Best Render Farm for CPU vs GPU Architecture Rendering: Which Costs Less?
GPU rendering is almost always cheaper per image for architecture — but CPU wins in two specific situations. On iRender’s same server (RTX 4090 GPU + 64-core Threadripper Pro CPU, ~$8.20/hour), V-Ray GPU renders a 4K interior in 10–30 minutes ($1.40–4.10). V-Ray CPU on the same server takes 25–75 minutes ($3.40–10.25) — roughly 2.5× slower and 2.5× more expensive per image. GPU wins clearly. But CPU wins when: (1) Your scene exceeds 24GB VRAM (100M+ polygon exteriors with Forest Pack) — CPU uses the server’s 256GB RAM instead. (2) You use Corona, which is CPU-only — and on SaaS farms (RebusFarm), Corona’s multi-node distribution makes CPU rendering faster and cheaper than V-Ray GPU on a single iRender server.
| Scenario | V-Ray GPU (RTX 4090) | V-Ray CPU (64-core) | Corona CPU (SaaS) | Winner |
|---|---|---|---|---|
| Simple interior | 10–20 min / $1.40–2.70 | 25–50 min / $3.40–6.80 | 5–12 min / $1.50–3.50 | GPU (or Corona SaaS) |
| Complex exterior | 20–45 min / $2.70–6.15 | 50–120 min / $6.80–16.40 | 8–20 min / $2.50–6.00 | GPU (or Corona SaaS) |
| Heavy scene (100M+ polys) | ⚠️ VRAM crash risk | 90–240 min / $12–33 | 12–30 min / $4–9 (SaaS) | CPU (or Corona SaaS) |
| 10-image batch | 2.5–5 hrs / $20–41 | 8–20 hrs / $66–164 | 20–45 min / $15–40 (SaaS) | Corona SaaS for batch |
Why Is GPU Faster but CPU Handles Bigger Scenes?
Think of it this way: a GPU is like 16,000 workers in a small room (24GB VRAM). They work incredibly fast together, but if the project is too large to fit in the room, nobody can work. A CPU is like 64 workers in a warehouse (256GB RAM). They’re slower individually, but the warehouse can hold virtually any project size. When the project fits in 24GB? GPU every time. When it doesn’t? CPU is your only option.
The RTX 4090’s 24GB VRAM handles approximately 80 million polygons comfortably. Above 100 million — typical for urban masterplans with dense vegetation — VRAM overflow causes crashes without warning. On iRender, you can switch from GPU to CPU mode in V-Ray’s settings without changing servers. Same $8.20/hour, just a different render engine. The server has both GPU and CPU available simultaneously.
Where Does Corona Fit in the CPU vs GPU Debate?
Corona throws a wrench in the simple “GPU is cheaper” conclusion. Corona only uses CPU — it can’t access the GPU at all. On a single iRender server, Corona is slow and expensive (the RTX 4090 sits idle while the CPU does all the work). But on SaaS farms (RebusFarm, GarageFarm), Corona distributes each image across 4–8 CPU nodes simultaneously, making it competitive with — and often faster than — V-Ray GPU on a single iRender server. For batch rendering (10+ images), Corona on RebusFarm is the most cost-effective option in our testing.
See more: Test GPU and CPU rendering on the same server → Test GPU and CPU rendering on the same server → View GPU+CPU cloud servers on iRender
Frequently Asked Questions
- Can I switch between GPU and CPU rendering on iRender?
Yes — on the same server, in the same session, for the same $8.20/hour. In V-Ray’s render settings, toggle between “GPU (CUDA/RTX)” and “CPU” mode. No server restart needed. We recommend: start with GPU mode. If the render crashes (VRAM overflow), switch to CPU and re-launch. Your scene, materials, and settings stay identical — only the render engine changes.
2. Is Corona or V-Ray cheaper for architecture on cloud?
On iRender alone: V-Ray GPU is cheaper (2.5× faster than Corona CPU on the same server). When factoring in SaaS farms: Corona on RebusFarm often matches or beats V-Ray GPU cost through multi-node parallel rendering — especially for batches of 10+ images. The best strategy: V-Ray GPU on iRender for interactive work, Corona on RebusFarm for overnight batch output. Both renderers cost $350/year from Chaos.
3. At what polygon count should I switch from GPU to CPU?
Watch your VRAM usage in V-Ray’s log during rendering. If it approaches 22–23GB on the RTX 4090, the next complex scene may overflow. As a general rule: under 80M polygons = GPU is safe and fast. Between 80–100M = test on GPU, have CPU as fallback. Above 100M = switch to CPU proactively or use V-Ray proxies to reduce VRAM. The 80M threshold assumes standard arch-viz materials; heavy displacement maps push this lower.
Related post: Best Render Farm for Architecture Competitions: Fast Deadline Rendering on Cloud
