Render Farm Cost per Frame Breakdown (2026)

If you've ever uploaded a project to a cloud render farm and watched credits drain faster than expected, you already know that "cost per frame" is not a fixed number. It depends on your scene, your render engine, how many samples you're pushing, and which pricing model the farm uses. Generic marketing pages rarely spell this out.

We run a farm with over 20,000 CPU cores and a growing fleet of RTX 5090 GPUs. About 70% of our jobs are CPU renders — V-Ray, Corona, Arnold, and the occasional CPU-mode Blender Cycles project. The remaining 30% are GPU jobs. That mix gives us a clear picture of what frames actually cost across different workloads, and it's the basis for everything in this article.

This is a 2026 snapshot. Pricing shifts as hardware evolves and competition adjusts. The goal here is to give you enough math to estimate your own costs before committing credits anywhere.

How Render Farms Charge (And Why "Cost per Frame" Is Misleading)

Most cloud render farms don't charge per frame. They charge per compute-hour, then your per-frame cost falls out of the equation: total compute cost divided by total frames rendered. The confusion comes from marketing pages that quote a single per-frame number without disclosing the scene behind it.

The two dominant pricing models in 2026 are:

GHz-hour (CPU rendering): You pay for the clock speed × time your job consumes. A 3.0 GHz core running for one hour = 3.0 GHz-hours. If the farm charges $0.004/GHz-hr and your frame takes 15 minutes on a 64-core machine at 3.5 GHz, the math is: 64 cores × 3.5 GHz × 0.25 hr × $0.004 = $0.224 per frame.

OBh (GPU rendering): OctaneBench-hours measure GPU throughput. An RTX 4090 scores roughly 700 OB; an RTX 5090 lands around 1,050–1,100 OB in production benchmarks. If the farm charges $0.003/OBh and your frame takes 4 minutes on a single RTX 5090 at 1,050 OB, the math is: 1,050 OB × (4/60) hr × $0.003 = $0.21 per frame.

A few farms still use node-hour pricing — a flat rate per machine per hour regardless of specs. This model is simpler to understand but harder to compare across farms with different hardware.

CPU Rendering: Where Most of the Budget Goes

CPU rendering dominates archviz, broadcast motion graphics, product visualization, and any pipeline that relies on V-Ray, Corona, or Arnold. These engines are mature, deterministic, and scale linearly with core count — which makes cost estimation straightforward.

Here's what we see across real jobs on our farm. These numbers assume $0.004/GHz-hr on nodes with 64 physical cores (128 threads) at 3.2–3.5 GHz — the standard CPU configuration on our fleet.

Scenario	Resolution	Avg. Frame Time	Cost per Frame	Typical Project Size
Archviz interior (V-Ray, moderate)	3000×2000	8–12 min	$0.11–$0.18	5–20 camera angles
Archviz exterior (Corona, GI)	4000×2250	12–20 min	$0.16–$0.30	5–15 angles
Product shot (V-Ray, studio lighting)	4K	5–10 min	$0.07–$0.15	10–50 frames
Broadcast animation (Cinema 4D + Arnold)	1920×1080	3–6 min	$0.04–$0.09	1,500–3,000 frames
Character animation (Maya + Arnold, SSS)	1920×1080	10–20 min	$0.14–$0.30	2,000–5,000 frames
Heavy VFX comp (Nuke + V-Ray, volumetrics)	4K	20–45 min	$0.27–$0.67	500–2,000 frames
Forest Pack/RailClone dense scene	4000×2250	25–40 min	$0.34–$0.60	10–30 angles

The pattern: archviz projects are moderate per frame but there aren't many frames — a typical 15-angle exterior runs $2.40–$4.50. Animation flips the ratio — lower per-frame cost but thousands of frames, so total spend adds up quickly.

A common budget range we see for archviz studios: $50–$300/month. Animation studios doing regular broadcast work tend to spend $500–$2,000/month depending on output volume.

GPU Rendering: Faster Frames, Different Math

GPU rendering is growing fast in production. Redshift, Octane, V-Ray GPU, and Blender Cycles GPU all benefit from the parallel architecture. The RTX 5090 specifically pushed GPU cost-efficiency past the point where it competes with CPU for many workloads.

GPU pricing is harder to compare across farms because OctaneBench scores vary by card, and some farms use custom benchmarks. Here's what GPU frames cost on our RTX 5090 nodes at $0.003/OBh (OB score ~1,050):

Scenario	Resolution	Avg. Frame Time (1× RTX 5090)	Cost per Frame	Notes
Archviz interior (V-Ray GPU)	3000×2000	2–5 min	$0.10–$0.26	Denoiser cuts time 30–50%
Motion graphics (Redshift)	1920×1080	30–90 sec	$0.03–$0.08	Redshift excels here
Product viz (Octane)	4K	1–4 min	$0.05–$0.21	Clean studio setups are fast
Blender Cycles GPU (moderate)	1920×1080	1–3 min	$0.05–$0.16	OptiX denoiser helps
VFX shot (V-Ray GPU, particles)	4K	5–15 min	$0.26–$0.79	VRAM limits can force CPU fallback
Houdini Karma XPU	4K	8–20 min	$0.42–$1.05	Still maturing; CPU fallback paths common

GPU per-frame costs look similar to CPU at first glance, but the total project cost is often lower because frames finish faster — you're renting the hardware for less wall-clock time. The catch is VRAM: if your scene exceeds GPU memory (32 GB on RTX 5090), the render either fails or falls back to CPU paths, which defeats the purpose.

CPU vs GPU: When to Pick Which

The choice isn't always about speed. It's about predictability, scene compatibility, and total cost.

CPU is the safer bet when:

Your scene has textures and geometry exceeding 32 GB, you're using Forest Pack or RailClone with millions of scattered instances, your pipeline is built around V-Ray or Corona CPU workflows, or you need deterministic output that matches local test renders exactly. CPU farms also tend to offer more machines in parallel — on our farm, a 128-thread CPU node is standard, and we can assign dozens of nodes simultaneously. That parallelism matters more than single-frame speed for animation.

GPU makes sense when:

Your engine supports it natively (Redshift, Octane, Cycles), your scene fits in VRAM, you're doing lookdev iterations where turnaround speed matters, or you're working with motion graphics where frame times are already short and GPU cuts them further.

The hybrid approach: Some studios render hero frames on GPU for speed, then switch to CPU for full-sequence batch rendering to keep costs predictable. We see this pattern especially with V-Ray users who do GPU lookdev but CPU final render.

What Drives Cost Up (And How to Control It)

Understanding cost drivers is more useful than memorizing price tables. Here are the factors we see causing the biggest cost variance, ranked by impact:

Resolution and sampling: Doubling resolution quadruples pixel count. Going from 1080p to 4K alone multiplies render time by roughly 3.5–4×. Cranking samples from 2,000 to 8,000 might improve noise by a barely visible margin while tripling cost. Use denoising (V-Ray's built-in denoiser, OptiX, or OIDN) and target the minimum samples that produce a clean result after denoising.

Displacement and subdivision: Heavy displacement maps with high subdivision levels are the single biggest cost multiplier in archviz. A rug with 4 levels of subdivision across a 10-meter floor area can double render time for the entire frame. Bake displacement where possible, or reduce subdivision on objects far from camera.

Light bounces and GI quality: Corona and V-Ray both default to high GI settings. For animation, you can often drop GI quality by 30–50% without visible impact at 24/30 fps playback speed. The eye doesn't catch per-frame noise in motion the way it does in a still.

Scatter density: Forest Pack and RailClone scenes with 10+ million instances consume RAM and inflate render times. Use distance-based density falloff aggressively. Objects more than 50 meters from camera can drop to 10% density with no visible difference.

Render region and passes: Don't render the full frame if you only need to update one element. Most engines support render regions and render passes (beauty, reflection, GI). Re-rendering a single pass is often 5–10× cheaper than re-rendering the full frame.

Estimating Your Project Cost Before You Upload

Here's a practical method we recommend to clients before they commit credits:

Render 3 representative frames locally. Pick an easy frame, a medium frame, and your heaviest frame. Time each one.
Note your local hardware. If your workstation has a Ryzen 9 7950X (16 cores, ~3.8 GHz avg), that's 60.8 GHz. A farm node with 128 threads at 3.5 GHz is 448 GHz — roughly 7.4× more compute.
Estimate farm frame time. Divide your local frame time by the compute multiplier. A 30-minute local frame becomes ~4 minutes on a farm node.
Calculate cost. Frame time × node GHz × rate. For a 4-minute frame on our 448 GHz node at $0.004/GHz-hr: 448 × (4/60) × $0.004 = $0.12/frame.
Multiply by frame count. 1,000 frames × $0.12 = $120 total.
Add 15–20% buffer. Real jobs always have heavier frames than your test sample. Budget accordingly.

This method works for CPU. For GPU, replace GHz with OctaneBench scores and use your GPU's OB score as the baseline.

Priority Tiers and Their Real Impact

Most farms offer priority levels. On our farm, the tiers work like this:

Priority	Typical Cost Multiplier	Use Case
Low / Economy	1× (base rate)	Non-urgent batch renders, overnight jobs
Standard	1.5×	Normal production deadlines
High / Rush	2–3×	Same-day delivery, client revisions

Priority affects how quickly your job starts and how many nodes are assigned simultaneously. The per-frame compute cost doesn't change — you're paying for faster turnaround, not faster individual frames. If you have a flexible deadline, low priority saves 30–50% compared to rush.

The Build-vs-Cloud Break-Even Point

At some spend level, building your own render hardware starts making financial sense. The crossover depends on utilization.

A single 64-core render node (AMD EPYC 9654, 128 threads, 3.55 GHz) costs roughly $8,000–$12,000 in 2026 including chassis, RAM, and storage. That node provides ~454 GHz of continuous compute. At $0.004/GHz-hr, renting equivalent capacity costs $1.82/hour, or ~$1,310/month at 100% utilization.

Break-even: ~7–9 months at full utilization. But most studios don't run 24/7. At 40% average utilization (typical for a small studio with project-based work), break-even stretches to 18–24 months — and that's before electricity, cooling, maintenance, and the opportunity cost of managing hardware.

The practical guidance: if you spend under $1,000/month consistently, cloud is almost certainly cheaper. Between $1,000–$3,000/month, it depends on your utilization pattern. Above $3,000/month sustained, start evaluating a hybrid setup — local nodes for baseline load, cloud for burst capacity. For a deeper analysis, we've written a dedicated build vs cloud cost comparison.

What Other Farms Charge in 2026

Pricing across the industry has converged toward similar ranges. Here's a snapshot of publicly listed rates from major farms (as of early 2026):

Farm	CPU Rate	GPU Rate	Free Trial
Super Renders Farm	$0.004/GHz-hr	$0.003/OBh	$25 credit
GarageFarm	$0.024/GHz-hr (low priority)	$1.49+/node-hr	$25 credit
RebusFarm	$0.0141/GHz-hr	$0.0053/OBh	25 RenderPoints
FoxRenderFarm	$0.0306/core-hr (Diamond tier)	$0.90/node-hr (Diamond tier)	$25 credit
Ranch Computing	Contact for quote	€0.005–0.009/OBh	€30 credit

Note that these rates aren't directly comparable without normalizing for node specs, priority tiers, and how each farm measures "GHz" or "OB." A farm quoting $0.024/GHz-hr on faster hardware might deliver the same per-frame cost as one quoting $0.004/GHz-hr on older machines. Always use the farm's cost calculator with your actual scene data.

Keeping Costs Predictable Month to Month

Cost surprises usually come from three sources: scope creep (more frames than planned), unoptimized scenes uploaded in a rush, and priority upgrades during crunch. Here are patterns we've seen work:

Set a monthly budget cap. Most farms (including ours) let you set spending alerts or hard caps. Use them. It's better to hit a cap and re-prioritize than to discover a $2,000 bill you didn't expect.

Optimize before uploading. Spend 30 minutes checking subdivision levels, texture sizes, and unnecessary geometry before submitting a job. That 30-minute optimization pass often saves 20–40% on render cost.

Batch similar frames. If you have 10 camera angles for an archviz project, submit them as a single batch rather than 10 individual jobs. Batching reduces overhead and lets the farm allocate resources more efficiently.

Use low priority when you can. If the deadline is next week, there's no reason to pay rush rates today. Submit on low priority and let it run overnight.

FAQ

Q: How much does it cost to render one frame on a render farm? A: It depends heavily on scene complexity and resolution. A moderate archviz interior might cost $0.11–$0.18 per frame on CPU, while a heavy VFX shot with volumetrics at 4K could run $0.50–$2.50 or more. The key variables are render time, the farm's hourly rate, and your chosen priority level.

Q: Is CPU or GPU rendering cheaper per frame? A: Neither is universally cheaper. CPU tends to be more cost-effective for complex scenes with high memory requirements (large texture sets, millions of scattered objects). GPU is typically faster and cheaper for scenes that fit within VRAM limits, especially with engines like Redshift or Octane. For a deeper look at how the two compare in production, see our pricing models guide.

Q: Why does my per-frame cost vary between render farms? A: Farms use different hardware, pricing models, and measurement units. One farm may quote GHz-hours while another uses core-hours or node-hours. The underlying hardware speed also differs. Always use each farm's cost calculator with your actual project file for an accurate comparison rather than comparing sticker rates.

Q: How can I estimate render farm costs before uploading? A: Render 2–3 representative frames locally and time them. Then divide your local render time by the compute ratio between your workstation and the farm's node specs. Multiply the estimated farm frame time by the farm's hourly rate and your total frame count. Add a 15–20% buffer for heavier-than-average frames.

Q: Does render priority affect per-frame cost? A: Not directly — the compute cost per frame stays the same. Priority affects how quickly your job enters the queue and how many machines are allocated. Higher priority means faster turnaround but at a 1.5–3× cost multiplier. If your deadline is flexible, low priority can cut total spend by 30–50%.

Q: At what point should I build my own render farm instead of using a cloud service? A: As a rough guide, if you consistently spend over $3,000/month on cloud rendering with high utilization, a hybrid setup may save money over 2–3 years. Below $1,000/month, cloud is almost certainly more economical. The middle range depends on how evenly your workload spreads across the year — seasonal spikes favor cloud, steady loads favor owned hardware.

Q: What's the biggest factor that increases render farm cost? A: Resolution and sampling quality are the top cost drivers. Doubling resolution roughly quadruples render time. After that, displacement/subdivision levels and scatter density (Forest Pack, RailClone) have the largest impact on CPU jobs. For GPU, VRAM overflow that forces CPU fallback is the single most expensive scenario.

Q: Do render farms charge extra for plugins like Forest Pack or V-Ray? A: Most major render farms include licenses for common plugins like V-Ray, Corona, Arnold, Forest Pack, and RailClone in their base pricing. You don't pay separately for these. However, niche or very new plugins may not be supported — always check the farm's supported software list before uploading.

Render Farm Cost per Frame in 2026: What You'll Actually Pay