NVIDIA H20

DeepSeek-V4-Pro, a 1.6-trillion-parameter Mixture-of-Experts model, achieves 271 output tokens/s per node on H20 GPUs at batch size 1, compared with 383.7 tokens/s on B300, a 1.42× performance gap, ac…

ByteDance Seed and Tsinghua AIR released CUDA Agent, an agentic reinforcement learning system that trains a large language model to generate CUDA kernels outperforming a compiler. On KernelBench, the …

Researchers introduced CoMem (Comprehension Memory), a method that leverages the depth division of labor in large language models to enable unbounded-context memory. In evaluations, CoMem achieved 97.…