NVIDIA's GeForce RTX 5090 is one of the first consumer graphics cards in the GeForce RTX 50 Series, which is based on the Blackwell architecture. This top-end model offers a large boost in performance over the previous-generation RTX 4090, and includes 33% more VRAM (32GB vs 24GB), but all of that comes at the cost of a substantial increase in power consumption and heat output. The RTX 5090 is rated for up to 575W, which is 125W higher than its predecessor, and that increase will make multi-GPU configurations much more difficult to achieve.
Alongside the rise in performance and power, the RTX 5090 also includes several major feature upgrades. There are improved video encoding and decoding capabilities, with added support for 4:2:2 chroma subsampling, as well as new generations of CUDA, tensor, and ray tracing cores. The 32GB of VRAM is also made up of new, high-speed GDDR7 memory. For detailed performance info across a wide range of professional applications, check out our RTX 5090 Content Creation Review.
This variant of the RTX 5090 is from Gigabyte and uses a typical triple-fan cooling layout. It works very well for keeping the GPU cool but is quite large. That size plus the massive amount of heat it puts off means that just one of these can be installed in most systems... and smaller form factor computers probably can't fit it at all.
Finally, in regards to connectivity, there are one HDMI 2.1a and three DisplayPort 2.1 plugs supporting a total of up to four displays at a maximum resolution of 7680x4320 (8K).
Alongside the rise in performance and power, the RTX 5090 also includes several major feature upgrades. There are improved video encoding and decoding capabilities, with added support for 4:2:2 chroma subsampling, as well as new generations of CUDA, tensor, and ray tracing cores. The 32GB of VRAM is also made up of new, high-speed GDDR7 memory. For detailed performance info across a wide range of professional applications, check out our RTX 5090 Content Creation Review.
This variant of the RTX 5090 is from Gigabyte and uses a typical triple-fan cooling layout. It works very well for keeping the GPU cool but is quite large. That size plus the massive amount of heat it puts off means that just one of these can be installed in most systems... and smaller form factor computers probably can't fit it at all.
Finally, in regards to connectivity, there are one HDMI 2.1a and three DisplayPort 2.1 plugs supporting a total of up to four displays at a maximum resolution of 7680x4320 (8K).
Model: Gigabyte GV-N5090GAMING OC-32GD
Specifications
| Chipset Manufacturer | NVIDIA |
| Product Line | GeForce |
| Series | RTX 50 Series |
| Model | 5090 |
| Motherboard Connection | PCI Express x16 |
| Interface | PCI Express 5.0 x16 |
| Maximum Digital Resolution | 7680x4320 |
| Maximum HDMI Resolution | 7680x4320 |
| Maximum DisplayPort Resolution | 7680x4320 |
| Maximum Monitors Supported | 4 |
| HDCP Compliant | Yes |
| Multi-GPU Technology | Not Available |
| TGP | 575 W |
GPU Specifications | |
| Core Speed | 2010 MHz |
| Boost Speed | 2550 Mhz |
| CUDA Cores | 21760 |
| AI Cores | 680 |
| RT Cores | 170 |
Memory Specifications | |
| Video Memory | 32GB |
| Memory Type | GDDR7 |
| Memory Bus Width | 512-bit |
| Bandwidth | 1,790 GB/s |
Performance | |
| Double Precision Floating Point (Peak) | 1.637 Tflops |
| Single Precision Floating Point (Peak) | 104.8 Tflops |
| Half Precision Floating Point (Peak) | 104.8 Tflops |
Additional Information | |
| Cooling System | Open Air Type |
| DirectX Support | DirectX 12 |
| CUDA Support | 12.8 |
| OpenGL Support | 4.6 |
| Removable Backplate | No |
| Slot Width | 3+ |
| Quadro Sync Support | No |
| Max Temp. | 90 C |
| Integrated GPU Brace Mounting | No |
Outputs | |
| Port 1 | HDMI 2.1 |
| Port 2 | DisplayPort 2.1b |
| Port 3 | DisplayPort 2.1b |
| Port 4 | DisplayPort 2.1b |
Power Connectors | |
| Plug 1 | 12V-2x6 |
Dimensions | |
| Length | 342 mm (13.5 in) |
| Height | 152 mm (6.0 in) |
| Width | 70 mm (2.8 in) |
| Net Weight | 2.260 kg (5.0 lbs) |