At last year's Arete Technology Conference, Nvidia Vice President and Chief Director of Accelerated Computing Ian Buck reiterated Nvidia's plan to update the main GPGPU architecture every two years, confirming that Blackwell architecture GPUs will be launched in 2024 and will first be used in data center products, while GeForce graphics cards will have to wait until 2025.
Recently, Moore's LawisDead stated that the GeForce RTX50 series gaming graphics cards equipped with Blackwell architecture GPUs have been released in the fourth quarter of 2024. The specific time depends on the sales of AdaLovelace architecture products and the performance level of competitor AMD's new generation products.
It is understood that the improvement of Blackwell architecture compared to AdaLovelace architecture is smaller than that of AdaLovelace architecture compared to Ampere architecture. This is definitely not good news for gamers. The current revenue of NVIDIA data center GPUs is five times that of gaming GPUs, and it is rumored that AMD's next-generation RDNA4 architecture will not have a flagship model. As the business focus shifts and competitive pressure decreases, it seems that NVIDIA's pace of upgrading has also slowed down.
It has been previously reported that the GeForce RTX50 series GPUs based on the Blackwell architecture will be manufactured using the TSMC 3nm process. According to TSMC's past statements, compared with the existing 5nm process node, the 3nm process node can bring about a 15% performance improvement under the same power consumption and number of transistors. There are rumors that the GeForce RTX50 series GPU will have five chips: GB202, GB203, GB205, GB206 and GB207, using PCIe5.0 interface, supporting DisplayPort2.1 display output, and will be equipped with a new "12V-2×6" connector.