Hardware that fits the problem
Hardware choices should start from the workload, model and compute resources the problem actually needs.
THE TECHNOLOGY WE BUILD
Connect compute to the demands of AI.
A research direction for running AI across different hardware, including infrastructure the business controls.
Explore how applications, models and compute resources connect through an AI inference runtime.
XPU RUNTIMEKXR / 02
AI APPLICATIONModel + Workload
KXR
KEVIQ XPU RUNTIMEInference runtime layer
CPU: a starting point for examining how a workload uses general-purpose compute.
Concept illustration · Select a component to explore
WHAT BUSINESSES CONSIDER
Choosing an AI foundation?
Different kinds of hardware
KXR is aimed at running AI across different hardware platforms, from CPUs and GPUs to accelerators.
Infrastructure you control
Also aimed at running AI on infrastructure the business controls, where the problem calls for it.
FROM TECHNOLOGY TO APPLICATION
The workload is where
the design starts.
KXR focuses on AI inference runtime across different hardware platforms. A relevant evaluation starts with the model, workload characteristics and compute resources involved.
Put simply, KXR is the layer between an AI application and the hardware, being researched to connect one application to different kinds of compute.
Application & model
Define inference requirements and workload characteristics.
Runtime layer
Explore the connection between applications, models and compute resources.
Resources & evaluation
Evaluate hardware choices under clearly specified conditions.
SCOPE & BOUNDARIES
Clear scope
before evaluation.
The Keviq ecosystem includes products, research projects and future directions. Not every item is commercially available; applicability is discussed problem by problem.
See all projects and the ecosystemSupport determined per environment
Support is determined for the specific model, hardware and evaluation environment.
Applied where it fits
KXR is considered where the workload fits; it is not the default for every problem.
HOW WE START
Four steps, starting
from the workload.
Describe the workload
The application, model and compute resources you are considering.
Agree a fitting scope
Agree what to evaluate first and the environment it will run in.
Evaluate for the workload
Runtime performance needs to be considered with the model, hardware and measurement conditions.
Decide on the evidence
How far KXR applies to a given problem is assessed against real evidence.
How we work together and the scope are agreed case by case.
EXPLORE BY INTEREST
How are you approaching KXR?
Businesses
You have an AI problem and want to weigh which hardware to run it on, and where.
Discuss one of your workflowsTechnology & research
Start with the workload, model and compute resources the problem calls for.
Discuss the technologyPartners & investors
Learn about the direction of KXR and how to get in touch about working together.
Partnership & investmentEXPLORE TOGETHER
What AI problem
are you working on?
Start with your context. Evaluation scope, integration possibilities and acceptance criteria are discussed for the specific problem.
Discuss one of your workflows