Seonglae ChoAmdahl's Curse
Even with a program that can be 50% parallelized, there is little benefit beyond 10 processors, whereas a program that can be 95% parallelized continues to show significant performance improvements even as thousands of processors are added
The limits of parallelization are explained by Amdahl's Law, while the concept of weak scaling describes what happens when problem size increases, and the cost of data movement is much higher than the cost of computation.
NVIDIA Tensor Core Evolution: From Volta To Blackwell
In our AI Scaling Laws article from late last year, we discussed how multiple stacks of AI scaling laws have continued to drive the AI industry forward, enabling greater than Moore’s Law grow…
https://semianalysis.com/2025/06/23/nvidia-tensor-core-evolution-from-volta-to-blackwell/
