Salient characteristics for ACQ-26-A47 AI_ML Work station.pdf

PDF 93 KB Posted

Attached to
Combined Solicitation Notice - High-Performance AI/ML Workstation (NVIDIA Grace/Blackwell Architecture)
Solicitation number
ACQ-26-A47
Issued by

About this file

This document is a Salient Characteristics (Technical Specifications) document issued by the Department of the Air Force, 412th Test Wing at Edwards Air Force Base for a High-Performance Artificial Intelligence and Machine Learning Localized Workstation (Requirement Tracking Number: UR-ACQ-26-A47).

The government requires one enterprise-grade deskside AI/ML workstation for fine-tuning, optimizing, and executing offline Large Language Models and deep learning neural networks. The system is restricted to NVIDIA Grace/Blackwell hardware and CUDA architecture to ensure compatibility with existing laboratory software. The Exxact Valence Workstation (Part Number: VWS-158270643) is established as the baseline technical standard, with "or equal" alternatives considered if they meet or exceed the specified salient characteristics.

Minimum required specifications include: a 72-Core NVIDIA Grace CPU integrated with an NVIDIA B300 GPU (GB300) on a single board with 1400W TDP support; 496GB LPDDR5X system memory combined with 252GB HBM3e providing 7.1 TB/s bandwidth; exactly two 2TB M.2 NVMe PCIe 5.0 SSDs pre-configured in a hardware-integrated OS MDADM RAID 1 array (locked and non-reconfigurable); one NVIDIA RTX PRO 2000 Blackwell secondary GPU with 16GB GDDR7 ECC memory in a dual-slot, half-height design; PCIe 5.0 expansion slots including one x16 FHHL slot, two x16 single-wide slots (x8 bandwidth), and one x1 M.2 slot; two additional M.2 NVMe PCIe 6.0 x4 SSDs routed through an NVIDIA CX8 switch; minimum two 400G QSFP112 ports with 800 GB/s peak bandwidth, one 10GBase-T port, and one GbE management port; 1600W 80 PLUS Titanium PSU with C20 inlet; standalone tower chassis with maximum 595mm depth; Ubuntu 24.04 (LTS) preinstalled with NVIDIA AI Development Tools including CUDA Toolkit and drivers; and a 3-Year Parts and Labor hardware warranty provided by the OEM or authorized system integrator.

View the file

Other files

File Type Posted
DF Below SAT -NVIDIA_signed.pdf PDF

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

DEPARTMENT OF THE AIR FORCE

412TH TEST WING | EDWARDS AIR FORCE BASE, CALIFORNIA

Document Type: Salient Characteristics (Technical Specifications)

Requirement Tracking Number: UR-ACQ-26-A47

Core Subject: High-Performance Artificial Intelligence (AI) and Machine Learning (ML) Localized Workstation

1. Scope and Technical Overview

The government requires one (1) high-performance, enterprise-grade deskside AI/ML workstation. This system is a critical laboratory asset used to fine-tune, optimize, and execute offline Large Language Models (LLMs) and deep learning neural networks.

To ensure software compatibility with existing laboratory models, pre-compiled deep learning frameworks, and proprietary software packages, this system is restricted to the NVIDIA Grace/Blackwell hardware and parallel computing (CUDA) architecture.

The Exxact Valence Workstation (Part Number: VWS-158270643) is established as the baseline technical standard. A quotation for an "or equal" system will be considered if it meets or exceeds the physical, functional, and performance salient characteristics detailed in Section 2 below.

2. Salient Characteristics (Minimum Required Technical Specifications)

Any proposed system must meet or exceed the following technical thresholds:

A. Core Processor & Computational Superchip (Brand Name NVIDIA)

Integrated CPU/GPU Superchip: Must feature a tightly integrated co-processor architecture combining a minimum of a 72-Core NVIDIA Grace CPU and an NVIDIA B300 GPU (GB300) on a single cohesive board, supporting high-bandwidth interconnectivity.

Thermal Design Power (TDP): Must support a power envelope of up to 1400W TDP for the primary superchip.

B. System Memory (High Bandwidth / Low Latency)

System and GPU Memory: Must include a minimum of 496GB of LPDDR5X system memory combined with a minimum of 252GB of High-Bandwidth Memory (HBM3e), providing an ultra-high memory bandwidth of at least 7.1 TB/s for local AI model hosting.

C. Primary Operating System Storage (Non-Changeable Configuration)

Physical Solid-State Drives: Must include exactly two (2) 2TB M.2 NVMe PCIe 5.0 SSDs.

OS Array Configuration: The system integrator/contractor must pre-configure these two drives in a hardware-integrated OS MDADM RAID 1 array prior to shipment. This configuration is a critical security and local accreditation baseline; it must be delivered locked and cannot be subject to post-delivery reconfiguration or modification.

D. Secondary Accessory Graphics Processing Unit (GPU)

Secondary GPU: Must include one (1) NVIDIA RTX PRO 2000 Blackwell (or technically superior) desktop graphics card.

GPU VRAM: Minimum of 16GB GDDR7 ECC dedicated memory.

Form Factor / Slot Constraint: Dual-slot, half-height active fan design drawing a maximum of 70W to ensure compatibility with standard workstation power and airflow layouts.

E. Accessory Storage & Expansion Slots

Expansion Slot Capacity: The system board must provide:

Minimum of 1x PCIe 5.0 x16 Full-Height Full-Length (FHHL) slot (dedicated to offboard GPU support).

Minimum of 2x PCIe 5.0 x16 Single-Wide slots (wired with x8 signal bandwidth).

Minimum of 1x PCIe x1 22x32 M.2 slot for wireless networking and Bluetooth connectivity.

Accessory Storage: Must include an additional two (2) M.2 NVMe 22x80mm PCIe 6.0 x4 SSDs routed through an on-board high-speed switch (NVIDIA CX8 or equivalent).

F. Networking & Input/Output Interfaces

Ultra-High-Speed Networking: Must feature a minimum of two (2) 400G QSFP112 ports utilizing an integrated high-speed controller (NVIDIA ConnectX-8 SuperNIC or equivalent), supporting up to 800 GB/s peak bandwidth.

Standard Networking: Must feature a minimum of one (1) 10GBase-T (RJ45) standard network port and one (1) dedicated GbE management port.

G. Electrical, Power, and Enclosure

Power Supply Unit (PSU): Must feature a minimum of a 1600W 80 PLUS Titanium certified power supply with a C20 inlet.

Power Cable: Must include one (1) heavy-duty C19 to NEMA 5-15P 6-foot power cable designed to support a 2000W load.

Enclosure Form Factor: Standalone tower chassis with a maximum case depth of 595mm (23.4 inches) to fit standard laboratory workbenches.

H. Operating System & Software Preinstallation

Base OS: Must be delivered with Ubuntu 24.04 (LTS) preinstalled.

AI Developer Toolkit: Must include standard NVIDIA AI Development Tools (including CUDA Toolkit, NVIDIA drivers, and parallel computing libraries) preinstalled and fully tested for immediate out-of-the-box local model training.

I. Warranty and Support

Coverage: Must include a comprehensive 3-Year Parts and Labor hardware warranty and maintenance agreement provided directly by the OEM or authorized system integrator.

File details come from the government source that posted it. Updated .