8GB RAM speed up Convolutional Neural Network?

[nick8967]

I have built a CNN on tensorflow and I run it on Rpi 4B 4GB. My question is if the validation on 4GB Rpi takes x minutes, can a 8GB RAM Rpi reduce that in y minutes (y<<x) ? Does anyone has an idea? experience on that?

[RGJ-PI]

I've re titled thread to more clearly describe question

[DougieLawson]

I've re titled thread to more clearly describe question

Good idea because I thought it was Trump's least favourite TV station.

https://www.tensorflow.org/lite/perform ... timization

[B.Goode]

I have built a CNN on tensorflow and I run it on Rpi 4B 4GB. My question is if the validation on 4GB Rpi takes x minutes, can a 8GB RAM Rpi reduce that in y minutes (y<<x) ? Does anyone has an idea? experience on that?

{Disclaimer: no experience.}

Since it is unlikely that anyone will have built and run your particular implementation of a Neural Network, it seems that ideas are all we have to contribute.


My idea is simply to question what resource is constraining the performance of your application? In broad categories: processing power; i/o throughput; memory (RAM)?.

Only if your code is memory-limited, and perhaps swapping heavily, will a hardware upgrade help. AIUI there is no difference in processing power or i/o capability between members of the RPi4B product range.