVIDIA Grace CPU and an NVIDIA B300 Tensor Core GPU, with a combined maximum power draw of 1400W.
System Memory 496 GB LPDDR5X RAM integrated with the CPU.
GPU Memory 252 GB HBM3e High-Bandwidth Memory integrated with the GPU.
Additional GPU NVIDIA RTX 2000 Ada Generation (or newer), 16GB GDDR7 ECC; PCIe 5.0 x8 interface.
Operating System Storage
Redundant high-speed NVMe storage: 2x 2TB PCIe 5.0 NVMe M.2 SSDs configured in a RAID 1 array for OS and critical applications.
High-Speed Data Storage
2x M.2 NVMe PCIe 6.0 x4 slots for additional high-throughput storage.
Networking - 2x 400Gb/s QSFP112 ports- 1x 10GBase-T Ethernet port- 1x dedicated Gigabit Ethernet port for system management
Expansion Slots - 1x PCIe 5.0 x16 Full-Height, Full-Length slot- 2x PCIe 5.0 x16 Single- Wide slots (operating at x8 bandwidth)
Power Supply 1600W 80+ Titanium certified power supply unit.
Chassis Tower workstation form factor with a maximum depth of 595mm (23.4").
3. Software Requirements
• Operating System: Ubuntu 24.04 LTS (or a more recent stable version).
• Development Tools: The system must come pre-installed with the NVIDIA AI Development Toolkit, including necessary drivers, CUDA libraries, and AI frameworks.
File details come from the government source that posted it. Updated .