What is the performance of Particle Photon in FLOPS?

I am running computer vision code on a Raspberry Pi on images and considering moving it to a Particle Photon. A Raspberry Pi can do 3.1 GFLOPS (3 billion Floating Operations Per Second). I could not find the performance of the STM32 processor on a Photon.

What is the performance in FLOPS of the particle photon?

GFLOPS would be Giga Floating Point Operations Per Second - that’d be billions :wink:

The core clock of an RPi is > 1GHz while the Photon runs at 120 MHz and I’m not convinced it even features an FPU.

2 Likes

Would running matrix algebra such as Sobel derivatives and algorithms such as edge detection be impossible on the Particle Photon?

Without HW floating point calculations it would all need to be done in software.

Comparing an Photon to an RPi 3B+ with 1.2GHz and four cores and a dedicated FPU you’d be looking at a factor of considerably less than 1:40

2 Likes

The STM32F205 in the Photon does not have hardware floating point and would not be a good choice for floating point calculations.

1 Like

Though he platform has no dedicated FPU, you can of course run floating point calculations just fine albeit slower than with FPU. So you “can” but you may not “want to” given your performance expectations (which you did not mention). Knowing a platform’s GFLOPS does little to understand how performance of your calculations will be - your specific algorithm may be more impactful than raw computing power. Lastly, with matrix calculations, memory may be more a limitation depending on how large your matrices are.

If the photon’s speed is not what you need, look for a better algorithm first then for a better platform. This is a ARM cortex m3, next one up would be a cortex m4 but most of those do not have an FPU either (there are a few, if memory serves from NXP). At some point you might be lured into doing hand assembly optimization - i’d stay clear from that as long as you can.

2 Likes