Nvidia RTX 5090 is an Enthusiast-level desktop graphics card that will be released on Jan 30th, 2025. It is a 4K gaming GPU based on Nvidia Blackwell Architecture which belongs to the RTX 5000 series graphics.
The latest and upcoming RTX 5090 graphics cards will be announced at CES 2025; So the RTX 5090 announcement date will be between Jan 7-10, 2025.
I have collected all the data from reputable sites like Techpowerup, VideoCardz, and YouTube channels like Hardware Unboxed, Gamers Nexus, etc. for the specs of the RTX 5090 Graphics card.
The post will be updated after the release of RTX 5090 at a later date. Let’s dive into all the specs of the RTX 5090 GPU.
Nvidia RTX 5090 Info
All the basic information about the Nvidia RTX 5090 GPU is presented such as GPU Architecture, market segment, codename, and price.
GPU Name | GB202 |
GPU Variant | GB202-400-A1 |
Series | RTX 5000 |
Architecture | Blackwell |
Release date | Jan 30th, 2025 |
Generation | GeForce 50 |
Foundry | TSMC |
Market Segment | Desktop |
Production | Announced |
Predecessor | GeForce 40 |
Launched Price | $1999 |
Current Price | — |
Nvidia RTX 5090 launch price is expected to be around $1999. The RTX 5090 architecture is based on Blackwell.
Technical Specifications
The Nvidia RTX 5090 Specs include compute units, number of shaders, pixel rate, texture filter rate, die size, density, TDP, process size, Vram, and more.
5090 Graphics Card include 4th Gen Ray Tracing cores and 5th Gen Tensor cores. RTX 5090 AI TOPS is 3342 TOPS and Ray Tracing TFLOPS is 318 TFLOPS.
SM Count | 170 Units |
Shaders / Cuda Cores | 21760 Units |
TMUs (Texture Mapping Units) | 680 |
ROPs (Raster Operation Pipeline) | 176 |
RayTracing Cores | 170 |
Tensor Cores | 680 |
L1 Cache | 128 KB (per SM) |
L2 Cache | 96 MB |
Transistors | 92.2 billion |
Process Size | 4 nm |
Die Size | 750 mm² |
Density | 122.9M / mm² |
Pixel Rate | 462.7 GPixel/s |
Texture Rate | 1,638 GTexel/s |
RTX 5090 cuda cores are 21,760 units instead of 24,576. The process size is 4nm and the transistors are 92.2 billion.
Board Design
TDP (Total Draw Power) | 575W |
Minimum Power Supply Req. | 1000W |
Tjunc. Max Temp | 90°C |
Slot Width | 2- Slot |
Card Dimensions (L X W X H) | 12 x 5.4 x 1.9 (inches) |
Outputs | 1x HDMI 2.1a, 3x DisplayPort 2.1b |
Bus Interface | PCIe 5.0 x16 |
Power Connectors | 1x 16-pin |
Nvidia RTX 5090 TDP is 575W and its Dimensions are 12 x 5.4 x 1.9 (inches) or 304mm x 137mm x 48mm. RTX 5090 idle power consumption is 46W and peak power draw is 670W.
The bus interface for RTX 5090 is PCIe 5.0 but it does support ATX 3.1/PCIe 5.1 standards according to VideoCardz. The RTX 5090 may support a 1x 16-pin connector to a 4x 8-pin type.
GPU Frequencies
In desktop/notebook GPUs, the graphics cards are tuned according to their manufacturers/Board partners, depending on which frequencies they want in their notebook/desktop GPUs.
If they want power efficiency, they want to tune the frequencies lower than normal, resulting in less power consumption and loss of performance.
If they want performance, they will either let the frequencies default or tune the frequencies a little higher for more performance. Hence, the GPU will consume more power.
The higher the clock speed, the better the GPU performance.
Base Clock | Boost Clock | FP16 (Half) | FP32 (Float) | FP64 (Double) |
---|---|---|---|---|
2.01 GHz | 2.41 GHz | 104.8 TFLOPS | 104.8 TFLOPS | 1.638 TFLOPS |
Video Memory (VRAM) capacity and Type
It tells how much VRAM GPU has, as well as VRAM type, size, bus, and bandwidth. This is a discrete video card so RTX 5090 VRAM is 32GB GDDR7.
Memory Size | 32GB |
Memory Type | GDDR7 |
Memory Bus | 512-bit |
Memory Clock | 28Gbps |
Bandwidth | 1792 GB/s |
RTX 5090 memory size is 32GB GDDR7 and memory bandwidth is 1792GB/s.
API Features Support
Graphical API features supported by the Nvidia Geforce RTX 5090 graphics card are:
DirectX | 12 Ultimate (12_2) |
CUDA | 12.8 |
OpenGL | 4.6 |
OpenCL | 3.0 |
Vulkan | 1.4 |
Shader Model | 6.8 |
Hardware Codec Support
Hardware codecs supported by the RTX 5090 32GB graphics card are:
h264 | Decode / Encode |
AV1 | Decode / Encode |
h265 / HEVC (8-bit) | Decode / Encode |
h265 / HEVC (10-bit) | Decode / Encode |
VP8 | Decode / Encode |
VP9 | Decode / Encode |
NVENC / NVDEC (3x 9th gen / 2x 6th gen) | Yes |
Other Features Support
HDMI 4K Support | Yes (4K at 480Hz), HDR |
HDCP-Version | 2.3 |
Max. Display Support | 4 |
Max. Resolution Support | 8K at 120Hz with DSC, HDR |
Nvidia DLSS | DLSS 4, Super Res, DLAA, Ray Reconstruction, Frame Generation |
Nvidia Reflex | Yes |
Nvidia Broadcast | Yes |
Resizable BAR | Yes |
Nvidia Geforce Experience | Yes |
Nvidia Ansel | Yes |
Nvidia FreeStyle | Yes |
Nvidia ShadowPlay | Yes |
Nvidia Highlights | Yes |
Nvidia G-Sync | Yes |
Nvidia Omniverse | Yes |
Nvidia Studio Drivers | Yes |
Game Ready Drivers | Yes |
VR Ready | Yes |
Nvidia GPU Boost | Yes |
Nvidia NVLink (SLI-Ready) | No |
Synthetic Benchmark Performance
These benchmarks are measured by GPU benchmarking software like 3DMark Time Spy graphics score, 3DMark Fire strike graphics, Passmark Performance Testsuite, Port Royal, and Geekbench 6.2.
What is the Passmark Performance Test suite?
The Passmark performance test suite is a 3D benchmark designed for video cards by using the DirectX feature.
It puts a load on the GPU for various DirectX versions like DX9, DX10, DX11, and DX12 provides separate benchmark results, and gives you the average G3D mark score.
What is the 3D Mark Time Spy benchmark?
It calculates both the CPU and GPU performance by rendering on 1440p resolution using DX12 API and gives you the system’s overall score. It also ranks your overall score on the internet and your system accordingly.
Time Spy Extreme also uses DX12 API to benchmark a gaming PC to its fullest capabilities by rendering it on 4K Ultra HD resolution.
What is the 3D Mark Fire Strike benchmark?
It is a standard to test your gaming PC graphics render capability in detail and complexity which is quite taxing to the gaming hardware. It uses DirectX 11 API and renders at 1080P resolution.
Fire Strike Extreme benchmark uses 1440p as a rendering resolution for gaming PCs. It enhances all the visuals by increasing the rendering load to measure the accurate performance of extreme hardware setups.
What is the 3D Port Royal benchmark?
The 3D Port Royal benchmark measures the ray-tracing capability of a graphics card that supports Microsoft DXR. It renders the scenes at 1440p resolution.
What is the Geekbench 6 GPU benchmark?
Geekbench 6 GPU benchmarks your graphics card computing power in gaming, image processing, video editing, photography, and machine learning. It tests your GPU Open CL, metal, and Vulkan score.
Benchmark Tools | RTX 5090 32GB |
---|---|
Time Spy Graphics | 48732 |
Time Spy Extreme Graphics | 25485 |
Firestrike Graphics | 101767 |
Firestrike Extreme Graphics | 60778 |
Firestrike Ultra Graphics | 33114 |
Port Royal | 36673 |
Open CL | 367740 |
Vulkan | 359742 |
Passmark | — |
Nvidia RTX 5090 Time Spy graphics score is 48,732, Time Spy Extreme graphics score is 25,485, Firestrike graphics score is 101,767, Firestrike Extreme graphics score is 60,778, and Firestrike Ultra is 33,114.
The 3d mark Port Royal score for RTX 5090 GPU is 36,673 (1440p) (Source: Tomshardware).
Nvidia Geforce RTX 5090 Open CL score is 367,740 and Vulcan score is 359,742 (Source: VideoCardz).
GeForce RTX 5090 Gaming Benchmark
The Test System specs used for this gaming benchmark are Ryzen 7 9800X3D CPU, 32GB (2x 16GB) DDR5-6000 MHz RAM, MSI 1 TB M470 SSD, RTX 5090 Founders Edition, and Win 11 Pro. (Source: Hardware Unboxed)
4K Native Performance
Games | 4K Avg. FPS | 4K 1% Low | Graphics Settings |
---|---|---|---|
Cyberpunk 2077 PL | 124 | 102 | High/Ultra |
Delta Force | 160 | 123 | Ultra |
Space Marine 2 | 125 | 99 | Ultra, 4K tex. |
Counter Strike 2 | 472 | 273 | V.High |
Marvel Rivals | 101 | 61 | Ultra, Lumen |
Hogwarts Legacy | 144 | 82 | High |
Star Wars Jedi Survivor | 102 | 88 | Epic |
God of War Ragnarok | 195 | 166 | Ultra |
A Plague tale: Requiem | 128 | 88 | Ultra |
STALKER 2 | 71 | 52 | Epic |
Dragon Age: The Veilguard | 96 | 78 | Ultra |
Dying Light 2 | 122 | 107 | High |
Marvel’s Spiderman Remastered | 212 | 171 | V.High |
The Last of Us | 125 | 115 | Ultra |
1440p Ray Tracing Performance
Games | 1440p Avg. FPS | 1440p 1% Low | Graphics Settings |
---|---|---|---|
Cyberpunk 2077 | 129 | 90 | DLSS Q, RT Ultra |
Metro Exodus Enhanced | 248 | 191 | Ultra, RT normal, hybrid, Physx, hair, Tess |
Alan Wake 2 | 105 | 89 | High, RT High, DLSS Q |
Dying Light 2 | 166 | 143 | RT High, DLSS Q |
Black Myth Wukong | 98 | 79 | V.High, RT V.High, DLSS Q |
Marvel’s Spiderman Remastered | 128 | 106 | V.High, RT V.High, Obj 10, DLSS Q |
Nvidia GeForce RTX 5090 GPU by following 3rd Party Companies
The following 3rd Party companies or board partners of Nvidia GPUs are ASUS, Gigabyte, Zotac, Colorful, Inno3D, Galax, MSI, Palit, PNY, Yeston, Gainward, KFA2, Kuroutoshikou, Leadtek, Manli, and Maxsun.
There are some OEM Brands like Dell, Lenovo, and HP also use Nvidia graphics cards.
Nvidia RTX 5090 Founders edition and 3rd party board partners’ GPUs will be available on Jan 30th, 2025.
The difference between the Founders Edition and 3rd Party board partner GPUs is that they have higher clock speeds, much better and beefier coolers with triple fan design, and RGB.
FAQs
How much faster is RTX 5090 compared to 4090?
RTX 5090 is barely 20-31% faster than RTX 4090.
What is the power requirement of RTX 5090?
RTX 5090 power requirement is at least 1000W as the 5090 TDP is about 575W.
What will be the price of RTX 5090?
RTX 5090 launch price is $1999 which is significantly $400 more than RTX 4090.