Overview
Flow's Parallel Processing Unit solves one of computing's most fundamental dilemmas - parallel processing.
The Parallel Processing Unit (PPU) is an IP block that integrates tightly with the CPU on the same silicon. It is designed to be highly configurable to specific meet requirements of numerous use cases.
Customization options include:
- Number of cores in PPU (4, 16, 64, 256, etc.)
- Type and number of functional units (such as ALUs, FPUs, MUs, GUs, NUs)
- Size of on-chip memory resources (caches, buffers, scratchpads)
- Instruction set modifications tailored to complement the CPU’s instruction set extensions
CPU modifications are minimal, involving the integration of the PPU interface into the instruction set and an update to the number of CPU cores in to unlock new levels of performance.
Our parametric design enables extensive customization, including the number of PPU cores, the variety and number of functional units, and the size of on-chip memory resources. Your performance enhancement scales with the number of PPU cores. A 16-core PPU is ideal for small devices like smart watches; a 64-core PPU is well-suited for smartphones and PCs; and a 256-core PPU is excels in high-demand environments such as AI, cloud, and edge computing servers.
Core benefits of Flow's technology
100x performance boost
- Flow's innovative Parallel Processing Unit (PPU) enhances CPU performance by up to 100 times, ushering in the era of SuperCPUs.
- Designed for full backward compatibility, the PPU boosts the performance of existing software and applications after recompilation. The more parallel the functionality, the greater the boost in performance.
- Flow's technology even enhances the entire computing ecosystem. While the CPU gains direct benefits, ancillary components—matrix units, vector units, NPUs, and GPUs—also see improved performance through the boosted CPU capabilities, all thanks to the PPU.
2X faster legacy software and applications
- Flow’s PPU enhances legacy code without altering the original application and greater performance gains when paired with recompiled operating system or programming system libraries.
- The result? Significant speed improvements across a wide range of applications, especially those that exhibit parallelism but are constrained by traditional thread-based processing. Our PPU unlocks the full potential of these applications, delivering significant performance gains where previous architectures fell short.
Parametric design
- The configurable, parametric design of the PPU allows it to adapt to diverse uses. Everything can be tailored to meet specific requirements of multiple use cases. And we do mean everything. The number of PPU cores—4, 16, 64, 256, or more— and the type and number of functional units, such as ALUs, FPUs, MUs, GUs, and NUs. Even the size of on-chip memory resources—caches, buffers, and scratchpads—can be tailored to suit specific requirements.
- The performance scalability is directly linked to the number of PPU cores. A PPU with 4 cores is ideal for small devices like smart watches, a 16-core PPU is perfect for smartphones, and a 64-core PPU delivers excellent performance for PCs. For servers, a PPU with 256 cores is recommended, enabling them to handle the most demanding computing tasks with ease.
Learn more about Vector Processor IP core
See how Codasip Studio enables SYCLOPS to implement a customizable RISC-V Vector processor, accelerating the future of open-source, high-performance AI compute at the Edge.
Real-time systems, particularly those used in domains like automated driving, are increasingly adopting neural networks. From this trend arises the need for high-performance hardware exhibiting predictable timing behavior. While state-of-the-art real-time hardware often suffers from limited memory and compute resources, modern AI accelerators typically lack the crucial predictability due to memory interference. The authors present a new hardware architecture to bridge this gap between performance and predictability.
Scalable IP Core of Vector Stream Cipher
As system requirements evolve toward multi-standard, reconfigurable platforms, signal processing architectures are under pressure to deliver both ASIC-class performance and software-like flexibility. Semiconductor engineers face a fundamental tradeoff: fixed logic yields, unmatched throughput, and efficiency, but cannot adapt once taped out. Software-programmable solutions offer flexibility but often miss hard real-time performance constraints and can consume more power.
Spiking Neural Networks (SNNs) are a promising, energy-efficient alternative to standard Artificial Neural Networks (ANNs) and are particularly well-suited to spatio-temporal tasks such as keyword spotting and video classification. However, SNNs have a much lower arithmetic intensity than ANNs and are therefore not well-matched to standard accelerators like GPUs and TPUs. Field Programmable Gate Arrays (FPGAs) are designed for such memory-bound workloads and here we develop a novel, fully-programmable RISC-V-based system-on-chip (FeNN-DMA), tailored to simulating SNNs on modern UltraScale+ FPGAs.
Much like the human visual system, embedded computer vision systems perform the same visual functions of analyzing and extracting information from video in a wide variety of products.
