AMD's Day-Zero DeepSeek Support Still Costs Up to 42x More Than NVIDIA
- Source
- x.com
- Date
POWER OF CUDA MOAT ALERT🚨: 2 days after CUDA vLLM supported DeepSeekv4.1 Flash, AMD finally publicly released its DeepSeek v4.1 Flash image. Functionally, it works out of the box, but performance-wise, it is currently up to 14.8x worse perf per dollar than H200 and up to 42x worse perf per dollar than B200/B300 currently. The 🚀 POWER OF THE CUDA MOAT 🚀 is that NVIDIA's collaboration with its massive 6 million-developer community ecosystem means that CUDA is optimized on day 0. As AMD Anush said, "Speed is the Moat," and day 0 model support shows CUDA is the speed.
Context
In a post benchmarking day-zero support, SemiAnalysis tracks how quickly hardware makers support brand-new AI models with working, optimized software on release day. Two days after NVIDIA's software stack supported DeepSeek v4.1 Flash, AMD released its own image for the model.
SemiAnalysis reports AMD's version works out of the box but delivers up to 14.8 times worse performance-per-dollar than NVIDIA's H200 and up to 42 times worse than B200 or B300, and attributes the gap to NVIDIA's much larger developer ecosystem optimizing its software immediately around new model releases. SemiAnalysis has separately argued this recurring gap reflects software and tooling investment rather than a fundamental silicon disadvantage.
Checking sign-in…
Loading comments…