AI Discussions

[JediJack]

Dear AMD Team,

I hope this message finds you well. I would like to share some thoughts on the current AI hardware landscape and offer suggestions for future development directions.

1. Nvidia's Dominance in the AI Hardware Market:
   As we all know, Nvidia currently dominates the AI hardware market, primarily due to its CUDA architecture. The vast support for CUDA within various AI libraries, both open-source and proprietary, has made it the default choice for AI development. This widespread adoption has created a significant barrier to entry for other hardware providers, making it difficult to surpass Nvidia in the AI space.

2. The Opportunity for AMD and Intel in Mid-to-Low-End AI Chips:
   While Nvidia focuses heavily on high-end GPUs, there is an emerging opportunity for AMD and Intel to concentrate on mid-to-low-end AI chips. There is a large market for affordable AI-powered PCs and smartphones. Developing chips optimized for these devices, while still maintaining sufficient AI performance, could create a significant market share. By offering efficient and cost-effective solutions, AMD and Intel could potentially gain traction in the AI personal computing market.

3. AI PCs and Smartphones – The Future of Personal AI Devices:
   In terms of AI PCs and smartphones, the key to widespread adoption lies in affordability without sacrificing AI functionality. These devices should not aim to replace AI servers but rather enhance the user experience by supporting offline AI capabilities in situations such as internet disconnection. Additionally, implementing SLM (Simple Language Models) tailored for device diagnostics and troubleshooting will be a huge differentiator. SLM can be used to guide users through basic troubleshooting steps in plain language, making AI more accessible to a broader audience.

I believe these suggestions could help AMD and Intel explore new avenues in the rapidly growing AI hardware market, especially as the demand for AI capabilities in personal devices continues to rise.

Thank you for your time and consideration.

Best regards,
Jack from Taiwan

[naturewalks]

---

@JediJack wrote:

While Nvidia focuses heavily on high-end GPUs, there is an emerging opportunity for AMD and Intel to concentrate on mid-to-low-end AI chips. There is a large market for affordable AI-powered PCs and smartphones. Developing chips optimized for these devices, while still maintaining sufficient AI performance, could create a significant market share. By offering efficient and cost-effective solutions, AMD and Intel could potentially gain traction in the AI personal computing market.

---

which is why AMD strengths are in their CPUs and GPUs

"AI powered" PCs do not exist. there is no code execution let alone logical reasoning within neural networks, it is just procedural generation over a set of neurons. with model powered AI you can create an abstracted memory of one state frame of an OS, but we can barely gen a few pages of consistent text.

so all those videos of AI minecraft are just as if we scanned your brain while you imagine what playing minecraft is like and projected it using an imagegen model trained on youtube videos.

there is no cost effective solution to this problem, which is why AMD should lean into CPUs and GPUs harder and let nvidia dig their own hole, however I think nvidia has industrial strength that is not to say it is a hole they are digging rather a side project they let the public maintain as this is all based on advances in python libraries anyway.

what will the AI run on?

specialized chips become e waste as everyone forgets about LLMs like the chatbot crazes of the past two decades, whereas x86 will always dominate the market, as I explained computer instructions cannot be executed by what we are all shouting about AI

besides, more VRAM for AMD GPUs means more future proofing for CUDA layers over wasting time on special LLM chips which serve no other purpose than to tell you good morning in different paragraph permutations

AI is a problem of throwing as much data at it you can to navigate the 3D latent space inbetween what you already have "learned" using the GPU, it's a software and energy issue not so much a hardware one

"AI powered" laptops will be a huge failure as it will add overhead and cost not necessary when a regex script will do exactly what you need