Support NeoGAF

[MrTroubleMaker]

will these go on fire as well?

 

[Von Hugh]

will these go on fire as well?

On fire or on fire sale? Yes to both.

 

[Danknugz]

having caught the tail end of the lga1700 with my 13900kf puts me into the position of a whole motherboard upgrade if i were interested, doesn't seem worth it but rest to pull the trigger on the 5090.

 

[Codeblew]

Well, NPUs are used to accelerate Neural Network processing. It have several use cases, and one of them is the Generative IA and voice recognition (the gimmick you're talking about). I think that, once it becomes more popular, it could be broadly used in gaming, but now, it isn’t. It could be used in upscaling, accelerate physics calculations and so on.

So basically what a GPU can already do.

 

[Danknugz]

So basically what a GPU can already do.

what GPUs are we talking about? if you're running AMD then not really. AMDs don't support CUDA which you need for pytorch and tensor flow, and training AI models. they can handle inference which is not the same as training your own models, albeit much slower. on that same note i doubt this NPU stuff with a handful of CPU cores could perform as well a CUDA supporting GPU.

 

[Codeblew]

what GPUs are we talking about? if you're running AMD then not really. AMDs don't support CUDA which you need for pytorch and tensor flow, and training AI models. they can handle inference which is not the same as training your own models, albeit much slower. on that same note i doubt this NPU stuff with a handful of CPU cores could perform as well a CUDA supporting GPU.

Yeah, AMD GPU's are behind but they are working on catching up. ROCm is promising. The only thing I would really use Inference for is LLM's for coding (currently pay for GPT so no need to run locally) or Leela Chess Zero. I have ran some of the open source LLM's locally on my 6700XT but I am limited to the smaller ones due to memory limitations on the card.