I am trying to decide on getting a pi ai hat vs an nvidia grahics card for a robotics project.
if i go for graphics card then inference will run on pc that pi speaks to over lan.
the compexity of running custom models on ai hat puts me off doing it on the pi.
also my current robot has a pi 4 on it and it would be a hassle to change it. or may even require a rebuild if the 'lego build hat' i use isnt comaptible.
i am trying to compare performance and i cant find any comparisons
a lot is made about the number of tops for the ai hat but with graphics cards tops is rarely mentioned when i can find somthing about tops the ai hat seems much higher
I will probably use the cloud for training and am looking only for a way of doing inference. mainly i will be doing computer vision. e.g. yolo
acording to this page on nvidia website
https://www.nvidia.com/en-gb/geforce/ne ... processor/
geforce gtx 1660 Ti has 11tops
according to this website
https://irendering.net/what-are-teraflo ... ight%20now.
i realise it is just some website i have no idea of the authority of the source the nvidia 2080 ti has 13.45 teraflops. i cant find anything quoting tops but for the 1080 the page above quotes twice as many tflops as tops so presumably its the same with the 2080ti
there are also a bunch of other cards quoted at the 2nd site above all of them with less than half the number of tflops as the pi has tops
can someone help me to do a meaningfull comparison?
is there something i am missing about what tops mean on the ai hat and a nvidia graphics card?
if i go for graphics card then inference will run on pc that pi speaks to over lan.
the compexity of running custom models on ai hat puts me off doing it on the pi.
also my current robot has a pi 4 on it and it would be a hassle to change it. or may even require a rebuild if the 'lego build hat' i use isnt comaptible.
i am trying to compare performance and i cant find any comparisons
a lot is made about the number of tops for the ai hat but with graphics cards tops is rarely mentioned when i can find somthing about tops the ai hat seems much higher
I will probably use the cloud for training and am looking only for a way of doing inference. mainly i will be doing computer vision. e.g. yolo
acording to this page on nvidia website
https://www.nvidia.com/en-gb/geforce/ne ... processor/
geforce gtx 1660 Ti has 11tops
according to this website
https://irendering.net/what-are-teraflo ... ight%20now.
i realise it is just some website i have no idea of the authority of the source the nvidia 2080 ti has 13.45 teraflops. i cant find anything quoting tops but for the 1080 the page above quotes twice as many tflops as tops so presumably its the same with the 2080ti
there are also a bunch of other cards quoted at the 2nd site above all of them with less than half the number of tflops as the pi has tops
can someone help me to do a meaningfull comparison?
is there something i am missing about what tops mean on the ai hat and a nvidia graphics card?
Statistics: Posted by gregorianranter — Tue Jan 28, 2025 4:44 pm — Replies 2 — Views 36