Google Coral TPU, is it still the best and most efficient object detection hardware available?
-
It took me month to get in hold of a TPU during pandemic, nearly 1 year. I finally got one and started using it for object detection in Blue Iris. I started to use Scrypted and playing with other NVR software and then I was informed Coral TPU is already considered old news, and I could possibly run better object detection in software or using intel quick sync. Is that true? What about comparing Coral TPU to an Nvidia GPU? Can it be similar or better than a standalone GPU? I'm considering to build another system with newer hardware to replace my ageing 10-year-old system, and I was wondering if I should go for another Coral TPU, or get a second hand GPU like Nvidia Tesla that several people recommend. Any thoughts please?
-
It took me month to get in hold of a TPU during pandemic, nearly 1 year. I finally got one and started using it for object detection in Blue Iris. I started to use Scrypted and playing with other NVR software and then I was informed Coral TPU is already considered old news, and I could possibly run better object detection in software or using intel quick sync. Is that true? What about comparing Coral TPU to an Nvidia GPU? Can it be similar or better than a standalone GPU? I'm considering to build another system with newer hardware to replace my ageing 10-year-old system, and I was wondering if I should go for another Coral TPU, or get a second hand GPU like Nvidia Tesla that several people recommend. Any thoughts please?
Myself I would consider why one needs the one that is most "efficient" right now? If you have a process to develop that requires this which works then presumably a better one would only make it faster? And if so then in a couple of years you would need to upgrade again. So perhaps better to get several older ones, because you save money, and then develop the process for each. Then you can design the process now to insure that it can support differences in the future. That is because you would have already been able to see what needed to be changed.
-
Myself I would consider why one needs the one that is most "efficient" right now? If you have a process to develop that requires this which works then presumably a better one would only make it faster? And if so then in a couple of years you would need to upgrade again. So perhaps better to get several older ones, because you save money, and then develop the process for each. Then you can design the process now to insure that it can support differences in the future. That is because you would have already been able to see what needed to be changed.