I have a few questions about the future landscape of AI hardware and architectures:
- Can GPU architectures eventually rival the speeds of Groq and similar specialized hardware? Is there a foreseeable future where GPU inference with hardware sub-systems (e.g. TPU) and/or algorithmic enhancements can match or surpass these speeds?
- Is there a post-Transformer AI world? Are there any new architectures or enhancements on the horizon? Which of these could continue the GPU dominance?
- Are approaches like Groq easy to optimize for future AI models?
Insights and thoughts appreciated!
0 comments
—
Ask HN: Future of AI Hardware – GPUs vs. Groq and Beyond Transformers · HackerTrans
I have a few questions about the future landscape of AI hardware and architectures:
- Can GPU architectures eventually rival the speeds of Groq and similar specialized hardware? Is there a foreseeable future where GPU inference with hardware sub-systems (e.g. TPU) and/or algorithmic enhancements can match or surpass these speeds?
- Is there a post-Transformer AI world? Are there any new architectures or enhancements on the horizon? Which of these could continue the GPU dominance?
- Are approaches like Groq easy to optimize for future AI models?
Insights and thoughts appreciated!