Nvidia GeForce RTX 50 series lineup includes five chips
At last year’s Arete Technology Conference, NVIDIA’s Vice President and Chief of Accelerated Computing, Ian Buck, reaffirmed NVIDIA’s commitment to updating its primary GPGPU architecture every two years, confirming that the Blackwell architecture GPU will be launched in 2024. It is anticipated that the Blackwell architecture will make its debut in server products, with GeForce graphics cards not appearing until 2025.
Recently, online users have stated on CHH that NVIDIA’s next-generation GeForce RTX 50 series lineup, based on the Blackwell architecture, will consist of five chips: GB202, GB203, GB205, GB206, and GB207. In a departure from previous patterns, there is no GB204 x04 chip following AD104, but instead, an x05 GB205 chip, possibly reflecting a change in NVIDIA’s market positioning strategy.
The anticipated positioning for the GeForce RTX 50 series lineup is as follows:
- GB202 – Ultra Enthusiast
- GB203 – Enthusiast
- GB205 – High-End
- GB206 – Mainstream
- GB207 – Entry-Level
Rumors suggest that the Blackwell architecture GPU used for the GeForce RTX 50 series will retain a single-chip design, with data center products adopting MCM packaging. The primary advantage of continuing the single-chip design at the consumer level is time control and low risk, allowing for timely production. NVIDIA is expected to make significant modifications to the Blackwell architecture. Although not a wholly new design, there will be a new structure for the SM and CUDA, the RT unit may be replaced by a PT unit, and ray tracing performance will be further optimized and enhanced.
Additionally, the GeForce RTX 50 series will employ the PCIe 5.0 interface, with GPU core frequencies reaching 3GHz or higher. The RTX 5090, based on GB102, will have 144 SM clusters, or 18,432 CUDA cores, with an L2 cache of 96MB. Blackwell architecture GPUs are likely to support GDDR7 video memory. Currently, Samsung has completed the development of the industry’s first GDDR7 chip, with a data I/O interface speed reaching 32Gbps. Coupled with a 512-bit memory bandwidth, this will result in a video memory bandwidth exceeding 2TB/s.