Kalray DPU: Purpose-built Acceleration For Modern Workloads.

Purpose-built acceleration across AI, Networking, Storage, and Security.

Today’s Compute Challenges

When CPUs And GPUs Aren’t Enough

Modern workloads are growing exponentially — larger AI models, connected devices, and real-time data demand far more from compute systems than ever before. Yet traditional architectures can’t keep up. CPUs struggle to scale, while GPUs deliver power at the cost of efficiency and flexibility.

As performance needs outpace general-purpose processors, the industry faces new bottlenecks in scalability, energy use, and adaptability. The future requires specialized architectures — designed to accelerate what today’s systems can’t.

Power utilization constraints

Energy limits cap AI scale

Support complex workloads

AI workloads strain system architectures

Massive compute requirements

AI demands extreme compute density

Kalray’s DPU architecture bridges the gap

The Need for Data Processing Units

AI trends make it clear: the market needs a new class of compute architecture. One that is not general-purpose, but purpose-built to address the efficiency, scalability, and coordination demands of modern workloads.

Data Processing Units, domain-focused accelerators, are architected not to replace CPUs or GPUs, but to complement them. They are designed to process data flows efficiently, remove infrastructure bottlenecks, and deliver higher performance per watt. By focusing on system-level efficiency rather than generic compute, DPUs enable organizations to scale AI and data-intensive workloads without relying solely on increasingly power-hungry processors.

DPU Advantages

Greater power efficiency

Delivering higher performance per watt by offloading infrastructure workloads from CPUs and optimizing GPU utilization in AI systems.

Compute Optimization

Offload data-centric processing from CPUs and feed GPUs efficiently, removing bottlenecks and improving overall system performance.

Complex Workload Management

Purpose-built for data-intensive workloads, orchestrating AI, networking, storage, and security functions within a balanced system architecture.

Software Programmability

Open, developer-friendly tools enable flexible optimization and deployment without proprietary lock-in or rigid frameworks.

Kalray’s MPPA: The architecture behind smarter DPU acceleration.

Kalray’s response to this change is its Massively Parallel Processing Architecture (MPPA) - a scalable, modular architecture designed to power the next generation of customized DPU accelerators.

Customizable

Purpose-built for customization, allowing domain-specific optimizations without starting from scratch.

Modular

Allowing different compute, memory, and I/O blocks to be combined and optimized for each use case.

Programmable

Enabling developers to tune and adapt acceleration behavior through standard software environments.

Scalable

Supporting deployments from low-power edge devices to high-throughput data center systems.

Kalray is recognized by industry analysts and certified by leading quality and interoperability organizations.

Analyst Recognition

GigaOm recognized Kalray as a Leader and Fast Mover in the 2023 Data Processing Units (DPU) Sonar Report — highlighting our innovative architecture and strong performance across AI and data workloads.

Quality Certifications

Kalray’s ISO 9001 certification and interoperability validations demonstrate our commitment to high-quality standards and sustainable performance in every solution we build.