The GeForce RTX 5090 represents the pinnacle of NVIDIA's upcoming GPU offerings. It is built on the GB202-300-A1 die architecture and is equipped with 21,760 FP32 CUDA cores, providing substantial parallel processing capabilities essential for intensive computational tasks and high-resolution gaming. The RTX 5090 features a GPU configuration of 170 Streaming Multiprocessors (SM), which contribute to its robust performance metrics. Memory-wise, this top-tier model is outfitted with 32 GB of GDDR7 memory operating on a 512-bit memory bus. Each GDDR7 memory module runs at 28 Gbps, resulting in a total memory bandwidth of 1,568 GB/s. Despite its high-performance components, the RTX 5090 maintains a thermal design power (TDP) of 600 watts, indicating efficient power management relative to its computational power. This balance of high memory bandwidth and efficient power usage underscores NVIDIA's focus on delivering powerful yet manageable GPU solutions.
In contrast, the GeForce RTX 5080 further delineates the xx80 and xx90 SKU categories with its distinct specifications. The RTX 5080 is built on the GB203-400-A1 die and features 10,752 FP32 CUDA cores, which is a significant reduction compared to the RTX 5090. This GPU is paired with 16 GB of GDDR7 memory on a 256-bit bus, maintaining the memory speed at 28 Gbps but resulting in a lower memory bandwidth of 784 GB/s. The RTX 5080 operates within a 400-watt TDP, making it a more power-efficient option suitable for users who require high performance without the maximum power draw of the top-tier model. Comparatively, the RTX 4090 boasts 68% more CUDA cores than the RTX 4080, and the RTX 5090 is rumored to have approximately 102% more CUDA cores than the RTX 5080. This significant increase in CUDA cores between the xx80 and xx90 models suggests NVIDIA's intention to clearly differentiate performance tiers, catering to a broader range of user needs and applications.
The segmentation of the RTX 50 series into the RTX 5090 and RTX 5080 models reflects NVIDIA's response to diverse market demands, offering varying levels of performance and power efficiency. The detailed specifications indicate that NVIDIA is enhancing the performance capabilities of its high-end GPUs while also addressing power consumption concerns. As the release date approaches, the pricing strategy for these GPUs will be critical in determining their competitiveness and accessibility in the market.
Unless there are unexpected changes, two new graphics cards will be launched together at the CES 2025 exhibition in early January next year. NVIDIA might also introduce the RTX 50 mobile graphics card at this event.
Taiwanese media reported that the GeForce RTX 5090 has a total graphics power (TGP) consumption of 600W, and the RTX 5080 consumes 400W TGP. Both models can only use a single 12V-2×6 connector. The RTX 5090 will feature a 14-layer PCB design and support advanced technologies such as PCIe 5.0 and DisplayPort 2.1a UHBR20.
Sources: kopite7kimi (RTX 5090), kopite7kimi (RTX 5080)
