This company claims that it can multiply the performance of any CPU by 100

It sounds too good to be true: a new chip called PPU that multiplies the performance of the CPU by a hundred. They call it the new era of SuperCPUs.

The Finnish company Flow claims to have found the Holy Grail of computing. Basically, a miracle. Multiply the performance of any CPU by 100, regardless of the architecture. And incidentally, at a lower level, also the GPU, the NPU (artificial intelligence chip), etc. Too good to be true?

The magic is produced by a new type of chip called PPU (Parallel Processing Unit)although its creators they prefer to see it as a coprocessor that is attached to the CPU’s own silicon.

And that’s it. Magically, by installing this chip in any device, from a mobile phone to a PC, its performance is increased by up to 10,000%. How does it work?

The secrets of the PPU

The keys to this PPU are in your parallel data processingand in the large number of cores it uses.

It is completely scalable. For a smartwatch, a quad-core PPU is enough; for a mobile, 16 cores; for a PC, 64 cores; and for a supercomputer or cloud server, 256 cores.

Here you can see a representation of this PPU: a chip that attaches directly to one corner of the CPU, on the same silicon. So the connection speed with the processor is maximum:

As the Flow company explains on its website, this incredible increase in performance is due, on the one hand, to the limitations of current CPUs. And on the other hand, parallel data processing that the PPU provides.

This chip reduces latency with RAM, because it is capable of processing other data threads while accessing RAM. Instead, the CPU has to wait.

It also increases performance reducing synchronization between cores, which the PPU performs only once per step. Synchronization overlaps with code execution, there is no waiting in a queue, like the CPU does.

Finally, it overcomes the CPU’s limitation of executing instructions in parallelonly if they are independent, allowing to work at the same time with dependent instructions.

The only handicap of this technology is that to obtain maximum performance, the code of each application needs to be rewritten. Even so, without rewriting anything, performance doubles, which is already impressive. If certain parts are rewritten, it is multiplied by 10, and if the application is designed from scratch to take advantage of this new technology, it is multiplied by 100.

The theory doesn’t sound strange. It looks like an optimization of the parallel architecture of traditional CPUs. But it remains to be seen how these changes are applied without interfering with the operation of the CPU, and to see if that performance is real.

Flow plans to do a demo in August. Then we will see if This new PPU actually multiplies the performance of the CPU by 100and we entered the era of SuperCPUs.

Known how we work on Computertoday.

Tags: CPU, Processors