Global Graphics Cards
NVIDIA
NVIDIA GeForce RTX 5090
32 GB • Blackwell • 2410 MHz • 575 W
32 GBMemory
BlackwellArchitecture
2410 MHzBoost
575 WPower
32 GB • Blackwell • 2410 MHz • 575 W
A summary of NVIDIA's official product description.
Announced at CES on January 6, 2025 and released on January 30, the RTX 5090 is built on the Blackwell architecture with 32 GB of GDDR7 memory. NVIDIA frames it around three uses at once: full ray-traced gaming at the lowest latency, demanding creative workloads, and running AI models locally, with DLSS 4 neural rendering used to raise frame rates and cut latency.
Manufacturer's official processing unit and architecture information.
Blackwell — Architecture · 21760 — Shading Unit · 170 RT — Ray Tracing Core · 680 Tensor — AI core
Values decisive at high resolutions and high textures.
32 GB — GDDR7 · 512 bit — Memory bus · 1792 GB/s — Memory bandwidth
Official values determining power supply and case selection.
575 W — Card Power Consumption · 1000 W — Recommended Power Supply · 16-pin (12V-2x6) — Power Connector
A usage guide built from the specs and the editors' tags; not a test result.
For gaming; tagged from the display's refresh rate and the chip's class. We have not measured frame rates or heat ourselves.
Good fitFor heavy work use; tagged from the battery and display specs. We have not timed the battery ourselves.
Good fitThe Kacbu editors' assessment of the specifications the maker publishes. We do not test devices ourselves, so nothing here is a measured result. Anything taken from an independent review is shown with its publication and date in the Press section.
Instead of tracking today's retail price, Kacbu keeps the product's official launch information as a reference.
The manufacturer's official announcement date.
The start of the initial sales period.
NVIDIA is headquartered in the US and designs the GPU there; it does not fabricate it itself — the GB202 die is made at TSMC's fabs in Taiwan. Card assembly varies by board partner.
Click a linked value to see other products that share that same spec, and leave a comment on that spec.
What the numbers on the spec sheet actually mean in day-to-day use.
Other products in the same category.
Whether a model runs on your machine is decided first by the graphics card's memory: the model file has to fit in VRAM with room left for the context window and the runtime. The states below are computed from memory alone; software support (CUDA, ROCm, Vulkan) is a separate matter.
Largest model that runs comfortably: Qwen3.5 27B
Rule: 1.5× the requirement in memory is Comfortable, 1.15× Runs, 0.6× Limited; the requirement is the default package size, or the maker's stated minimum where there is one (a stated minimum counts as fitting at exactly that figure). Package sizes are the Ollama library's, minimums the maker's own statement; read on 2026-09-11. Sources: ollama.com/library/qwen3.5, ollama.com/library/gpt-oss, openai.com/index/introducing-gpt-oss