Enclosure No. 2B - Appendix B Statement of Work.docx

Appendix B Argonne National Laboratory Statement of Work Artemis Compute Cluster May 6, 2026

Background The Argonne Leadership Computing Facility (ALCF) requests proposals for the acquisition and installation of a compute cluster – to be named Artemis -- meeting the specifications below.

The primary purpose of the cluster is to serve as a Kubernetes and Virtualization environment for persistent services. The cluster will host both user and facility services with both types needing connectivity to existing facility resources, including (but not limited to) production HPC systems, storage, and networking.

Keywords Used in this Document The ALCF adheres to RFC-2119 as the governing document for the definitions of important words used in this Statement of Work. Proposals shall be evaluated using the following five definitions copied verbatim from RFC-2119. The ALCF does not adhere to RFC-8174 capitalization in this Statement of Work: the following five definitions apply regardless of the use of lowercase or uppercase.

· MUST: This word, or the terms "REQUIRED" or "SHALL", mean that the definition is an absolute requirement of the specification.

· MUST NOT: This phrase, or the phrase "SHALL NOT", mean that the definition is an absolute prohibition of the specification.

· SHOULD: This word, or the adjective "RECOMMENDED", mean that there may exist valid reasons in particular circumstances to ignore a particular item, but the full implications must be understood and carefully weighed before choosing a different course.

· SHOULD NOT: This phrase, or the phrase "NOT RECOMMENDED" mean that there may exist valid reasons in particular circumstances when the particular behavior is acceptable or even useful, but the full implications should be understood and the case carefully weighed before implementing any behavior described with this label.

· MAY: This word, or the adjective "OPTIONAL", mean that an item is truly optional. One vendor may choose to include the item because a particular marketplace requires it or because the vendor feels that it enhances the product while another vendor may omit the same item. An implementation which does not include a particular option MUST be prepared to interoperate with another implementation which does include the option, though perhaps with reduced functionality. In the same vein an implementation which does include a particular option MUST be prepared to interoperate with another implementation which does not include the option (except, of course, for the feature the option provides.)

Scope & Requirements

· Offerors must provide a proposal for the system and services specified in this Statement of Work; line items must be sub-totaled into the following categories:

· Hardware

· Installation

· Maintenance/Support

· Selection and procurement of the Kubernetes software distribution will be done by ALCF as part of a separate process and is not in scope of this Statement of Work.

Hardware The Artemis cluster is expected to consist of 36 individual nodes; each node should meet the following specifications:

· CPU:

· Physical core count at least 128

· CPUs should support a mode of confidential computing, e.g. AMD SEV-SNP, ARM CCA, etc.

· Memory: at least 1TB DDR5

· Storage:

· 2x240GB OS SSD disks in Hardware RAID 1

· 12-16TB Local Storage in appropriate Hardware RAID configuration.

· Local storage will be used for I/O performance-sensitive applications (databases, image caching, etc.); vendors should propose redundancy level appropriate to application use-cases.

· Networking:

· A minimum of one 10/100/1000Base-T Ethernet interface visible to the Linux operating system for in-band build and maintenance.

· A minimum of one separate 10/100/1000Base-T Ethernet interface not shared with the above required interface on an ILO, LOM, BMC, or equivalent hardware implementation for out-of-band build and maintenance.

· Support for speeds faster than 1000Base-T is permitted, but the termination must be an 8P8C modular connector (commonly referred to as “RJ45”) with Base-T autonegotiation.

· One port dedicated 1Gbps NIC for out-of-band management network.

· 1G Switches will be part of a separate procurement

· 2xNVIDIA Bluefield-3 B3240 Dual-Port QSFP112 adapters.

· Cabling for the 400G interconnect is not part of this procurement and will be provided and performed by ALCF.

· Power:

· Offerors must provide peak and expected power consumption per node as part of their response.

· Power cabling between servers and PDUs must be single cables, without the use of extenders, and must be of the correct ampacity.

· Power cabling and relevant devices must be NRTL listed.

Other requirements:

· Offerors must provide a comprehensive rack diagram of the proposed deployment.

· Offerors must provide a quote for both a full set of 36 nodes, as well as for individual nodes in the expectation of future expansion.

Software Compatibility

· The purpose of this cluster is to run Kubernetes in support of persistent service workloads. Several distributions of Kubernetes are being considered, including:

· Red Hat OpenShift

· Rancher Harvester/RKE2

· Talos Linux -- cluster compatibility with Talos Linux is explicitly desirable.

· Standard Kubernetes on top of a bespoke immutable OS layer

· Acquisition of the Kubernetes software distribution is not in the scope of this Statement of Work.

· Responses should include a description of known incompatibilities between offered hardware solutions and each of the Kubernetes distributions described above. Responses should also include remediations for any incompatibilities, if known.

Installation

· The offeror will be required to provide on-site installation services into a set of four (4) existing racks (HPE P9K58A) in the Building 240 Data Center, each with four (4) PDUs (HPE R4N79A).

· Argonne is extremely conscious to safety concerns; offerors will be required to adhere to all Argonne safety regulations while working on-site

· A Job Safety Analysis (JSA) will be jointly prepared by Argonne safety staff and the offeror prior to arrival for installation. The JSA will outline all safety requirements applicable to the installation.

· Offerors will be required to rack the nodes evenly across the provided racks, starting at the bottom of each rack working their way upwards.

· Offerors will be responsible for installation of copper cabling between server nodes and top-of-rack 1G switches for the management network (RU47, 48).

3.3.1 Acceptance Testing

Prior to taking ownership and making payment, the following lightweight acceptance tests must be successfully completed:

· The hardware must be racked per the agreed upon rack elevation and inventoried to ensure it matches the agreed upon Bill of Materials (BOM).

· Each server must successfully boot into a standard Linux distribution (the latest stable version of RHEL at the time of acceptance)

· All components must be discoverable by the OS and functional.

· All firmware versions must be current and consistent across the servers.

Maintenance, Support and Monitoring

· Offerors must describe the hardware support process as part of their response (how requests are to be submitted, the structure of the response process, etc.)

· ALCF requires five years of 9x5 next business day (NBD) service with on-site parts replacements. Next business day is defined in this case as the offeror committing to a maximum time of 24 hours from ticket creation to shipment of hardware replacement components if a hardware RMA is requested by ALCF personnel.

· While not required, maintaining a small spare parts cache on-site is highly desired. The Offeror may choose to quote it as a separate (optional) line item or choose not to offer it.

· The support contract must not prohibit the ALCF Operations Team from performing any applicable hardware and software maintenance for the life of the equipment following acceptance.

· The Offeror must provide on-premises licensing and software to enable advanced hardware monitoring. Any software provided/used for this purpose must be hostable on-premises and must not be exclusively cloud-based.

Project Management

· Offerors must provide contact information for primary points of contact for sales and technical queries.

· Offerors must provide a plan of installation actions prior to delivery.

Government Owned Property No government-owned property will be provided.

Security Access to the Argonne National Laboratory campus is controlled and requires a pre-arranged gate pass. Individuals who are not U.S. citizens are allowed but require a security investigation, will need to be identified, and must provide the required documentation a minimum of two weeks prior to installation.

Access to the building 240 data center is also controlled. An ALCF staff member will be assigned to work with the offeror and enable access while the offeror's installation team is working on-site.

Cybersecurity The offeror shall describe the process for remediating security vulnerabilities in hardware and software delivered by the offeror, including vulnerabilities discovered by third parties, vulnerabilities in upstream components, and vulnerabilities discovered by the company but not yet publicly known. In particular, the offeror shall describe timelines and the process for obtaining Common Vulnerability Enumeration (CVE) identifiers for vulnerabilities discovered within offeror's products.

Any products provided by the offeror shall have patches and fixes released in a timely manner that are compatible with updates provided to the standard operating system. It is expected that disclosure and remediation practices will follow industry best practices and government directives. Examples of those that are in effect at the time of publishing this document are BOD 22-01: Reducing the Significant Risk of Known Exploited Vulnerabilities and BOD 23-01: Improving Asset Visibility and Vulnerability Detection on Federal Networks. E.g. zero-day exploits should have mitigations in frame of days, and Critical/High CVEs may need to be patched in the order of weeks.

The offeror shall provide the ability, documentation, and automation required to rotate every secret in all components in the system to a unique non-shared credential without requiring physical interaction with the hardware. This includes but is not limited to passwords, tokens, SSL certificates/keys, signing keys, BMC passwords, and node/rack security configurations.

The offeror shall describe how they will disclose all default accounts and credentials upon system delivery. Any credentials added after delivery of the system will be disclosed in release notes.

Post-acceptance, Argonne requires that all the offeror’s staff who review logs be U.S. persons.

Place of Performance The Artemis cluster will be installed in the building 240 data center on campus at Argonne National Laboratory. The offeror must rack on-site into the existing racks as mentioned above.

Period of Performance System delivery should be as soon as possible based on component lead time/availability and the vendor manufacturing/assembly schedule. Installation and acceptance testing should occur as soon as is practical after system delivery but no more than 30 days after delivery. Hardware/software support, maintenance, and warranty services shall commence upon system acceptance and shall continue for a period of five (5) years thereafter.