At CES in Las Vegas, Nvidia unveiled, along with other products, the new generation of graphics cards, the GeForce RTX 50 Series. In a nutshell, it can be described as pushing the performance and technology capabilities in all areas that Nvidia has been addressing with recent GeForce generations. It offers higher performance, generally better pricing than the last generation, and a host of new and improved technologies like DLSS 4 and Reflex 2.
GeForce RTX 5070
The RTX 5070 looks the same size as the more powerful models due to the proportions of the cooler, but when you compare them by PCIe connector size, it’s similar to the current generation. On top is the RTX 5090, below is the RTX 5070.
| RTX 5070 Ti | RTX 5070 | RTX 4070 Ti Super | RTX 4070 Ti | RTX 4070 | |
| Shader cores | Blackwell | Blackwell | Ada Lovelace | Ada Lovelace | Ada Lovelace |
| Nvidia CUDA cores (SP) | 8960 | 6144 | 8448 | 7680 | 5888 |
| RT cores | 133 TFLOPS | 94 TFLOPS | 102 TFLOPS | 93 TFLOPS | 67 TFLOPS |
| Tensor Cores | 1406 AI TOPS | 988 AI TOPS | 706 AI TOPS | 641 AI TOPS | 466 AI TOPS |
| Boost Clock | 2,45 GHz | 2,51 GHz | 2,61 GHz | 2,61 GHz | 2,48 GHz |
| Base Clock | 2,3 GHz | 2,16 GHz | 2,34 GHz | 2,31 GHz | 1,92 GHz |
| Memory | 16 GB GDDR7 | 12 GB GDDR7 | 16 GB GDDR6X | 12 GB GDDR6X | 12 GB GDDR6 12 GB GDDR6X |
| Memory Bus Width | 256-bit | 192-bit | 256-bit | 192-bit | 192-bit |
| Memory Bandwidth | 896 GB/s | 672 GB/s | 672 GB/s | 504 GB/s | 504 GB/s |
| Max. resolution | 4K at 480Hz or 8K at 120Hz with DSC | 4K at 480Hz or 8K at 120Hz with DSC | 4K při 240 Hz at 8K or 60 Hz with DSC, HDR | 4K at 240 Hz or 8K at 60 Hz with DSC, HDR | 4K at 240 Hz or 8K at 60 Hz with DSC, HDR |
| Standard Display Connectors | 3× DisplayPort, 1× HDMI | 3× DisplayPort, 1× HDMI | 3× DisplayPort, 1× HDMI | 3× DisplayPort, 1× HDMI | 3× DisplayPort, 1× HDMI |
| PCI Express | Gen 5 | Gen 5 | Gen 4 | Gen 4 | Gen 4 |
| Nvidia Encoder | 2× 9th gen. | 1× 9th gen. | 2× 8th gen. | 2× 8th gen. | 1× 8th gen. |
| Nvidia Decoder | 1× 6th gen. | 1× 6th gen. | 1× 5th gen. | 1× 5th gen. | 1× 5th gen. |
| length | by manufacturer | 242 mm | by manufacturer | by manufacturer | 244 mm |
| width | by manufacturer | 112 mm | by manufacturer | by manufacturer | 112 mm |
| height | by manufacturer | 2 slots | by manufacturer | by manufacturer | 2 slots |
| Total Graphics Power | 300 W | 250 W | 285 W | 285 W | 200 W |
| Supplementary Power Connectores | 2× 8pin PCIe (adapter) or 1× 300 W PCIe Gen 5 | 2× 8pin PCIe (adapter) or 1× 300 W PCIe Gen 5 | 2× 8pin PCIe (adapter) or 1× 300 W PCIe Gen 5 | 2× 8pin PCIe (adapter) or 1× 300 W PCIe Gen 5 | 2× 8pin PCIe (adapter) or 1× 300 W PCIe Gen 5 |
The following table summarises how the individual cards of the recent generations differ in terms of technology support.
| Technologie | RTX 50 family | RTX 40 family | RTX 30 family | RTX 20 family | GTX 16 family |
|---|---|---|---|---|---|
| Architecture | Blackwell | Ada Lovelace | Ampere | Turing | Turing |
| Streaming multiprocessors | 2× FP32 | 2× FP32 | 2× FP32 | 1× FP32 | 1× FP32 |
| RT Cores | 4th gen | 3th gen | 2nd gen | 1st gen | – |
| Tensor Cores (AI) | 5th gen | 4th gen | 3rd gen | 2nd gen | – |
| NVIDIA DLSS | DLSS 4 | DLSS 3.5 | DLSS 2 | DLSS 2 | – |
| Super Resolution | Super Resolution | Super Resolution | Super Resolution | – | |
| DLAA | DLAA | DLAA | DLAA | – | |
| Ray Reconstruction | Ray Reconstruction | Ray Reconstruction | Ray Reconstruction | – | |
| Frame Generation | Frame Generation | – | – | – | |
| Multi Frame Generation | – | – | – | – | |
| NVIDIA Reflex | Reflex 2 | Reflex 2 | Reflex 2 | Reflex 2 | Reflex |
| Low Latency Mode | Low Latency Mode | Low Latency Mode | Low Latency Mode | Low Latency Mode | |
| Frame Warp (Coming Soon) | Frame Warp (Coming Soon) | Frame Warp (Coming Soon) | Frame Warp (Coming Soon) | – | |
| NVIDIA Broadcast | Yes | Yes | Yes | Yes | – |
| NVIDIA App | Yes | Yes | Yes | Yes | Yes |
| Game Ready Drivers | Yes | Yes | Yes | Yes | Yes |
| NVIDIA Studio Drivers | Yes | Yes | Yes | Yes | Yes |
| NVIDIA ShadowPlay | Yes | Yes | Yes | Yes | Yes |
| NVIDIA Highlights | Yes | Yes | Yes | Yes | Yes |
| NVIDIA Ansel | Yes | Yes | Yes | Yes | Yes |
| NVIDIA Freestyle | Yes | Yes | Yes | Yes | Yes |
| VR Ready | Yes | Yes | Yes | Yes | GTX 1650 Super or higher |
| NVIDIA Omniverse | Yes | Yes | Yes | Yes | – |
| RTX Remix | Yes | Yes | RTX 3060 Ti or higher | – | – |
| PCIe | 5th generation | 4th generation | 4th generation | 3rd generation | 3rd generation |
| NVIDIA Encoder (NVENC) | 9th generation | 8th generation | 7th generation | 7th generation | 6th generation |
| NVIDIA Decoder (NVDEC) | 6th generation | 5th generation | 5th generation | 4th generation | 4th generation |
| AV1 Encode | Yes | Yes | – | – | – |
| AV1 Decode | Yes | Yes | Yes | – | – |
| CUDA capability | 12.8 | 8.9 | 8.6 | 7.5 | 7.5 |
| DX12 Ultimate | Yes | Yes | Yes | Yes | – |
I guess that’s the most important stuff. Nvidia didn’t say much at the presentation itself, but a lot of material appeared on the web and youtube. I’ve gone through a number of them and tried to summarize and filter the information into this article and not speculate on something Nvidia has already made public. That’s why I’m releasing it a bit later than the gunslingers who only summarized the basic info from the presentation and press conference. I’m sure I haven’t covered everything though, so more info will be added on an ongoing basis. Neural network materials, improved ray reconstruction and other technologies will be worth mentioning, I expect we’ll learn more as the cards go on sale, which should be on January 30 for the RTX 5090 and RTX 5080 and during February 2025 for the RTX 5070 Ti and RTX 5070.
English translation and edit by Jozef Dudáš
