ButSpeak.com
News which Matters.
Finnish startup Flow Computing has unveiled a revolutionary chip that can potentially double CPU performance and increase it up to 100 times with further software optimization. This breakthrough could significantly impact AI technologies and autonomous vehicle systems, providing the computational power needed for these advanced applications.
Originating from Finland’s VTT Technical Research Centre, Flow Computing introduces the Parallel Processing Unit (PPU). Despite skepticism about these bold claims, Flow’s co-founder and CEO, Timo Valtonen, remains confident.
The PPU is a companion chip designed to optimize processing tasks in real-time, transforming traditional serial processing into more efficient parallel operations without additional power consumption or excessive heat. This technology is likened to expanding a CPU from a one-lane road to a multi-lane highway, dramatically enhancing efficiency and processing speed.
Flow’s PPU enhances CPU functionality by managing tasks at nanosecond intervals, allowing multiple processes to occur simultaneously. This innovation increases throughput without altering the CPU’s clock speed or architecture. However, the success of this technology hinges on its adoption by chip manufacturers, requiring integration at the design stage which could disrupt current production methods. Despite these challenges, the minimal modifications needed and the potential for substantial performance improvements make it an attractive option for chipmakers facing growing computational demands.
Flow Computing has demonstrated the effectiveness of its technology in FPGA-based tests. With €4 million in initial funding and backing from several venture capital firms, the startup is now seeking industry partnerships to further develop and scale its solution.
At the heart of Flow’s groundbreaking technology is its patented Parallel Processing Unit (PPU) architecture and versatile compiler ecosystem. The PPU can seamlessly integrate into any design architecture or process geometry, providing immediate performance boosts for future CPUs and eliminating the need for costly GPU acceleration.
Flow’s technology significantly enhances not only individual devices but also embedded systems and data centers. Its applications span edge and cloud computing, AI clouds, multimedia processing over 5G/6G networks, autonomous vehicles, and military-grade computing, setting new standards for CPU capabilities.
Core Benefits of Flow’s Technology
- 100x Performance Boost: Flow’s PPU architecture amplifies CPU performance by 100 times, introducing an era of SuperCPUs. This enhancement applies to existing software and applications, with parallelism yielding greater performance gains. Notably, matrix units, vector units, NPUs, and GPUs also benefit from the PPU’s enhanced CPU capabilities.
- 2x Faster Legacy Software and Applications: The PPU improves legacy code without modifications and significantly boosts performance when paired with recompiled operating systems or programming libraries. This results in substantial speed improvements, especially for applications constrained by traditional thread-based processing.
- Parametric Design: The PPU’s configurable and parametric design allows it to adapt to multiple uses, from smartwatches to high-demand environments like AI and cloud servers. Users can tailor the number of PPU cores (4 to 256 or more), the type and number of functional units (ALUs, FPUs, MUs, GUs, NUs), and the size of on-chip memory resources (caches, buffers, scratchpads) to meet specific requirements.
What is the Parallel Processing Unit?
The PPU is an IP block that integrates tightly with the CPU, designed for high configurability across various use cases. Customization options include the number of cores, functional units, and on-chip memory resources. Minimal CPU modifications are required, leveraging substantial performance gains through the PPU interface integration with the CPU’s instruction set.
How is the 100x Boost Possible?
Flow’s PPU addresses critical challenges around CPU latency, synchronization, and virtual level parallelism:
- Latency Hiding: Unlike traditional multicore CPUs suffering from memory access delays and coherence issues, Flow’s PPU hides latency by executing other threads during memory access, eliminating coherence problems and enhancing scalability via a high-bandwidth network-on-chip.
- Synchronization: Flow’s PPU reduces synchronization costs significantly. Traditional CPUs require expensive synchronizations for inter-thread dependencies, while the PPU synchronizes threads once per step, drastically lowering costs and improving efficiency.
- Virtual ILP/LLP: Flow’s PPU optimizes instruction execution by organizing functional units in a chain, allowing dependent code execution within a step and eliminating pipeline hazards, thus enhancing parallelism and performance.
Performance Boost for Existing Software & Applications
Flow’s technology is fully backward compatible, enhancing all existing legacy software and applications. The PPU’s compiler identifies and executes parallel code segments in PPU cores, and Flow is developing an AI tool to help developers optimize code for maximum performance.
Why Better CPU Performance is Vital for Future Industries
Investments in compute accelerators have transformed customer capabilities, yet general-purpose compute remains critical. Enhancing CPU performance is essential to maximize performance, reduce costs, and meet sustainability goals. Without improvements in general-purpose compute, infrastructure capabilities will be limited despite advances in accelerators.
