Beyond Nvidia: Daisytuner Secures €6M to Optimise AI Across Hardware Platforms

Avatar photo

The artificial intelligence industry is facing a new challenge as the demand for computing power grows faster than the ability of traditional hardware systems to keep up. While companies are investing heavily in advanced processors, getting AI models to perform efficiently across different chips remains a complicated engineering task. German deeptech startup Daisytuner wants to simplify this process through artificial intelligence powered software that automatically optimises computing performance. The company has raised €6 million in seed funding to expand its technology and bring its platform to more enterprise customers.

The funding round was co led by Elaia and LEA Ventures, with participation from Amadeus APEX Technology Fund and Seedcamp. The investment will support the expansion of Daisytuner’s enterprise business, development of its compiler technology and recruitment of additional engineering talent.

Solving the AI Hardware Efficiency Problem

Founded in 2024 by Lukas Trümper and Adrian Schmitz, Daisytuner develops technology that helps AI models operate efficiently across different processor architectures.

Most AI applications require specialised computing hardware, particularly graphics processing units (GPUs). However, different chip manufacturers use different architectures and software systems. As a result, developers often need to rewrite or optimise their applications for each hardware platform.

This process requires highly specialised engineering expertise and can delay the adoption of new processors. It can also leave expensive computing infrastructure operating below its full potential.

Daisytuner aims to automate this performance engineering process, allowing companies to use different hardware without repeatedly rebuilding their software for every new chip.

Introducing Daisyflow and AI Powered Compilation

The company’s technology combines an open source compiler called DOCC with its inference platform, Daisyflow.

The compiler translates software code into versions optimised for the target processor. It is designed to support different hardware architectures, including Nvidia and AMD GPUs, alongside emerging chip technologies.

Artificial intelligence agents continuously test potential optimisations on real hardware. The compiler checks whether these changes are correct before they are used, while verified improvements are stored in a performance database for reuse across similar workloads.

Daisyflow brings these capabilities into production environments. It enables companies to run language, vision, speech and recommendation models using familiar application interfaces.

An important feature is that businesses can deploy the platform within their own data centres or private infrastructure. This gives organisations greater control over their data, computing resources and operating costs instead of depending entirely on external AI cloud providers.

Benchmark Shows Faster AI Inference

Daisytuner has reported promising early performance results from its technology.

In a company benchmark, Daisyflow ran Alibaba’s Qwen 2.5 artificial intelligence model up to 18 per cent faster than vLLM, a widely used open source inference engine, on AMD Instinct GPUs.

The result highlights the potential of software optimisation to improve the performance of existing hardware without necessarily requiring companies to purchase more powerful processors.

However, performance improvements can vary depending on the model, hardware configuration and workload. The reported benchmark represents the company’s own testing rather than an independently verified industry wide result.

A Research Driven Engineering Team

Daisytuner has built its technology around expertise in computer science, compiler engineering and high performance computing.

Its team includes researchers and engineers associated with institutions such as ETH Zurich, the Technical University of Munich, RWTH Aachen University, TU Darmstadt and Lawrence Livermore National Laboratory.

The company’s founders are focused on making advanced hardware optimisation more accessible to enterprises that cannot maintain large specialist engineering teams.

Funding to Accelerate Enterprise Adoption

The newly raised €6 million will help Daisytuner bring Daisyflow to more enterprise customers, strengthen its engineering team and expand the capabilities of its compiler.

As businesses explore alternatives to traditional AI hardware providers, software that allows models to run efficiently across different processors could become increasingly important.

Daisytuner is positioning itself at the intersection of AI infrastructure, chip optimisation and enterprise computing, with the ambition of helping companies extract more performance from every processor they use.

Total
0
Shares
Previous Post

From Dirt Tracks to Unicorn Status: Stark Future Raises €150M

Next Post

Building the Future of Defence: Singapore Hackathon Brings 300 Innovators Together

Related Posts