Source item

Recent Updates on Nvidia Hardware: HGX B300 NVL8, DGX Spark, and N1X AI PCs/Laptops 1. HGX B300 NVL8: About two weeks ago, Nvidia transitioned from HGX B300 NVL16 to NVL8 to meet the demand for faster GPU-to-GPU interconnects required by latest LLMs with sparse MoE architectures, such as DeepSeek and LLAMA4. Both NVL8 and NVL16 feature 16 GPU dies, but NVL8 integrates two GPU dies per package, connected via Nvidia’s high-bandwidth interface (NV-HBI), delivering bi-directional bandwidth of up to 10 TB/s. Compared to the NVL16, whose GPU dies communicate only through NVLink, the NVL8’s NV-HBI is far better suited to sparse MoE workloads. This shift will enable Nvidia to predominantly adopt CoWoS-L packaging going forward. Since the AI chips of GB and HGX platforms share same packaging, Nvidia can optimize supply chain management. 2. DGX Spark & N1X: The DGX Spark uses the GB10 chip, which is scheduled for mass production around June-July, with mass shipments of end products expected in August-September. The N1X is Nvidia’s Windows-on-ARM processor. Although its architecture basically similar to GB10, significant additional software optimization is required due to its application in laptops and supporting Windows. This extra optimization effort has impacted the timeline for mass shipment schedule of the end product. While market rumors suggest potential delays of N1X laptops into 2026, my latest supply chain research indicates Nvidia and laptop brands currently remain committed to launching before the end of 2025.