NVIDIA’s next-gen GeForce RTX 5080 graphics card will be faster than the RTX 4090 in ray-tracing. According to @XpeaGPU, the GB203 die powering the 5080 will be faster than the RTX 4090 (AD102) in ray-traced gaming, while offering similar performance in rasterization. The RTX 5080 is expected to feature a 256-bit with half as many cores as the RTX 5090. Consequently, the Blackwell 80-class GPU will likely have around 10,000 cores (FP32) paired with a 256-bit bus.
On the memory side, both the RTX 5080 and 5090 are expected to leverage 28Gbps GDDR7 memory. While the latter will offer 24GB, the former will likely retain the 16GB memory buffer of its predecessor. The top three RTX 50 series GPUs are said to feature high-speed GDDR7 memory. This grants high-end Blackwell a 30-40% bandwidth upgrade without even considering the architectural side.
@XpeaGPU claims that the GeForce RTX 5080 and 5090 will offer the biggest generational uplift in NVIDIA’s history, in part because they were designed to fight “a monster 3nm MCM RDNA 4 flagship.” Additionally, it is claimed that the RTX 50 series “Blackwell” graphics cards will compete against AMD’s RDNA 4 and RDNA 5 offerings. This could mean a few things.
We already know that the Radeon RX 8900 XTX (monster MCM GPU) has been scrapped, making the RX 8800 XT the RDNA 4 flagship. So, either NVIDIA will release an RTX 40 Refresh, or AMD will launch the RDNA 5 family (RX 9900 XTX) a short while after the RX 8800 XT. We’ll update this story as we hear more.