Nvidia today announced the GeForce RTX 50 series, the company’s next generation of desktop and mobile graphics processors. Based on the latest Blackwell architecture, the product stack currently includes the RTX 5090, RTX 5080, RTX 5070 Ti, and RTX 5070 in both desktop and mobile variants.
There are obvious specific differences between all four GPUs but let’s first focus on the things they have in common. All new Blackwell GPUs are built on the TSMC 4NP process. These include the new 5th generation Tensor cores and 4th generation ray tracing cores. There is a new 9th generation NVENC encoder and 6th generation NVDEC decoder.
| rtx 5090 | rtx 4090 | rtx 5080 | rtx 4080 | rtx 5070 ti | rtx 4070 ti | rtx 5070 | rtx 4070 | |
|---|---|---|---|---|---|---|---|---|
| generation | blackwell | ada lovelace | blackwell | ada lovelace | blackwell | ada lovelace | blackwell | ada lovelace |
| CUDA core | 21760 | 16384 | 10752 | 9728 | 8960 | 7680 | 6144 | 5888 |
| tensor core | 3352 tops | 1321 tops | 1801 tops | 780 tops | 1406 tops | 641 tops | 988 tops | 466 tops |
| ray tracing core | 318 TFLOPS | 191 TFLOPS | 171 TFLOPS | 113 TFLOPS | 133 TFLOPS | 93 TFLOPS | 94 TFLOPS | 67 TFLOPS |
| boost clock | 2.41GHz | 2.52GHz | 2.62GHz | 2.51GHz | 2.45GHz | 2.61GHz | 2.51GHz | 2.48GHz |
| memory configuration | 32GB GDDR7 | 24GB GDDR6X | 16GB GDDR7 | 16GB GDDR6X | 16GB GDDR6X | 12GB GDDR6X | 12GB GDDR7 | 12GB GDDR6/GDDR6X |
| memory bus | 512-bit | 384-bit | 256-bit | 256-bit | 256-bit | 192-bit | 192-bit | 192-bit |
| Power | 575W | 450W | 360W | 320W | 300W | 285W | 250W | 200W |
| launch price | $1999 | $1499 | $999 | $1199 | $749 | $799 | $549 | $599 |
All four announced models feature fast GDDR7 memory. Connectivity is improved, with all cards supporting PCIe 5.0 with the latest DisplayPort 2.1b UHBR20 (80Gbps) and HDMI 2.1b (48Gbps). All 50-series cards will also support the upcoming Reflex 2 with frame warp as well as DLSS4.
The flagship RTX 5090 has 32% more CUDA cores than the RTX 4090, with 154% more AI TOPS and 66% more ray tracing TFLOPS. You also get 8GB more video memory that’s even faster on a 512-bit memory bus. Power consumption has also increased, with Nvidia targeting 575W for the 5090, although the new Founders Edition card somehow only uses a dual-slot design with double flow-through fans.
The price of the RTX 5090 is a staggering $1999, which is 33% more than the price of the RTX 4090. It will be available from January 30.
The RTX 5080 has a modest 5% increase in CUDA core count compared to the 4080 Super and a comparatively small increase in tensor and ray tracing performance compared to the 5090. Memory also remains at 16GB but with faster GDDR7 on the same 256-bit memory bus. As you can imagine, power has also seen a relatively small increase, going from the 4080 Super’s 320W to 360W.
The RTX 5080 is priced at $999, which is the same as the RTX 4080 Super and lower than the RTX 4080’s launch price of $1199. The 5080 will also be available from January 30th.
Finally, the RTX 5070 Ti is priced at $749, which is $50 less than the launch prices of the 4070 Ti and 4070 Ti Super, while the RTX 5070 is priced at $549, which is $50 less than the 4070 Super and the same as the 4070. The 5070 Ti and 5070 will be available in February.
Nvidia also announced DLSS4, the latest generation of Nvidia’s deep learning super sampling image reconstruction technology. The main new component of DLSS 4 is multi-frame generation, which goes from inserting one AI generated frame to generating and inserting three AI generated frames between every two conventionally generated frames. This technology takes advantage of the significantly improved tensor processing of the 50-series cards to generate these new frames with greater accuracy and quality than before. By the way, this feature is only available on 50-series cards.
However, Nvidia is also improving the DLSS experience on older RTX cards. A new Enhanced Single Frame Generation feature improves the quality of the single-frame frame generation feature on RTX 50 and 40-series cards, further increasing frame rates while reducing VRAM usage.
The company is also switching from CNN-based models to Transformer-based models for features like ray reconstruction, super resolution, and DLAA. The result is an image with more image detail and greater temporal stability, reducing artifacts such as ghosting, aliasing, and smearing in the reconstructed images. These new DLSS 4 enhancements are coming to all RTX cards from the 20-series up. Users can enable these features in existing games through the Nvidia app without waiting for a game update.
DLSS 4.0 with Multi-Frame Generation will be available in 75 apps and games, including Unreal Engine 5, on day 5. Users can also enable the feature in games that have not been updated through the Nvidia app by using the new DLSS override feature.
