Nvidia’s Christmas Present: GB300 & B300 – Reasoning Inference, Amazon, Memory, Supply Chain – SemiAnalysis
Merry Christmas has come thanks to Santa Huang. Despite Nvidia’s Blackwell GPU’s having multiple delays, discussed here, and numerous times through the Accelerator Model due to silicon, packaging, and backplane issues, that hasn’t stopped Nvidia from continuing their relentless march.
They are bringing to market a brand-new GPU only 6 months after GB200 & B200, titled GB300 & B300. While on the surface it sounds incremental, there’s a lot more than meets the eye.
The changes are especially important because they include a huge boost to reasoning model inference and training performance. There is a special Christmas present from Nvidia to all the hyperscalers, especially Amazon, certain players in the supply chain, memory vendors, and their investors. The entire supply chain is reorganizing and shifting with the move to B300, bringing many winners presents, but also some losers get coal.
The B300 GPU is a brand-new tape out on the TSMC 4NP process node, IE it is a tweaked design, for the compute die. This enables the GPU to deliver 50% higher FLOPS versus the B200 on the product level. Some of this performance gain will come from 200W additional power with TDP going to 1.4KW and 1.2KW for the GB300 and B300 HGX respectively (compared to 1.2KW and 1KW for GB200 and B200).
