RTX 5090 for Blender: 32 GB of VRAM, Cycles speed and supported versions
Short answer
The GeForce RTX 5090 has 32 GB of GDDR7 video memory and 21,760 CUDA cores. Blender supports it from version 4.4. On Blender Open Data, with OptiX and Blender 5, it is the fastest consumer card, about 40% ahead of the RTX 4090. Its 32 GB let it render scenes on the GPU that are too big for the 24 GB of a 4090.
The RTX 5090 is the top card of NVIDIA's GeForce RTX 50 series, released in January 2025. For 3D rendering, two of its specs matter more than the others: its video memory, which decides whether a scene fits on the card, and its speed with OptiX in Cycles. This guide goes through its official specs, the Blender versions that support it and what public benchmarks measure.
RTX 5090 specifications
| Specification | RTX 5090 |
|---|---|
| Video memory | 32 GB GDDR7 |
| Memory bus | 512-bit |
| CUDA cores | 21,760 |
| Architecture | Blackwell |
| Card power | 575 W |
| Power supply recommended by NVIDIA | 1,000 W |
Source: the RTX 5090 specifications page on NVIDIA's website.
32 GB of VRAM: what it changes in Blender
To render on the graphics card, Cycles has to load the scene onto it: geometry, textures, the acceleration structure (BVH, page in French), hair, volumes. Video memory therefore sets the size of the scenes the card can handle. With 32 GB, against 24 GB on the RTX 4090, the 5090 takes heavily textured architectural interiors, instanced forests or heavy simulations that fill up a smaller card.
Memory does not make a render faster on its own: a scene that fits in 12 GB renders at the speed of the chip, not of the memory. What it does prevent, as the scene grows, is the out-of-memory error that forces you to simplify the scene or render on the CPU. The guide which PC for 3D rendering helps estimate how much memory a scene needs.
Which Blender version for the RTX 5090?
The Cycles 4.4 release notes add support for GeForce RTX 50 cards, built on the Blackwell architecture. With Blender 4.4 or newer, the card shows up in Edit > Preferences > System, under CUDA and OptiX. Earlier versions do not officially support it.
Once the card is ticked, set Device to GPU Compute in Render Properties. On an RTX card, OptiX is usually faster than CUDA: the OptiX vs CUDA guide compares the two. The Blender 5 guide (in French) covers the recent versions.
Speed in Cycles: what Blender Open Data shows
Blender Open Data is the public database of Blender Benchmark results, published by the Blender Foundation. The figures below are the median scores of submissions run with OptiX, for Blender 5.0.0, 5.1.0, 5.1.1, 5.2.0, collected on 12 September 2026. A higher score means a faster card.
| Card | Median score | Submissions |
|---|---|---|
| RTX 5090 | 16,171 | 485 |
| RTX 4090 | 11,534 | 270 |
| RTX 5080 | 9,164 | 818 |
| RTX 5090 Laptop GPU | 8,163 | 101 |
On this measure, the RTX 5090 is about 40% ahead of the RTX 4090. The laptop version, despite its name, sits far behind the desktop card: its power draw is much more limited. A benchmark is still an average, and the gap on your own scene may be smaller or larger. The full ranking of cards is on the Blender benchmarks page (in French).
Buy an RTX 5090 or rent one for rendering?
An RTX 5090 makes sense when rendering keeps the machine busy for days on end, month after month. For peaks in workload, an animation with a deadline or a scene that is occasionally too heavy, the power can be rented by the hour: several cards in parallel then cut the render time, which a single card cannot do. The render time calculator gives an order of magnitude for an animation.
Frequently asked questions
How much VRAM does the RTX 5090 have?
The GeForce RTX 5090 has 32 GB of GDDR7 video memory on a 512-bit bus, against 24 GB for the RTX 4090. It is the NVIDIA consumer card with the most video memory.
Does the RTX 5090 work with Blender?
Yes, from Blender 4.4: the Cycles 4.4 release notes add support for the RTX 50 series (Blackwell architecture). Earlier versions do not officially support it.
Is the RTX 5090 much faster than the RTX 4090 in Blender?
On Blender Open Data, with OptiX and Blender 5, the median score of the RTX 5090 is about 40% higher than that of the RTX 4090 (data collected on 12 September 2026). The real gap depends on the scene.
What power supply does an RTX 5090 need?
NVIDIA rates the card at 575 W and recommends a 1,000 W power supply for a reference system; what you actually need depends on the rest of the machine.
Rendering the animation on TURNA
When a scene is already well optimised and the frame count is what takes the time, the remaining lever is rendering several frames at once. TURNA does this for Blender: an add-on (Blender 4.0 to 5.2, Cycles and Eevee) sends the scene in one click, and the frames are split across 1, 2, 4 or 8 GPUs in parallel, RTX 5090 first and RTX 4090 when none is free. The add-on shows the estimated cost before anything starts. Prices start at €2.90 per hour, billed on the time actually used, with a 30-second minimum, and new accounts get €25 of free credit. A single still image is still rendered by one machine, so the gain is on animations. Tiers and worked examples are on the Blender render farm page.
TURNA renders Blender scenes on RTX 5090 first, RTX 4090 otherwise, on 1 to 8 machines in parallel, with the 5090 used only from Blender 4.4. The card you get depends on which machines are free when the render starts: it is not guaranteed.
