NVIDIA vs AMD for Blender in 2026: OptiX, HIP and HIP RT Compared

Blender’s Cycles renderer runs on both NVIDIA and AMD graphics cards, so the question is not whether an AMD card works but what you give up or gain with each brand. This guide covers how Cycles talks to each kind of card, which features differ, what current benchmarks show, and when an AMD card is a reasonable choice for Blender.

You choose the GPU backend in Preferences › System › Cycles Render Devices, then set each scene to use the GPU under Render Properties › Device (Blender Manual). Each brand has its own backend:

HIP RT has a short history worth knowing. It was experimental for several releases, Blender 4.4 marked it as no longer experimental and dropped it for RDNA1 cards, and Blender 5.1 turned it on by default. If you compare an AMD card using an older Blender version, or with hardware ray tracing switched off, expect slower results than a current build gives.

Most scenes render the same on both brands, but a few features do not:

• Open Shading Language (OSL): only supported on OptiX among GPU backends. If your shaders use OSL, an AMD card has to fall back to CPU rendering for them.
• GPU denoising: OpenImageDenoise runs on the GPU on discrete Radeon RX 6000 and newer cards on Linux, and on Radeon RX 7000 and newer on Windows. On NVIDIA it runs on RTX cards.
• Shadow caustics: not supported when using HIP without hardware ray tracing.
• Path guiding: not supported on any GPU, so this one does not decide between the brands.

All four points come from the Blender Manual. For both brands, each GPU uses only its own memory, and when VRAM is full Cycles falls back to system RAM, which is slower.

The most recent independent test we found is Puget Systems’ AMD GPU Rendering Performance: Is Radeon a Viable Alternative? (Puget Systems, June 2026). NVIDIA cards used OptiX and AMD cards used HIP; the article does not state the Blender version or whether HIP RT was on. Results:

The same Classroom times as a chart, with each card’s VRAM and launch price:

Two things stand out. First, the cheapest NVIDIA card in the test, the RTX 5070, beat both AMD cards in all three measurements. Second, the gap depends on the scene: on Junkshop the RX 7900 XTX reached 88% of the RTX 5070’s speed, on the Blender Benchmark score 83% (our calculation from Puget’s figures). Puget’s own conclusion is that AMD is no longer automatically ruled out for Blender, but that Radeon suits mixed-use systems and occasional rendering better than high-volume production.

• You mostly model, sculpt or animate and render occasionally. Viewport work and EEVEE run on any current card.
• You need VRAM on a budget: the RX 7900 XTX has 24 GB at a lower launch price than NVIDIA’s 24 GB+ cards, which can decide whether a heavy scene fits at all.
• The machine also serves games or Linux-first work where the AMD card is the better fit for the rest of your use.

Choose NVIDIA when rendering is your main job, when you use OSL shaders, or when you want the widest support across render engines. For the specs that decide which NVIDIA card, see our guide to choosing a GPU for Blender rendering; for the rest of the build, see best PC specs for Blender.

A render farm cannot settle the brand question for your own desk, but it can show you how your scene behaves on a given card before you buy. On a SaaS render farm you upload the .blend file and the farm picks the hardware. On an IaaS render farm such as iRender you rent a specific GPU server, open the scene yourself and measure it. iRender servers are NVIDIA only, so they cover the NVIDIA side of this comparison.

A simple rule: if you only need frames, a SaaS farm is the shorter path whatever card you own. If you are deciding whether an RTX card is worth buying for your work, time your own scene on one first, which is what an IaaS server lets you do.

1. Upload first. Put the .blend file and textures in iRender Drive before you start a server; transfer and storage are free and no rental time runs while you upload.
2. Pick the card to test. RTX 4090 servers (3S, 4S, 5S, 9S) have 24 GB per GPU; RTX 5090 servers (3i, 4i, 5i, 9i) have 32 GB. A 1-GPU server gives you a per-card result; 1-GPU servers have a one-hour minimum, multi-GPU servers bill per minute.
3. Install Blender. Blender is not pre-installed: download the exact version you use from blender.org (the portable .zip runs without installing) and add your add-ons.
4. Measure. Set OptiX in Preferences, render the heaviest frame at final settings and note the time and the memory figure in the render window.
5. Render or decide. If it fits and the time works, render on a 2x, 4x or 8x server, or split frame ranges across several servers for a deadline. Blender needs no render licenses, however many servers you run.

1. Is NVIDIA or AMD better for Blender?

For Cycles rendering, NVIDIA. In Puget Systems’ June 2026 tests every NVIDIA card from the RTX 5070 up was faster than the RX 9070 XT and RX 7900 XTX. AMD works and can make sense for mixed use or occasional rendering.

2. Does Blender support AMD GPUs?

Yes. Cycles renders on AMD RDNA1 and newer cards through HIP, and on Radeon RX 6000 and newer with hardware ray tracing (HIP RT), which is on by default since Blender 5.1.

3. What is HIP RT in Blender?

HIP RT is hardware ray tracing for AMD cards in Cycles. It uses the ray accelerators on Radeon RX 6000 and newer cards, and it is enabled by default from Blender 5.1.

4. Should I use OptiX or CUDA on an NVIDIA card?

OptiX on RTX cards. It uses the RT cores for hardware ray tracing and it is the only GPU backend that supports Open Shading Language.

5. Can I mix NVIDIA and AMD GPUs in one Blender render?

No. Cycles uses one backend at a time, such as OptiX or HIP, so cards from different brands cannot render the same frame together.

6. How can I test my scene on an NVIDIA GPU before buying one?

Rent a server with that GPU, install your Blender version, select OptiX in Preferences and render your heaviest frame at final settings. Compare the time with your current card.

iRender is an IaaS GPU render farm: you rent a bare-metal server with 1, 2, 4 or 8 NVIDIA RTX 4090 (24 GB) or RTX 5090 (32 GB) GPUs, an AMD Ryzen™ Threadripper™ PRO CPU, 256 GB ECC RAM and a 2 TB NVMe SSD, connect by remote desktop and work on it like your own workstation, with full administrator access. Your server image is saved between sessions, so Blender and your add-ons stay installed.

Multi-GPU servers are billed per minute (1-GPU servers have a one-hour minimum), and sessions of 3 hours or more get 10% off. Our 4-GPU and 8-GPU servers are liquid-cooled; in our own measurements the GPUs hold 60–70°C under sustained render load. Elite support is available 7 days a week by hotline, live chat, Skype, WhatsApp and email.

This quick start shows how to start a server and render Blender on it:

Two Cycles runs we recorded on our servers: 120 frames at 1080p on 4x RTX 5090 at 72 seconds per frame, and the same 4K scene on 4x RTX 5090 and 4x RTX 4090 at 87 and 108 seconds per frame per GPU:

New users get a 100% bonus on the first top-up (minimum $50). See plans on the Blender render farm page or download the iRender GPU app.

Sources: Blender Manual: GPU Rendering · Blender 5.0 release notes: Cycles · Blender 4.4 release notes: Cycles · Blender 5.1 release page · Puget Systems: AMD GPU Rendering Performance · Blender Open Data · iRender: multi-GPU render benchmarks.