Speed is the
superpower of computing.
Flow enables next-gen CPUs.
[Parallel Processing Unit]
Hello, Flow PPU.
Progress in CPU performance has slowed dramatically over the past decades, even as the demand for greater processing power continues to rise, driven by AI workloads, edge computing, and cloud technologies.
Flow PPU breaks through this bottleneck. When coupled with a multicore CPU, it leverages the full power of parallelism to deliver substantial performance gains in general-purpose applications.
Built to be instruction-set independent, Flow PPU provides long-term architectural optionality as CPU ecosystems evolve The architecture supports integration with Arm, x86, RISC-V, and Open Power CPUs.
Fully backward compatible with existing code, Flow PPU can double the performance of many legacy applications after recompiling.
Flow PPU’s architecture exceeds the future demand for speed and efficiency, ushering in a new era of CPU performance and enabling the next generation of CPUs.
What is Flow PPU?
Parallel processing is a way to speed up the execution of computational problems by dividing them into subproblems that can be solved in parallel. In the best case, with P processors one can execute a problem on a data set P times larger, or process the original data set P times faster.
The ideal speedup with current parallel systems built from multicore processors is, however, difficult to achieve, especially when non-trivial memory access patterns or synchronization between parallel software components are required.
To address these problems, Flow Computing® is building the Parallel Processing Unit® (Flow PPU). It is a tightly integrated on-chip IP block (or chiplet) that works alongside the CPU on the same silicon (or package). It uses a novel architecture designed specifically for shared-memory parallel computation to significantly boost performance by eliminating the overheads of current multicore CPUs in non-trivial memory access patterns and the synchronization of parallel software components.
Using Flow PPU to accelerate execution is simple. Assign the parallel parts of the code explicitly to Flow PPU, or let our compiler handle it, while the sequential and joint parts continue to run on the CPU.
The parameters of Flow PPU are design-time adjustable to meet the specific requirements of a wide range of use cases.
Performance results that justify a closer look.
Flow PPU delivers scalable performance beyond the limits of multicore CPUs. From compute-heavy tasks to high-throughput workloads, our benchmarks show consistent, significant improvements in throughput, synchronization, and latency.
Where Flow PPU makes an impact.
Flow PPU delivers scalable performance on workloads where conventional multicore CPUs stop scaling, from hyperscale infrastructure to autonomous and embedded systems.
It provides high-throughput execution for demanding applications, like decentralized AI, edge AI inference, real-time signal/sensor data crunching that need high-performance general-purpose parallel execution. Flow PPU integrates into any current or future CPU architecture or process geometry, delivering immediate acceleration without the need for GPU offloading of CPU workloads.
These performance gains are enabled by our patented PPU architecture and its compiler ecosystem, which together enable efficient, high-throughput execution.
Extending beyond single devices, our architecture also scales across data centers, supercomputers and edge environments, improving compute efficiency across AI clouds, multimedia workloads, autonomous platforms, and mission-critical systems, setting a new benchmark for the future of CPU performance.
Server & Cloud AI CPUs
Scalable throughput for cloud, HPC, supercomputers, data centers and centralized AI workloads, accelerating preprocessing, inference, scientific compute, and large-scale services without GPU overhead. When most of the AI latency is serialization, a faster GPU won't help. A faster CPU will.
Industrial & Embedded AI CPUs
Deterministic, low-latency performance for decentralized AI, robotics, signal/sensor data processing, autonomous systems, defense, and real-time industrial compute, while accelerating scientific workloads and enterprise compute and simplifying parallel code.
Consumer & Edge AI CPUs
High-performance, energy-efficient acceleration for mobile AI, edge AI inference, and next-generation consumer devices like smartphones, laptops and even smartwatches.
The slowdown of Moore’s law.
All CPU markets (server industrial, and consumer) have benefited from decades of silicon technology-driven performance scaling, but the slowdown of Moore’s law and the rise of data-heavy AI workloads have exposed the limits of multicore CPUs. GPUs can help in well-behaving cases, but they significantly increase cost, power consumption, and data-movement overhead.
As CPU core counts grow, multi-core CPUs fail to scale efficiently, making Flow PPU a necessary way to restore parallel performance directly inside the CPU pipeline, at much lower hardware and programming costs than GPU-based solutions.
Science
Efficient realization of general-purpose parallel computing has long been considered one of the hardest problems of computer science. Read more on our Science page about our approach to these problems, building on our involvement in more than 30 years of scientific research into parallel computing, memory systems, and processor design.
Highlights from press coverage
08/07/25
Flow Computing crowned as the KPMG Tech Innovator in Finland 2025
We are thrilled to announce that Flow Computing has been named the winner of the KPMG Tech Innovator in Finland 2025 competition!
Read more14/05/25
Flow Reaches Milestone: PPU Achieves End-to-End CPU Operations in Alpha Testing
Flow Computing hits a major milestone on its path to commercializing Flow Parallel Processing Unit (PPU). With its compiler now in Alpha, the company demonstrates significant end-to-end performance gains, pushing CPU innovation into a new era.
Read more10/03/25
Flow Computing Named One of Finland’s Most Promising Startups by Talouselämä
Flow Computing is honored to be named one of Finland’s "10 Most Interesting Startups" of 2025 by Talouselämä! This recognition highlights our Flow PPU technology, which boosts CPU performance up to 100x. Read more about how we’re driving innovation in high-performance computing
Read more23/09/24
Startup Says It Can Make a 100x Faster CPU Flow Computing aims to boost central processing units with their ‘parallel processing units’
In an era of fast-evolving AI accelerators, general purpose CPUs don’t get a lot of love. “If you look at the CPU generation by generation, you see incremental improvements,” says Finland-based Flow Computing.
Read more20/03/25
Flow Computing Featured in Tech Funding News: How International Teams Drive Innovation in Finland
Flow Computing is shaping the future of high-performance computing. Featured in Tech Funding News, we discuss how international teams secure 60% of Finland’s VC funding and why our PPU technology is gaining investor attention. Read more!
Read more13/08/24
CPU Performance Bottlenecks Limit Parallel Processing Speedups
Semiconductor Engineering's Bryon Moyer analyses how hardware optimizations and well-thought-out software architectures can help CPU performance bottlenecks that are currently limiting potential parallel processing speedups.
Read more