Attachment 1 AMDT 2 Technical Specifications Final.pdf

TECHNICAL SPECIFICATIONS

May 5, 2026

GPU Compute Servies for Network of Optimal Dynamic Energy Signatures

1. INTRODUCTION: The Biological Technologies Office (BTO) at Defense Advanced Research Projects Agency (DARPA) requests GPU compute services to support the Network of Optimal Dynamic Energy Signatures (NODES) Program.

2. BACKGROUND: The NODES program within DARPA/BTO seeks to enable distributed, data-driven biological discovery by integrating experimental platforms, large-scale biological datasets, and advanced computational methods. NODES performers generate and analyze substantial volumes of heterogeneous biological data and rely on computational workflows to enable rapid hypothesis generation, model development, and validation across a networked ecosystem.

Execution of NODES technical objectives requires access to scalable, high-performance computational resources to support activities such as large-scale data integration, AI-enabled biological analysis, model training, and iterative experimentation. Additionally, NODES requires specific levels of security and access control in order to safeguard its data.

During base period, NODES performers will require a virtual resource to store data and perform computational model testing, along with appropriate management and security services. The contractor is required to provide large-scale availability of highly sought-after AI-specific hardware, with the ability to meet urgent government needs without program delays.

3. PERIOD OF PERFORMANCE: The period of performance will be for a 5-month base period and a 5-month Option 1.

4. SCOPE: The Contractor shall provide a dedicated, managed high-performance computing

(HPC) solution that includes compute hardware, network-attached storage, high-speed networking, robust security controls, and comprehensive user support. The solution must meet the performance, availability, and security standards detailed herein to support the NODES program's scientific objectives. The current requirement does not involve CUI and therefore does not require a FEDRAMP-certified environment.

4.1. Future Compute Services and Contract Continuation

Quoters are advised that the Government may require the continuation of services beyond the initial period of performance. In such an event, the scope of work may be modified to include the handling of Controlled Unclassified Information (CUI). This may necessitate a higher level of security compliance, including but not limited to, a FEDRAMP-certified environment.

5. REQUIREMENTS: The Contractor shall provide all personnel, equipment, supplies, facilities, transportation, tools, materials, supervision, and other items and non-personal services necessary to deliver and maintain a secure, high-performance computing environment as defined herein.

5.1. Technical Requirements

Provide access to fifty (50) dedicated, private NVIDIA HGX H100 80GB 700W SXM5 8X GPU-baseboards, equivalent NVIDIA specifications or better, for model training and inference, which are required for the level of scientific computation at this program:

Hardware that allows direct access, without intermediate layers, to streamline performer access without additional month-long waiting periods.

Low-latency systems, which are essential for NODES need for large-scale, multi- GPU AI training that requires maximum bandwidth between GPUs.

Software tweaks that allow a little older GPUs to perform specific applications as fast as newer GPUs are also welcome as part of the solution.

GPU availability for a start in mid-May 2026.

1 x86 Compute virtual machine.

Dedicated Internet Access on 100G Port.

100 additional NVIDIA HGX H100 80GB 700W SXM5 8X GPU-baseboards (equivalent or better) servers available and ready to use in the case of server outages.

550TB of NFS storage that can be utilized by the performers to store data and run computational models.

Resources optimized to large-scale, high performance, scientific computation, and multi-GPU training requirements.

Provide resources for replacing inventory on demand, and ability to provide replacement GPUs as necessary.

5.2. Service requirements:

24/7/365 Customer Support (tier 1) for up to 100 individual users Scientific software support (tier 3) 24/7/365 for up to 100 individual users Provide software installation, if needed, to the required compute environment(s) specifications (e.g., rail interconnection, node provisions, clusterware provisions, VAST NAS provision and interconnection, cluster testing, NCCL testing, customer hand-off).

Provide monitoring of user allotment to ensure enforcement of allocation limits on NODES users, as defined by the Government.

5.3. Security Requirements:

Provide Isolated Virtual Private Cloud (VPC) deployments, to allow default security and to provide NODES performers full control over data flows.

The contractor shall hold SOC 2 type II certification and must implement security controls.

6. DATA RIGHTS: The contractor shall not have any data rights to the data.

7. REPORTING: The Contractor shall provide detailed monthly reports on resource utilization and system performance. These reports shall serve as the basis for invoicing. The Government will be billed solely for actual metered consumption (actual GPU-hours used) up to the established Not-to-Exceed (NTE) contract ceilings. The Contractor shall also provide other metrics as requested by the government to monitor user allotment.

8. DELIVERABLES:

Technical Specifications

Para.

Deliverable Frequency # of Copies

Medium/Format Submit To

7 Monthly Usage & Performance Report

Monthly, by the 10th business day of the following month.

1 Electronic via Email

Contracting Officer's Representative (COR) and the DARPA Program Manager

7 Ad-hoc Government- Requested Metrics

As Requested

1 Electronic via Email

Contracting Officer's Representative (COR) and the DARPA Program Manager