AMD Lux opens to scientific users at Oak Ridge | Magica
Lux opens to scientists at Oak Ridge, splitting capacity between Genesis and paid access
Editorial Team
••📖7 min read
AMD says the Lux AI supercomputer is open to scientific users at Oak Ridge. Half its annual capacity supports the Genesis Mission and half is reserved for paid access, including AMD development, while workload limits and undisclosed rates shape its practical value.
AMD says Lux is open to scientific users at the Department of Energy's Oak Ridge National Laboratory.
Half its annual 3.5 million node-hours support the Genesis Mission; half are reserved for paid access, including AMD development and testing.
The user guide documents limits on large MPI workloads and persistent GPU services under Kubernetes.
Chipmaker Advanced Micro Devices said on November 16 that Lux, an AI-focused supercomputer built by server maker Hewlett Packard Enterprise (HPE) at the U.S. Department of Energy's Oak Ridge National Laboratory, is now open to scientific users. The opening adds dedicated AI capacity through a partnership that gives both government researchers and its private-sector co-investor access to the machine.
AMD describes Lux as the first dedicated U.S. AI factory for the Genesis Mission, the federal effort to use AI for scientific discovery. In its announcement, AMD identified allocated projects in fusion power plant design, materials development and advanced manufacturing, and forecasting water availability, streamflow and river temperature in the Tennessee Valley Authority region.
The arrangement exchanges shared investment for shared access. The Oak Ridge Leadership Computing Facility (OLCF), which operates Lux on DOE's behalf, reserves half the machine's annual capacity for paid use. AMD will use these cycles for development and testing and make a portion available to public- and private-sector partners. Federal science gains dedicated AI capacity while its chip supplier gains a large system on which to develop its technology.
A shared investment with a substantial entry commitment
The facility's access terms allocate half of Lux's annual 3.5 million node-hours to the Genesis Mission and half to paid partners. A node-hour measures use of one compute node for one hour. Proprietary projects pay the DOE User Facility rate, can start in any month and can run from six months to five years.
Records detail Chinese data-center investments linked to Hexi founder Kang Wei, husband of former Megaspeed chief Alice Huang. Filings document financing and Huawei business, but do not establish a route for restricted Nvidia chips.
Editorial Team
The minimum commitment is 175,000 node-hours over six months. Oak Ridge illustrates that scale as roughly 400 nodes, or 3,200 GPUs, running for eighteen days spread over the allocation period. Partners can spend their allocation on jobs ranging from a single node to the full system, although Oak Ridge cautions that nodes dedicated to persistent services may leave fewer than 4,000 GPUs available for full-system jobs.
The billing rules also reward careful resource requests. Under the user guide, Slurm jobs are charged for the resources they reserve and their elapsed running time, even if an application leaves some requested GPUs idle. Users can request individual GPUs, each accompanied by a share of CPU cores and system memory.
DOE announced Lux on October 27, 2025, presenting co-investment and shared computing infrastructure as a way to shorten delivery times from years to months. It then targeted deployment in early 2026. The November opening announcement establishes scientific access, without specifying when installation was completed, so it does not settle whether that earlier deployment target was met.
The original announcement distinguished Lux's partnership model from the traditional competitive procurement used for Discovery, the forthcoming HPE-and-AMD supercomputer intended to succeed Frontier at Oak Ridge. DOE said Discovery would arrive in 2028 and described more than $1 billion in public-private investment across the two systems. That combined figure is not a disclosed price for Lux alone.
HPE says it has made an in-kind investment of $37 million to advance the Genesis program and support research projects. That is a contribution, not a total system cost. The announcements and access terms give no dollar rate per node-hour, preventing a price comparison with alternative computing services.
Dedicated AI capacity builds on Frontier's infrastructure
Lux has 504 nodes and 4,032 AMD Instinct MI355X GPUs, according to the user guide. Each node combines eight GPUs, each with 288 GB of high-bandwidth memory, with two AMD EPYC CPUs. HPE supplies the liquid-cooled ProLiant XD685 servers and Juniper networking; the system also uses AMD Pensando networking and AMD's ROCm GPU software.
The partners bring an existing scientific-computing relationship. AMD and HPE supplied the technology behind Frontier, which broke the exascale barrier. Oak Ridge's Leadership Computing Facility has operated major supercomputers since 2004, with a deployment history including Titan in 2012, Summit in 2018 and Frontier in 2022. Lux extends that infrastructure with capacity dedicated to AI research.
Shared storage gives that relationship a practical use: researchers with both Frontier and Lux allocations can generate simulation datasets on Frontier and train models on Lux using the same Orion filesystem. The user guide lists Orion's usable namespace at 679 petabytes, shared infrastructure rather than storage exclusive to Lux. Allocations on the two computers remain separate, and Orion files are subject to a 90-day purge policy.
The Genesis Mission itself was launched by executive order on November 24, 2025. The order makes DOE responsible for implementation and directs it to integrate computing, scientific data, foundation models and research agents through the American Science and Security Platform. Implementation is subject to available appropriations. Lux supplies computing capacity toward that broader effort; its opening alone does not establish that the integrated platform is complete.
HPE says Lux entered the TOP500 ranking at No. 14 with 220.79 petaflops of performance. The ranking supplies evidence of computing scale, while the user documentation gives researchers a more specific guide to which workloads are ready.
Lux supports both Slurm, used to schedule supercomputing jobs, and Kubernetes, used to manage containerized applications. Oak Ridge says the partition sizes will change with demand. But the guide describes Lux as primarily centered on AMD's ROCm Collective Communication Library and says MPI, the communication interface used by many parallel scientific applications, is not yet officially verified. Because of tested scaling limits, the facility strongly recommends restricting MPI workloads to 16 nodes.
Kubernetes GPU access also has boundaries. GPUs can currently be used only in time-limited Jobs, with a maximum requested runtime of 48 hours. The guide says they cannot be used for persistent services and that support for other GPU workload types remains in development. Researchers planning continuously running GPU applications therefore need to account for capabilities that are still pending.
Data eligibility is another condition. Oak Ridge says Lux uses DOE Moderate security controls, allowing export-controlled applications and data. Support for the stronger Moderate-Enhanced controls is planned; the facility describes support for International Traffic in Arms Regulations and protected health information as dependent on that future addition.
The next test is useful access and measurable science
AMD directs prospective users to the Oak Ridge Leadership Computing Facility to discuss computing needs and access pathways. Paid partners need the applicable rate and allocation terms; scientific teams need to establish whether their communication patterns, runtime requirements and data protections fit the available service.
The Genesis order supplies a near-term accountability point: under its reporting requirement, within a year of its November 24, 2025 signing, DOE must submit a report to the president covering the platform's operational capabilities, user participation, measurable scientific advances and public-private partnership outcomes. The requirement concerns the mission as a whole, rather than Lux alone.
For Lux, the evidence still needed is application-level performance and results from its allocated projects: whether models improve fusion design, materials development or regional water forecasts, and how efficiently the software uses the hardware. AMD's opening announcement identifies those projects but does not supply their results. Those measurements, alongside actual access costs and service availability, will determine how much scientific value the new capacity delivers.
ElevenLabs plans to acquire Kotoba, the startup behind Japanese speech technology used in Salesforce’s Agentforce Voice. The deal adds a 15-person team and deployed products to its existing Japanese operation, but financial terms and integration plans remain undisclosed.
Tesla plans €3 billion of expansion at its profitable Grünheide factory to build Cybercabs from 2029, eventually targeting up to one million annually. The driverless taxi still lacks German and EU approval, and its capacity target differs from an older plan for the whole site.
Our World in Data’s comparison finds negative views of AI exceed positive ones across five probability-based surveys. The primary results also reveal substantial mixed views, acceptance of some applications and a sharper divide over who controls development and bears its local costs.
NVIDIA's Nemotron for Science pairs an existing open model with nearly 50,000 chemistry questions and scientific post-training recipes. Partner results suggest scope for specialized agents, but the reported accuracy and inference savings do not establish the full cost of adapting them.
Apple is adding app-by-app and folder-specific permissions, repeat approval requests and clearer warnings in macOS 27.3 beta 2. Existing permissions remain intact, limiting the immediate reach of the privacy changes.
OpenAI reportedly discussed buying Goodfire for more than $5 billion, over four times its February financing valuation. No deal resulted. A disclosed paid assessment agreement offers cooperation without ownership, while the technology’s safety benefits remain to be demonstrated.
Gemini’s regional voice rollout includes more than ten UK options and applies to live conversations and spoken chat responses. Earlier accent consistency problems remain untested in the new release, while pricing and rollout eligibility are unspecified.
Hugging Face is moving language-model requests to OpenRouter’s provider router, adding automatic failover while ending custom provider keys on that route. Provider rates carry no Hugging Face markup, but the selected service—and the resulting bill—can change.
Nasiko’s OpenRuntime combines agent spending reports, model routing and gateway budget limits. Earlier releases and company benchmarks show what it can do, but metering disputes, workload trade-offs and runtime charges complicate the savings case.
A report citing business records identifies Anthropic cybersecurity researcher Paul Crowley as a manager of Red Means No. The disclosure adds to scrutiny of consent safeguards and professional networking at the Bay Area sex parties.
Reported internal Microsoft data puts Claude for Microsoft 365 usage below one-tenth of Copilot’s. Undisclosed measurement details and Claude’s availability inside Copilot limit what the gap says about Anthropic’s workplace business.
South Korea is strengthening device checks and considering public Wi-Fi shutdowns before its November 19 college entrance exam, after a tax-accountant exam top scorer admitted using AI to cheat. The response still depends heavily on human screening, with no electronic device detectors planned for this year's test.
Russia’s objections to observer participation stalled the opening of a Geneva weapons conference, where governments must decide whether to authorize negotiations on autonomous weapons and whether any resulting rules should be binding.
Apple’s iOS 27.3 beta ties higher allowances for server-based AI to existing iCloud+ plans. Basic Siri commands remain free, but undisclosed daily limits leave the value of paid access uncertain.
The partners plan to combine IBM quantum processors, NVIDIA computing and Fermilab physics expertise to study modular quantum systems. A separate decoder release promises faster error-correction research, with important limits on its performance comparisons.
Federal Reserve Chairman Kevin Warsh outlined AI scenarios spanning shared prosperity, job displacement and an investment downturn. The exercise offers no forecast or rate guidance as colleagues worry that the infrastructure buildout is adding to inflation.
OpenAI is offering approved organizations a free classifier for suspected child sexual abuse images, including AI-generated material. Early results suggest less review work, but existing detection alternatives and the need for human review qualify the launch.
Nvidia plans scientific computing racks for early 2027, extending Vera beyond its initial AI deployments. Its 2.4× performance-per-watt claim comes from one airflow simulation against AMD Turin; performance against newer rivals and delivered costs remain unresolved.
OpenAI’s CEO expressed empathy for Nvidia founder Jensen Huang but withheld judgment on reinstatement. His confidence in continued AI financing comes with a direct commercial stake and unresolved deployment risks.