AI Security A100 GPU Nodes

This is a Combined Synopsis / Solicitation for Commercial Items prepared in accordance with the format in FAR Subpart 12.6 - Streamlined Procedures for Evaluation and Solicitation of Commercial Items - as supplemented with additional information included in this notice. The solicitation is issued using simplified acquisition procedures under the authority of FAR Part 13.5. THIS ANNOUNCEMENT CONSTITUTES THE ONLY SOLICITATION. QUOTATIONS ARE BEING REQUESTED AND A WRITTEN SOLICITATION DOCUMENT (i.e. STANDARD FORM) WILL NOT BE ISSUED. This solicitation is a Request for Quotation (RFQ). The solicitation document and incorporated provisions and clauses are those in effect through Federal Acquisition Circular 2022-04 dated 01-30-2022. The associated North American Industrial Classification System (NAICS) code for this procurement is 334111, Electronic Computer Manufacturing. The small business size standard is 1,250 employees. This acquisition is 100% set aside for small businesses. DESCRIPTION OF REQUIREMENTS Background The Information Systems Group (ISG) at the National Institute of Standards and Technology (NIST) is expanding its research efforts into AI security, including both applications of AI algorithms to novel datasets, and foundational research into new AI methods. To support these efforts, new compute capabilities are required for training, inference, and evaluation of AI models. These new compute capabilities need to interoperate with the group's existing facilities. AI models are known for being extremely compute intensive, requiring dedicated computation accelerators attached to the machine. Modern graphics processing units (GPUs) are the most commonly used accelerators in the deep learning space. Current AI training frameworks leverage these accelerators directly to perform the majority of the heavy lifting and math required to train state of the art AI models. The group is conducting several broad investigations into the security of AI models. First, ISG is creating large scale novel security AI datasets. For example, injecting hidden behavior into AI models for the TrojAI competition. In addition, ISG is evaluating AI detectors that are trying to identify if an AI model has hidden behavior. Both of these tasks require significant compute resources. Purchasing devoted server class CPU and GPU nodes enables developers and researchers to make progress in cultivating trust in IT metrology and AI. This platform must provide not only compute capability, but high bandwidth networking interconnects to load the data AI models are trained on from the network attached storage (NAS) system. Objectives The scope of this procurement is the provision of one (1) 4U Dual-Socket GPU Server with 8x A100 80 GB SXM4 GPUs and 1 TB Memory, and one (1) 4U Dual-Socket GPU Server with 8x A100 80 GB SXM4 GPUs and 2 TB Memory, each with a warranty. Brand Name Only Justification: IAW FAR Authority 13.106-1(b) Solicit from a Single Source The central processing units (CPUs), graphics processing units (GPUs), and InfiniBand Network Interface Cards are brand name specific. The group's existing nodes cluster uses Advanced Micro Devices (AMD) - Epyc Rome and - Epyc Milan series CPUs. To interoperate with this cluster and enable compilation and software development, the nodes must have the same CPU architectures as the other nodes. The NVIDIA-based GPUs have thousands of GPU cores and high memory bandwidth. It is the only available product line that operates with the CUDA general purpose programming language to take advantage of these cores. NVIDIA is the only company that produces chips that target matrix multiplication throughput with its special-purpose - Tensor Cores that accelerate compute intensive workflows, in particular AI. The math library CUDA Basic Linear Algebra Subroutines (cuBLAS) and its general matrix multiplication functions have delivered tremendous performance when using these - Tensor Cores. NVIDIA GPUs are required to be able to use this functionality. The NVIDIA Mellanox InfiniBand network interface cards (NICs) are commonly used at NIST on other systems such as Enki and the Software and System Divisions network 200 Gbit/s network switch; the NVIDIA Mellanox InfiniBand NIC is a familiar card that is the only compatible cards with our current network switch. If another brand card is used, it will be incompatible with the current hardware and is unable to provide the required network throughput to our network attached storage arrays. Additionally, the NVIDIA Mellanox cards are the only cards on the Imarket that offer 200 Gbit/s InfiniBand for PCIe gen 4 systems. Procuring these brand name only components is essential for the NIST ISG as all of the above functionality is required in order to meet mission objectives. Required Specifications Please see the attached Statement of Requirements for details. DELIVERY Delivery is required F.O.B Destination at a mutually agreed time within 90 Days after contractor receipt of order. See Statement of Requirements for additional, specific delivery instructions. PAYMENT TERMS The Contractor shall be paid, in accordance with Net 30-day payment terms, upon receipt and acceptance of a proper invoice, in accordance with the following schedule: 100% after receipt and acceptance by the TPOC of the fully delivered systems. The Government anticipates inspection will occur upon: After the testing procedures set forth in this document have been complete At NIST's discretion: a. NIST may allow partial shipment of equipment depending on lead times. APPLICABLE PROVISIONS AND CLAUSES Please see the attached document for applicable provisions and clauses. QUESTION AND ANSWER PERIOD CAR 1352.215-73 INQUIRIES (APR 2010): Offerors must submit all questions concerning this solicitation in writing electronically to Ms. January Magyar, Contract Specialist, at january.magyar@nist.gov. Questions must be received by or before July 7, 2022 at 5:00 PM ET. Any responses to questions will be made in writing, without identification of the questioner, and will be included in an amendment to the solicitation. Even if provided in other form, only the question responses included in the amendment to the solicitation will govern performance of the contract. (End of Provision). DUE DATE FOR QUOTATIONS All quotations must be submitted via email to Ms. January Magyar, Contract Specialist, at January.magyar@nist.gov. Submission must be received not later than July 12, 2022 at 11:00 AM ET. REQUIRED SUBMISSIONS 1. For the purpose of evaluation of Technical Capability: Documentation and product literature, which clearly demonstrates that the items being quoted conform to the requirements identified herein. Documentation shall discuss specifications and warranty of the quoted products, demonstrating a line by line compliance with the specifications detailed in the Statement of Requirements. 2. For the purpose of evaluation of Price: A firm-fixed price quotation in accordance with the required line item structure noted below. Any shipping/delivery fees must be included in the quotation. 3. Contractor Unique Entity Identifier (UEI) UEI shall be used to verify the offeror's active System for Award Management (SAM) registration. Offerors must have an active registration at www.SAM.Gov to be considered for award. 4. Completed Provisions Please complete the highlighted areas in the applicable provisions and clauses documents and return with submissions. CONTRACT LINE ITEM STRUCTURE: Responsible Offerors shall provide pricing for the following line items: LINE ITEM 0001: Delivery of one (1) 4U Dual-Socket GPU Server with 8x A100 80 GB SXM4 GPUs and 1 TB Memory meeting in accordance with the Statement of Requirements, inclusive of any Delivery Fees LINE ITEM 0002: Delivery of one (1) 4U Dual-Socket GPU Server with 8x A100 80 GB SXM4 GPUs and 2 TB Memory in accordance with the Statement of Requirements, inclusive of any Delivery Fees EVALUATION CRITERIA AND BASIS FOR AWARD In accordance with FAR 52.212-2: Evaluation - Commercial Items as included in the attached Applicable Provisions and Clauses document, the Government intends to award a purchase order resulting from this solicitation to the responsible offeror whose quotation, conforming to the solicitation, is the Lowest Price, Technically Acceptable quote. Technically acceptable means that the quotation meets the following requirements: Quotation clearly includes all products, in accordance with the required specifications, quantities, and warranty features, as identified in the Statement of Requirements Quotation meets delivery requirements (Delivery within 90 Days ARO) Price will be evaluated for reasonableness.