The hidden costs of rendering locally include depreciation, power, cooling, blocked time and a VRAM ceiling, and a simple cost model shows when a cloud machine becomes cheaper.
Cost Audit // Local Workstation Ownership & Real Cost Per Render Hour
A look at the hidden costs of rendering locally on a high-end GPU workstation: what the hardware draws, what it loses in value, and when a cloud machine becomes the cheaper hour (specifications from NVIDIA product pages, cost model illustrative, October 2026).
- Utilisation Decides Everything: in our illustrative model a $4,500 RTX 4090 workstation that loses $900 a year costs about $4.50 per render hour at 200 hours a year but only $0.60 at 1,500 hours.
- Power Is Small, Power Delivery Is Not: electricity is about $0.12 per render hour at an assumed 600 W and $0.20 per kWh, yet NVIDIA lists an 850 W minimum system power for the RTX 4090 and 1000 W for the RTX 5090.
- Blocked Time Can Outweigh Both: if you wait on renders instead of working, blocked time can cost more per hour than depreciation and electricity combined.
- The Break-Even Depends on Your Inputs: with our assumptions, a cloud machine at $8.20 per hour only wins below roughly 112 render hours a year, or about 254 hours once blocked time is priced.
- Practical Production Fix: log your render hours for a month, price your own tariff and blocked time, then compare against a cloud rate. The cloud side has costs of its own, covered in render farm hidden costs. Measure first, then decide.
The hidden costs of rendering locally rarely show up on a receipt. You pay for the GPU once, so every render hour feels free. It is not. Depreciation, power, cooling, blocked time and a hardware ceiling all add up, and the right comparison with a render farm depends on how many hours a year the machine actually renders. This guide puts numbers on each one using a transparent model you can rerun with your own inputs.
Contents
What are the hidden costs of rendering locally?
The main hidden costs of rendering locally are hardware depreciation spread over too few hours, electricity, room cooling, power delivery upgrades, the time your workstation is blocked, crashes and restarts, and the VRAM ceiling that forces an early upgrade. None of them appear in the sticker price of the GPU.
| Cost | Why it hides | How to measure it |
|---|---|---|
| 1. Depreciation | Paid once, felt never | Purchase price minus expected resale, divided by years, then by render hours |
| 2. Electricity | Buried in the household or studio bill | Wall watts x hours x your tariff; use a plug-in power meter |
| 3. Cooling | Shows up as an AC bill or a hot room | Heat output of the machine plus the extra cooling load |
| 4. Power delivery | Shows up when you build or upgrade | PSU, cabling and circuit capacity |
| 5. Blocked time | Not invoiced | Hours you wait x your hourly value |
| 6. Failures and restarts | Looks like bad luck | Hours of render lost to crashes, throttling or a power cut |
| 7. Upgrade ceiling | Appears when a scene stops fitting | VRAM headroom against your largest scenes |
How does utilisation change the cost per render hour?
Depreciation does not depend on how much you render, so the cost per render hour falls as usage rises. In the model below, the same machine costs $4.50 per hour at 200 render hours a year and $0.60 per hour at 1,500. When you count the hidden costs of rendering locally, utilisation matters more than the GPU price.
The model is illustrative, not a measurement. It assumes a $4,500 RTX 4090 workstation, a resale value of $1,800 after three years (so $900 a year of lost value), a draw of 600 W at the wall under render load, a tariff of $0.20 per kWh, and a 25% uplift for extra cooling. Replace each input with your own.
| Render hours per year | Depreciation per hour | Electricity and cooling per hour | Total visible cost per hour |
|---|---|---|---|
| 200 | $4.50 | $0.15 | $4.65 |
| 500 | $1.80 | $0.15 | $1.95 |
| 1,500 | $0.60 | $0.15 | $0.75 |
Check the arithmetic: 0.6 kW x $0.20 = $0.12 per hour, and $0.12 x 1.25 = $0.15 once the cooling uplift is added.
What does electricity and cooling really add?
Electricity is the smallest of the hidden costs of rendering locally, but it is easy to calculate wrongly. Use wall draw, not the GPU board power, and multiply by your own tariff. Cooling matters mostly in a small room, where the machine heat becomes an air-conditioning load.
- Formula: kWh = watts / 1000 x hours, and cost = kWh x tariff. At 600 W and $0.20 per kWh that is $0.12 per hour; at $0.10 it is $0.06, and at $0.35 it is $0.21.
- NVIDIA lists total graphics power of 450 W for the GeForce RTX 4090 and 575 W for the RTX 5090. The whole system draws more, so measure at the wall. Our 600 W figure is an assumption for a single-GPU build, not a specification.
- 600 W of electrical load becomes about 2,050 BTU per hour of heat (600 x 3.412). Whether that costs you money depends on your climate and whether the room is air-conditioned.
- For tariffs, use your own bill or the US EIA Electric Power Monthly, which publishes average prices by state. We did not quote a national average because the table is only published as a spreadsheet.
Why do power delivery and hardware limits cost extra?
High-end GPUs set requirements for the rest of the machine. NVIDIA lists an 850 W minimum system power for the RTX 4090. For the RTX 5090 its page lists 1000 W in the specification table and 850 W in another section, so check your card maker guidance before you buy a power supply.
- RTX 5090: 575 W total graphics power and 32 GB GDDR7, with power connector requirements that the NVIDIA page describes in more than one way. RTX 4090: 450 W and 24 GB GDDR6X.
- A new PSU, a better power cable and a circuit that can carry the load are upgrade costs that only appear when you build or move to a new card.
- VRAM is a hard ceiling. When a scene stops fitting, render time stops being the problem because the job fails or slows down. See the VRAM discussion in our multi-GPU rendering guide and the out-of-memory fixes for heavy comps.
What does a blocked workstation cost you?
When one machine is both your workstation and your render node, every render hour is an hour you cannot use it. If you wait, that time has a price. If you keep working on another machine, blocked time is close to zero. Put your own number on it, because it is often the largest of the hidden costs of rendering locally.
- Formula: blocked cost per render hour = share of render time you are actually blocked x your hourly value.
- Example: if you are blocked for 15% of render hours and your time is worth $30 an hour, blocked time adds $4.50 per render hour.
- Failed and restarted renders cost twice: the lost hours and the power. See our recovery workflow for overnight crashes and our guide to thermal throttling while rendering for slowdowns that quietly stretch jobs.
When does a cloud machine become cheaper than a local one?
A cloud machine becomes cheaper when you render few hours a year. Against the $8.20 per hour iRender publishes for one RTX 4090 machine (October 2026), our model break-even is about 112 render hours a year without blocked time and about 254 with it. Above that, the local machine is cheaper per hour in this model.
| Render hours per year | Local cost per hour, visible only | Local cost per hour, with blocked time ($4.50) | Cloud (1x RTX 4090, published rate) |
|---|---|---|---|
| 200 | $4.65 | $9.15 | $8.20 |
| 500 | $1.95 | $6.45 | $8.20 |
| 1,500 | $0.75 | $5.25 | $8.20 |
- Break-even: $900 / ($8.20 – $0.15) = 112 hours, and $900 / ($8.20 – $0.15 – $4.50) = 254 hours.
- The iRender price page also lists $7.38 per hour for rentals of three hours or more. At that rate the break-even moves to about 124 hours without blocked time and about 330 with it.
- The comparison is per machine hour. It ignores that a cloud job can use many machines at once, and it ignores the cloud side hidden costs: upload and setup time, idle sessions, re-renders and tax. Those are covered in render farm hidden costs.
- Other farms bill in different units, so compare on cost per delivered frame rather than per hour.
How do you measure the hidden costs of rendering locally?
Measure your own numbers instead of trusting a generic calculator. The six steps below take about a month of logging and one plug-in power meter.
- Log your render hours for a month and multiply by twelve.
- Measure wall watts during a render with a plug-in meter.
- Take your tariff from the electricity bill.
- Estimate resale value from current used prices for your card.
- Decide honestly how often you are blocked.
- Put the numbers into the tables above and compare against a cloud rate from a test frame. Our cost estimator guide shows how, and the build-your-own render farm article covers the self-managed option.
The hidden costs of rendering locally are real, but they do not make cloud rendering automatically cheaper. They make the comparison depend on your hours, your tariff and how you use the machine. Measure those three, and the answer is usually clear.
FAQ
Is it cheaper to render locally or in the cloud?
It depends on how many hours a year you render. In our illustrative model, a local RTX 4090 workstation beats a cloud machine at $8.20 per hour above roughly 112 to 254 render hours a year, depending on whether blocked time is counted. Below that, cloud is cheaper per hour.
How much electricity does GPU rendering use?
Use wall draw times hours times your tariff. NVIDIA lists 450 W total graphics power for the RTX 4090 and 575 W for the RTX 5090, and the whole system draws more. At an assumed 600 W and $0.20 per kWh, one hour costs $0.12.
What power supply does an RTX 4090 or RTX 5090 need?
NVIDIA lists an 850 W minimum system power for the RTX 4090. For the RTX 5090 its page lists 1000 W in the specification table and 850 W in another section, so confirm with your card maker before buying.
Does local rendering block my workstation?
Yes, if the same machine renders and runs your software. The cost depends on how long you wait; if you keep working elsewhere it is close to zero. Price it as blocked hours times your hourly value.
How do I compare local cost with a render farm?
Convert both sides to cost per delivered frame: local cost per hour times hours per frame, against the farm billed cost for a test sample divided by the number of frames.
COMMENTS