Matrix-free linear-system and extreme-eigenvalue computation with independent true-residual checking.
Discuss krylov_scaleTHE PRODUCT, IN PRACTICE
Some scientific workloads can apply a very large operator without storing its full matrix. krylov_scale is built for that setting: it solves supported linear systems and extreme-eigenvalue problems using the operator interface, then recomputes the true residual to check the result. The customer receives a numerical component for a specific operator, not a claim of limitless matrix capacity.
An application may know how to apply an operator even when storing every matrix entry is impractical.
krylov_scale accepts the supported matrix-free operation rather than requiring the full operator to be retained as a dense matrix.
Internal convergence reporting can differ from checking the final candidate against the actual problem.
The independent result check recomputes the true residual for the returned numerical candidate.
Setup and execution overhead can matter alongside numerical convergence.
Evaluate the operator path, memory and end-to-end time to the required tolerance. The separate Krylov upgrade targets the execution overhead identified in the campaign.
The product scope below describes the capability and outputs to confirm for your release. Custom integration, operated delivery and additional engineering are separately agreed.
Supported linear-system computation through operator application.
The additional supported eigenvalue computation path.
The solution or eigen-result returned for the selected problem.
Independent recomputation of the final numerical residual.
krylov_scale is fully available with CLI, MCP and headless access. Use the interface that fits your terminal, your agents or your automated workflow.
Run the product from a terminal or script, alongside the tools your team already uses.
Get CLI accessConnect MCP-compatible clients and agents to the product’s exposed capabilities.
Get MCP accessOperate without a graphical interface in unattended, scheduled or pipeline-driven work.
Set up headless accessProduct-specific inputs, hardware requirements and result guarantees still apply. An access mode does not change your license or deployment scope.
Explore all product interfacesA simulation team can apply its supported operator but does not want to materialize the entire matrix.
Illustrative use of the supported scope, not a customer case study or a measured result.Provide the operator, right-hand side or eigenproblem and tolerance.
Run the supported matrix-free numerical computation.
Recompute the true residual and compare total cost against the existing solve.
Simulation, computational-science and engineering teams with matrix-free numerical operators.
CLI, MCP and headless access are all available. Run from a terminal or script, connect an MCP-compatible client, or use the product without a graphical interface. Choose the mode during onboarding and receive its setup instructions. Runtime requirements and supported operations remain product-specific.
The billion-unknown Poisson result is a specific reported operator and H100 configuration. It does not establish capacity for arbitrary stored matrices. A small residual does not alone guarantee a small solution error for an ill-conditioned problem.
The agreed krylov_scale release, its supported interfaces and outputs, the applicable checks and the support stated in your order. The plan defines users, workflow, environment and capacity. Source-code transfer, unrestricted redistribution and bespoke engineering are not automatically included.
Request a scoped engineering or operated-service proposal. It identifies what our team delivers, where the workload runs, infrastructure costs, acceptance requirements and ongoing support. Your software subscription alone does not activate a hosted runtime.
Explore engineeringP42 / Company-reported. The supplied campaign reports a billion-unknown Poisson workload fitting on an H100. The new Krylov track targets execution overhead identified in that campaign; no unmeasured successor speedup is inferred.
These descriptions come from the supplied product and campaign material. Website checks are not an independent rerun of the engine. Qualification identifies the actual release, workload and available artifacts.
Read the evidence summaryCustomer-hosted access and a service operated by SpinForge are different scopes. Confirm the release, deployment and included support before activation.
Plans & deliveryBring a representative workload and the result your team needs. We’ll define the product, deployment and engineering scope around that task.
Activate this capability inside a qualified checked decisions workflow. Compatible methods and required shared checking are included once within that workflow; the product remains independently purchasable.
Capacity, release, environment, result contract and operating responsibilities are specified before activation. Choosing an additional method is not permission to add an unrelated workload or unqualified deployment.
Explore Checked Decisions Choose workflow capacityCommission the model, integration and acceptance work needed for your task. Existing products stay independently available.
Use the software directly or ask us to scope the implementation.
One capability. A connected workflow.
A team to put it into practice.