A Flow Computing Parallel Processing Unit can work with current software and dramatically boost performance.
Finnish startup Flow Computing has emerged from stealth mode with a groundbreaking claim: their new coprocessor can be added to a CPU to achieve “up to 100-fold” acceleration for highly parallelizable workloads without requiring any modifications to existing software and without the high power demands typical of traditional GPUs and accelerators. Co-founder and CEO Timo Valtonen states, “We firmly believe there have been only incremental improvements in CPU performance over the last few decades. This has led to a situation where the CPU has become the weakest link in computing due to its sub-optimal sequential architecture.”
The company’s claims are bold, emphasizing “up to 100-fold” performance gains through the integration of PPU technology into a CPU design. They assert that this technology is “fully backward compatible with every existing software application for that CPU architecture,” requiring merely a “recompilation for the PPU without any software changes needed.”
Without any software modifications, a system combining a traditional CPU with a PPU will see no speed-up. Recompiling software with a PPU-aware compiler can double performance, while “small refactoring” to offload certain bottlenecks to the PPU can yield a tenfold increase. The promised 100-fold boost is achievable only by rewriting software to specifically target the PPU. The company asserts that their performance increase is made possible through a design tailored for highly parallel workloads. This design avoids common pitfalls in rival accelerator approaches, such as cache coherence issues and inefficient flow synchronization, utilizing unique techniques for concurrent memory access, a flexible threading scheme, and a fully superpipelined design.
Despite these claims, the company has yet to prove its technology and, at the time of writing, has not secured any customers. “As we have just emerged from stealth mode,” the company states, “we are now looking to engage with leading-edge CPU companies like AMD, Apple, Arm, Intel, NVIDIA, and Qualcomm to co-develop the future of advanced CPU computing.”
Valtonen continues, “A new era in CPU performance is necessary to meet the ever-increasing demand for more computing power, driven largely by needs in AI, as well as edge and cloud computing. Flow aims to lead the SuperCPU revolution with its innovative Parallel Performance Unit (PPU) architecture, promising up to 100x the performance of any CPU, regardless of architecture, and maintaining full backward compatibility with existing software.”
