Drawbacks of AMD GPU technology?

[CatTNT]

I'm interested in upgrading to a new 40 series or AMD 7000 series GPU. I have an Nvidia GPU currently. My main concern about going for the AMD GPU (7900XTX) is that is doesn't support CUDA, and some other technologies that Nvidia does. 

 

Some thing I may want to use in the future, that AMD may not have the tech to allow:

I want to get experience with using Unreal Engine 5, is there anything AMD doesn't support that Nvidia does, and noticeably outperforms them in? 

 

How is blender  performance with an AMD GPU? I don't do anything super complex, just editing some models for 3d printing occasionally, no rendering animations or anything like that.

 

How does AMD do machine learning and AI? I've screwed around with stable diffusion recently, and I am not currently using or learning anything with AI or ML, but I don't want to lock myself out of that in the future. I heard that CUDA is the only real ML/AI platform that's being used, I might be wrong. 

 

 

 

 

[Shimejii]

With what you are doing, especially blender, just stick with NVidia. NVidia is the one dumping billions into AI and development, pretty much making most of those programs backbone based into CUDA. AMD simply doesnt have that same amount of money  to purely throw at devs, so their products tend to suffer because of it.

[igormp]

8 minutes ago, CatTNT said:

I want to get experience with using Unreal Engine 5, is there anything AMD doesn't support that Nvidia does, and noticeably outperforms them in? 

 

No, both will work fine for that.

 

8 minutes ago, CatTNT said:

How is blender  performance with an AMD GPU? I don't do anything super complex, just editing some models for 3d printing occasionally, no rendering animations or anything like that.

It works but nvidia is faster by a substancial margin.

 

8 minutes ago, CatTNT said:

How does AMD do machine learning and AI?

It doesn't, at all.

8 minutes ago, CatTNT said:

I heard that CUDA is the only real ML/AI platform that's being used, I might be wrong. 

You are not, most of the ML stuff is locked to CUDA indeed.