Quote:
|
NVIDIA's dominance in AI is largely due to CUDA, a dedicated architecture and software ecosystem that simplifies using their GPUs for AI workloads. While technically possible to use AMD (ROCm) or Intel (oneAPI) GPUs, the AI software world is built around CUDA. This makes it much harder for developers to use competitor hardware, as popular frameworks like TensorFlow and PyTorch are optimized for CUDA. It's not that "nobody knows how" to use AMD's platform, but rather that the ecosystem is less mature and requires more effort. AMD faces a real challenge in gaining traction, as fewer users lead to less software support, creating a difficult cycle. While AMD and Intel are working to improve their offerings, it's a tough competitive landscape, not a "hilarious" situation. The core issue is the software and developer familiarity, not just the hardware itself.
|
stream of conscious into'd the LLM and let it fact check what i remembered, so in theory you can buy time by just controlling the supply of Nvida chips, then it doesn't matter what you're doing with power plants or regulations, we're focusing on powering the data centers with natural gas because we have so much of it. and i doubt DARPA or whatever Black project is running behind the scenes gives a shit about the rest of the normal imbalances and advantages China would have. they didn't care about data rights after 9/11 with PRISM and all the other programs. i'd bet money we're doing more insane illegal shit as a country building this than China is right now