This reduces overhead, improves scalability, and enables more predictable parallel execution.

Flow’s architecture builds on these principles:
flow-computing.com/science

#DeepTech #Semiconductors #FlowComputing #AI #HPC #ParallelComputing #CPU #FlowPPU

Parallel execution is powerful, but coordinating threads efficiently is hard. Traditional models often introduce synchronization overhead, complexity, & unpredictable performance. TCF addresses this by executing groups of threads (fibers) in tightly coordinated steps w/ deterministic control.
#CPU

This research reinforces why rethinking #CPU architecture is critical for the future of #AI infrastructure. Learn more about the research behind our architecture:
flow-computing.com/science

Source: arxiv.org/pdf/2511.00739

#FlowPPU #Semiconductors #DeepTech #HPC #ParallelComputing

#AI often relies on agent workflows interacting w/ tools, memory & external systems, instead of single-pass inference. Orchestration/tool-side work can happen on CPUs. #FlowPPU addresses this. It's a gen. purpose parallel coprocessor w/ software & compiler, optimized to scale #CPU performance.

Diagram showing the Flow PPU (Parallel Processing Unit) as a 64-core general-purpose parallel co-processor integrated alongside a CPU on a chip. The visual highlights that it unlocks scalable parallel performance, is instruction-set independent, and works with existing CPU architectures. Additional text explains it is a licensable IP block with its own software and compiler, designed to scale CPU performance while remaining backward compatible.

CPUs play a big role in system performance. A recent paper analyzed representative agentic #AI workloads. Findings: #CPU tool processing accounts for (90.6%] of latency in some workloads, CPU-side tasks dominate runtime, & CPU dynamic energy accounts for (44%] of the total dynamic energy at large.

Flow’s architecture builds on these principles: eu1.hubs.ly/H0ttjwj0
#FlowPPU #CPU #ParallelComputing #HPC #AI #FlowComputing #Semiconductors #DeepTech

Shared memory is central to parallel computing, but difficult to scale efficiently. Traditional multicore systems rely on cache coherence, introducing overhead & limiting scalability. ESM addresses this, enabling high-throughput execution, latency hiding, & simplified parallel programming.

This makes it well suited for robotics and embedded systems, where low-latency and predictable performance are essential.
→ flow-computing.com/solutions/

#NationalRoboticsWeek #CPU #FlowPPU #HPC #Semiconductors #DeepTech #AI #FlowComputing #ParallelComputing

Black-and-white graphic featuring robotic arms in the background with a centered white paper-style overlay. Large text reads: “just a little reminder: scalable parallel execution drives real-time sensor & signal processing.” Smaller text around the edges references robotics systems, real-time control, parallel workloads, and notes that traditional multicore CPUs struggle to scale efficiently, while Flow enables scalable parallel execution for real-time data processing.

Robotics depends on fast, deterministic execution across parallel workloads. Sensor & signal processing workloads require efficient parallel execution to handle high-throughput data streams in real time. #FlowPPU enables nextgen performance, delivering scalable parallel execution w/ CPUs.
#CPU #HPC

Thank you to those we met for the insightful discussions, including Fuad Abazovic & Deepak Chugh, Hartti Suomela & @um.fi, & to #NordicInnovationHouse. We’re building momentum. See where we'll be next: eu1.hubs.ly/H0tfTQf0

#CPU #HPC #AI #VC #Semiconductor #DeepTech

Three men standing outdoors take a selfie in front of a residential home with a garden and trees, captured on a bright, sunny day.

Group of four men from Flow Computing standing outdoors in front of a Cadence office sign in Silicon Valley, smiling during a sunny day.

Our founders met w/ customers & investors in #SiliconValley. Signals:
→ Growing interest from both customers and investors
→ Highly engaged, technical conversations
→ Multiple discussions moving forward
It’s encouraging such traction w/ US-based investors focused on #semiconductor & compute tech.

#ParallelComputing #ComputerArchitecture #Semiconductors #AIInfrastructure #HighPerformanceComputing #CPU #DeepTech #SystemsEngineering #Innovation

Graphic titled “Modern processors face 3 major bottlenecks” listing: Memory access inefficiencies High synchronization overhead Poor scalability as core counts increase Flow Computing logo displayed at the bottom.

These limitations reflect deeper architectural challenges in modern parallel computing. They’re not easily solved w/ more cores / incremental improvements. #FlowPPU is designed from 1st principles to rethink how parallel workloads are executed. Explore science behind it: flow-computing.com/science

Graphic titled “Top early-stage deep tech startups in Europe” with Flow highlighted. Text reads “We’re on the DTM Watchlist!” and “Join DTM26, May 20–21, Berlin.” Deep Tech Momentum logo in the top right with a colorful cloud illustration background.

We've been selected from 5000+ startups for the #DeepTechMomentum Watchlist. We’ll be at #DTM26 in Berlin, meeting with investors, partners, & industry leaders across the #DeepTech ecosystem. If you’d like to connect, you request a meeting here: flow-computing.com/meet-timo-at...
#VC #CPU #AI #HPC

As system performance becomes increasingly important, improving parallel execution within the #CPU is key to unlocking overall system throughput.
Learn more: flow-computing.com

#Semiconductors #AI #ParallelComputing #DeepTech #FlowComputing #FlowPPU #HPC #AIInfrastructure #ChipDesign

Simple diagram titled “Newcomer’s guide to Flow PPU” showing: (1) the problem—CPUs struggle to efficiently utilize parallel workloads, (2) Flow PPU as a general-purpose parallel co-processor alongside the CPU, (3) how it works—CPU handles sequential tasks while PPU executes parallel workloads, and (4) why it matters—improved parallel execution increases overall system throughput.

New? Here’s a quick guide to what Flow PPU is & why it matters. Modern workloads are increasingly parallel. Efficiently utilizing that parallelism remains a challenge. Flow PPU is a general-purpose parallel co-processor designed to work alongside the CPU, enabling more efficient parallel execution.

🎧 Listen to the full episode below:
flow-computing.com/parallel-per...

#Semiconductors #AI #ParallelComputing #DeepTech #FlowComputing #FlowPPU #ComputeArchitecture #CPU #HPC #ChipDesign #Podcast #Torino #ItalianTech

Podcast graphic with three speakers discussing how microprocessor innovation has become incremental and why general-purpose parallelism is a key future direction.

🇺🇸On #DanteDay, we’re sharing an ep. of our podcast in Italian. We recorded this while staying near Via Dante. Our team discusses how #CPU architectures have reached a point of incremental improvements, limits of specialized accelerators, & why general-purpose parallelism is an important shift. #HPC

We’re focused on this challenge. #FlowPPU is a general-purpose parallel co-processor designed to work w/ the #CPU, enabling scalable, high-throughput parallel execution for modern workloads. Explore our technology: flow-computing.com/technology
Source: @cnbc.com
www.cnbc.com/2026/03/13/n...
#HPC

These systems rely on: orchestration across agents, data movement, &tool execution & control. System performance increasingly depends on how efficiently a #CPU can support & coordinate these workloads. Improving how parallel workloads are executed @ the CPU level is increasingly important. #AI #HPC

Minimalist graphic with the title “FLOW PPU” and key descriptors: “general-purpose parallel processing CPU co-processor” on the left, and “linear scaling,” “low synchronization,” and “instruction set independent” on the right, highlighting core architectural features of Flow PPU.

AI isn’t just about GPUs. A recent report highlights this. In a @cnbc.com interview w/ Dion Harris, #NVIDIA, he noted that CPUs are ‘becoming the bottleneck’ in AI & agentic workflows. As AI moves toward multi-step, agent-based systems, this reflects a broader change in how workloads are executed.

It’s exciting to see the ecosystem around efficient AI infrastructure continue to evolve. Learn more about our approach to parallel computing: flow-computing.com/solutions/
#EdgeAI #AI #ParallelComputing #Semiconductors #DeepTech #FlowComputing #FlowPPU #CPU #HPC

Minimal grey graphic featuring the statement “EDGE AI NEEDS PARALLELISM” centered inside a circular outline with Flow branding below. The visual emphasizes the importance of parallel computing for enabling efficient AI workloads on edge and consumer devices.

#AI moves from centralized training ➡️ deployment. Workloads shift to edge & consumer devices w/ strict constraints. Improving how CPUs execute parallel workloads is essential for a more capable AI. We were recently included in an Efficient & #EdgeAI Compute Market Map (parallelism category).

Black-and-white portrait of Flow Computing founders Dr. Martti Forsell, Timo Valtonen, and Jussi Roivainen standing together. The graphic headline reads “Meet the Founders of Flow” with text below noting their visit to Silicon Valley, California from March 19–27, 2026.

Our founders will be in #SiliconValley until 3-27-26. If you're interested in learning more about #FlowPPU / seeing a preview of our latest performance results, book a meeting here (meeting slots are limited): flow-computing.com/silicon-vall...
#Semiconductors #AI #DeepTech #HPC #CPU

We’re grateful to the SemiTO-V Student Team & the organizers for bringing us together to discuss the future of #HPC These conversations are important inshaping next-gen compute architectures. More events are coming soon: flow-computing.com/events/
#RISCV #CPU #Semiconductors #FlowPPU #DeepTech

Conference photo from World RISC-V Days in Turin featuring Flow Computing Head of CPU Development Marcello Ranone presenting Flow’s approach to scalable parallel performance.

Thank you to everyone who joined #RISC-V Days. It was a pleasure to see Marcello Ranone, presenting our approach to scalable parallel performance & engaging w/ the RISC-V community @ Politecnico di Torino. Thanks also to Andrea Coluccio, PhD for joining & capturing moments from the event.

Read the full article ↓
viewpoints.fov.ventures/p/europe-s-d...

#DeepTech #EuropeanTech #AI #Semiconductors #ParallelProcessing #Robotics #SpatialComputing #FlowComputing #FlowPPU #VentureCapital #VC

Minimal grey quote card featuring a statement from FOV Ventures’ “Europe’s Deep Tech Frontier” article describing Flow Computing as a research-driven company building a new chip architecture to accelerate parallel computing for real-time AI in physical systems. The graphic includes the FOV Ventures attribution and Flow branding.

Europe’s #DeepTech ecosystem is entering a new phase where advances in #computing, #robotics, & #AI infrastructure are converging to power real-world systems. We’re grateful to be featured in FOV’s Viewpoints article on Europe’s deep tech frontier. Thanks David Ripert & FOV team for the support.

Research in shared-memory systems, interconnection networks, and compiler techniques for explicit parallelism drive the #FlowPPU. The result is TCF, a programming and execution model that simplifies parallel software development. See the science behind Flow: flow-computing.com/science/

#CPU #HPC

We’re 2! Over the past 2 years, we’ve been developing #FlowPPU, a licensable co-processor designed to unlock efficient, general-purpose parallel execution across architectures. Thank you to our team, partners, & investors for being part of our journey.
#DeepTech #Semiconductors #Startup #HPC #CPU