Flow Computing Promises a 100-Fold Performance Gain by Adding a "PPU" to a CPU

Finnish startup Flow Computing has emerged from stealth with a bold claim: a coprocessor that can be added to a CPU to deliver "up to 100-fold" acceleration for highly-parallelizable workloads — without modifying existing software, and without the high power demands of traditional GPUs and accelerators.

"We firmly believe there have been only incremental improvements in CPU performance during the last few decades — in our opinion, this has led to a situation where the CPU has actually become the weakest link in computing due to its sub-optimal sequential architecture," claims co-founder and chief executive officer Timo Valtonen of the company's work.

"A new era in CPU performance has become a necessity in order to meet the continuously increasing demand for more computing performance, driven to large extent by needs in AI, as well as edge and cloud computing. Flow intends to lead the SuperCPU revolution through its radical new Parallel Performance Unit (PPU) architecture, enabling up to 100x the performance of any CPU, regardless of architecture and with full backwards software compatibility."

That's a bold claim, and one the company repeats: performance gains "up to 100-fold" through the integration of its PPU technology into a CPU design, which is "fully backwards compatible with every existing software application for that CPU architecture" requiring nothing more than "recompilation for the PPU without any software changes required."

Documentation published by the company, however, tempers Valtonen's claims: with no modification to the software at all, a system combining a traditional CPU with a PPU will see no speed-up at all. Only by recompiling the software from source with a PPU-aware compiler will deliver any performance benefits — and "only" a doubling, far from the promised 100-fold boost. With "small refactoring" on the user's part, to explicitly offload certain bottlenecks to the PPU, this can be increased to a tenfold performance gain — and the headline-grabbing two-orders-of-magnitude promise is only reachable, the company admits, by rewriting software to specifically target the PPU.

A spin-off from the VTT Technical Research Center of Finland, Flow Computing claims it can deliver such a performance increase through a design, which is specifically tailored for highly-parallel workloads and avoids common pitfalls to rival accelerator approaches including cache coherence issues, efficient flow synchronization, "unique and specific techniques/solutions" for concurrent memory access, a flexible threading scheme, and a "fully superpipelined design."

What the company has not yet done, however, is proven its claims — and, at the time of writing, had not found any customers for its technology. "As we just came out of stealth mode," Flow Computing admits in its briefing paper, "we are now looking to engage with the leading-edge CPU companies such as AMD, Apple, Arm, Intel, NVIDIA, and Qualcomm to co-develop the future era of advanced CPU computing."

More information on the PPU technology is available on Flow Computing's website; no date has been announced for the first "SuperCPU" to launch.