NVIDIA's Blackwell GPUs face potential design modifications that could increase power consumption, affecting GeForce RTX 50 launch timing.

NVIDIA's GeForce RTX 50 "Blackwell" GPUs represent the next major graphics market launch for PC platforms, highly anticipated by gamers and enthusiasts planning to upgrade from older RTX GPUs. Recently, the company made specific changes to its Data Center Blackwell AI chips involving a redesign of the mask layer. Blackwell is on track for volume production in Q4 2024, when these chips will generate billions of dollars in revenue for the company.

The redesign was prompted by significant technical issues identified in the data center variant. The main problems involved thermal expansion (CTE) mismatch between the two GPU dies, LSI bridge, RDL interposer, and the main substrate, which caused chip warping and system failure. The solution implemented was a redesign of the top metal layer and bumps. While the Data Center and Consumer tier Blackwell chips will differ significantly from one another, their fundamental architecture remains very similar, raising questions about whether the consumer-tier GeForce RTX 50 series, including the RTX 5090, was affected by these same issues.

NVIDIA has clarified that samples sent to customers were already fixed units, and the volume production heading into the market will include the updated design. However, the company is currently prioritizing Data Center Blackwell GPUs, as they constitute the bulk of its revenue and face enormous demand, with supply bottlenecked only by TSMC's production capabilities. There have been reports that the GeForce RTX 50 launch might shift from Q4 to CES 2025.

According to industry sources including Kopite7kimi, the upcoming RTX 50 GPUs will see increases in power consumption, with higher-end SKUs experiencing bigger increases. Kopite7kimi stated: "I know you someone got the details of GeForce of Blackwell recently. Both of them all have some increase in power consumption, with higher SKUs increasing more." Uncertainty remains about whether these power increases apply only to prototype and ES chips or to the final retail SKUs.