Performance optimization for modern compute

Demand morefrom your silicon.

The optimization platform, to run any model, on any hardware, at full speed.

Getting Started$pip install docc-compiler

Applications

Get more from every workload
you run.

Qwen

> Explain loop tiling in one sentence.

TOKENS

Get more speed from LLMs

Qwen compiled from PyTorch with optimized attention and KV-cache. Tuned for modern GPUs.

Tuned for modern GPUs

ROCm

CUDA

PyTorchTransformersGitHub

FRAMES

Get higher frame rates

Real-time object detection compiled from PyTorch into native code. One model, optimized for every chip.

Optimized for

ROCm

CUDA

PyTorchOpenCVGitHub

EXPERIMENTS

Get more science from fewer bits

Computational fluid dynamics on modern RISC-V-based TPUs. C++ source compiled directly with lower precision hardware.

Accelerator

Metalium

C++OpenMPRead more →GitHub

DETECTIONS

Get more efficiency, less entropy

Computing with light. Faster R-CNN with ResNet-50 backbone compiled directly from PyTorch to photonic hardware.

Photonic processor

NPU Gen 2

PyTorchtorchvisionRead more →GitHub

How it worksWe reinvented how performance engineering works, scalable for everyone.→

Core benefits

Get more performance and freedom for the applications you have already written.

Automatic

No rewrites, no custom kernels, no new language.

Products

Get more from the hardware
you already own.

Ready to use

Daisytuner Platform

Fast code on every chip, no kernels to write. Compile, benchmark and optimize in one place, ready when you sign in.

What you get

DOCC compilerMake your code run on every chip.
Performance DatabaseState-of-the-art code optimizations for popular algorithms.
Continuous BenchmarkingMeasure performance in CI on our hardware.
Web IDEDesign optimizations for your own database with advanced performance analyses.

Get started Learn more

Your infrastructure

Enterprise

Daisytuner Enterprise

Enterprise-grade performance for your stack. Every optimization is built by experts, and our engineers stand behind it.

What you getEverything in Platform

Auto-tuningYour custom optimizations, fully automated.
Self-hostedRuns on your own servers. Data stays inside.
SLA and supportGuaranteed response times and long-term support.

Learn more

Inference stack

Daisyflow

High-performance AI inference stack for your enterprise. Built on top of the Daisytuner Platform to run AI models at the best cost efficiency, on any hardware.

What you getEverything in Enterprise

Inference stackServing, routing and batching for any model.
SetupInstalled on your servers, on any accelerator.
OperationsWe run and tune it with you, end to end.

Learn more

AI's next leap won't be bought, it will be unlocked.

The world's answer to the hunger for compute is to build more. But the machines already in place run at a fraction of what they can. Daisytuner unlocks the rest, on any chip you choose.

Talk to an Engineer Getting started

Demand morefrom your silicon.