November 4, 2026 Linh Nguyen

Low GPU Usage When Rendering in Blender Cycles (2026): What the Meter Means and How to Fix It

Quick answerWhy is GPU usage low in Cycles?
Checked Nov 2026

Low GPU usage during a Cycles render usually has one of four causes: the GPU is not enabled in Preferences › System, the meter is showing the 3D engine rather than compute, the CPU-side steps of each frame (scene sync, BVH build, saving) leave the GPU idle, or several GPUs share one small frame. Check the device settings first, read utilization from nvidia-smi, and for animation run one Blender instance per GPU.

  • Device: Preferences › System: OptiX, all GPUs ticked; Render › GPU Compute.
  • Meter: nvidia-smi shows the share of time a GPU kernel was running.
  • Between frames: CPU steps idle the GPU; Persistent Data helps animation.
  • Several GPUs: one Blender per GPU keeps each card busy.

A GPU meter at 40% during a render looks like wasted money, but it is not always a problem. Cycles spends part of every frame on the CPU, and some monitoring tools do not show compute work at all.

This guide explains what the utilization number means, which parts of a frame run on the CPU, how to check that every GPU is really rendering, and what actually raises throughput on a multi-GPU machine.

Checked against the Blender Manual (GPU Rendering, Performance, Command Line Rendering) and NVIDIA nvidia-smi documentation, November 2026.

1. First check: is every GPU enabled?

Cycles only uses the devices ticked in Preferences › System › Cycles Render Devices, and the scene must be set to GPU Compute in Render Properties (Blender Manual: GPU Rendering). On NVIDIA RTX cards choose OptiX and confirm every GPU is ticked. These settings live in the user preferences of each machine, not in the .blend, so check them again on a new computer.

iRender
Two switches decide which GPUs Cycles uses
Schematic
This computer: Preferences › SystemCycles Render DevicesNoneCUDAOptiXHIPNVIDIA GeForce RTX 5090 (GPU 0)NVIDIA GeForce RTX 5090 (GPU 1)NVIDIA GeForce RTX 5090 (GPU 2) ← not tickedNVIDIA GeForce RTX 5090 (GPU 3)The .blend: Render PropertiesDeviceGPU Computesaved with the filenvidia-smi while renderingGPU 098%GPU 197%GPU 20%GPU 398%Device choice is saved in the .blend, but the ticked GPUs are a preference of each computerBlender Manual: GPU Rendering
per computerper .blendidle GPU
Example readings, not measured

A GPU that is not ticked in Preferences sits at 0% for the whole render, even though the file says GPU Compute. On a new machine, open Preferences first and tick every card.

2. Reading the meter correctly

NVIDIA defines GPU utilization in nvidia-smi as the percent of time over the sample period during which one or more kernels was executing (NVIDIA: nvidia-smi documentation). It is a time measure, not a measure of how much of the chip is working. Windows Task Manager splits the GPU into several engine graphs, and the default 3D graph may show little while Cycles is busy; switch a graph to the compute or CUDA engine, or run nvidia-smi -l 1 in a terminal.

iRender
What the GPU utilization number measures
Schematic
One sample period of nvidia-smitime →GPU A100%SMs in use while a kernel runs:GPU B40%SMs in use while a kernel runs:GPU A shows 100% while using a few SMs; GPU B shows 40% while fully busy part of the timeUtilization is the share of time a kernel was running, not how much of the GPU was doing workNVIDIA: nvidia-smi documentation (utilization.gpu)
GPU AGPU B
Illustrative: real GPUs have many more SMs

So a high number does not prove the card is working hard, and a dip does not prove it is idle for long. To watch every card at once, run this in PowerShell while Blender renders:

iRender
Watch every GPU while Blender renders
PowerShell
1# Refresh every second; press Ctrl+C to stop
2nvidia-smi --query-gpu=index,utilization.gpu,memory.used,memory.total `
3 --format=csv -l 1
One row per GPU: index, utilization, memory used and totalNVIDIA: nvidia-smi documentation

3. The CPU part of every frame

Before sampling, Blender syncs the scene, builds the acceleration structure and loads textures; after sampling it denoises, composites and writes the file. The CPU and disk steps leave the GPU idle, so on simple frames average usage looks low even when everything is correct.

iRender
Why the GPU meter dips during a Cycles render
Schematic
One frame, one Blender instance (schematic, not to scale)Scene syncCPUBVH buildCPUTexturesCPU / diskPath tracingGPUDenoiseGPUSave fileCPU / diskGPU busy:Simple frames: fixed CPU steps are a large share of the frame, so average GPU usage looks low.Heavy frames: path tracing dominates and usage stays high. Running one Blender per GPU overlaps the CPU steps.CPU steps in each frame leave the GPU idle; one Blender per GPU overlaps themBlender Manual: Performance; Command Line Rendering
GPU workdenoiseCPU or disk
Schematic; phases from the Blender Manual, Performance

For animation, Persistent Data keeps render data in memory after rendering for faster re-renders and animation renders, at the cost of extra memory (Blender Manual: Performance). It helps most when the scene changes little between frames.

4. Several GPUs: keep each card busy

One Blender process rendering one frame on several GPUs shares that frame between them, and on light frames the CPU steps and the split itself limit scaling. Running one Blender instance per GPU on different frames, started from the command line (Blender Manual: Command Line Rendering), overlaps the CPU steps of one instance with GPU work in another. Our guide to multi-GPU rendering in Cycles shows the setup; if usage stalls near the memory limit, see Cycles GPU memory.

FixesBlender 5.x + Cycles

Low GPU usage: causes and first checks

Go down the list in order; the first rows are the most common.

What you seeLikely causeFirst check
0% on one or more cardsGPU not enabled, or render device is CPUPreferences › System: OptiX or CUDA with every GPU ticked; Render › Device: GPU Compute
Usage looks low in Task ManagerGraph shows the 3D engine, not computeSwitch a graph to the compute/CUDA engine, or use nvidia-smi
Usage drops between framesScene sync, BVH and saving run on the CPUPersistent Data for animation; faster storage for outputs
Extra GPUs add little speedSmall frames do not fill several GPUsOne Blender instance per GPU on separate frames
Usage spikes then stallsTextures loading or memory near the limitCheck peak memory; texture cache or smaller textures
RuleMeasure frames per hour for the whole sequence, not the GPU meter.

A GPU usage checklist

  1. Preferences: OptiX selected and every GPU ticked on this machine
  2. Scene: Render › Device set to GPU Compute
  3. Meter: nvidia-smi or the compute graph, not the 3D graph
  4. Animation: Persistent Data tried on a short range
  5. Multi-GPU: one Blender instance per GPU for sequences
  6. Measure: frames per hour over a test range, before and after each change

For paths, caches and other scene checks, see our 10-point Blender Cycles pre-flight checklist.

On a managed farm, device setup is handled by the provider. On an IaaS server you set the Cycles devices in Preferences yourself and can watch nvidia-smi while you test, so you see exactly how each GPU is used.

Checking GPU usage on an iRender server

iRender
How iRender works: your computer controls your own multi-GPU server
IaaS render farm
Your PC or MacNo powerful GPU neededRemote desktopLive screen1:1 like your PCYOUR OWN GPU SERVERbare-metal · not sharedGPUGPUGPUGPUGPUGPUGPUGPU1, 2, 4 or 8 × RTX 5090 (32 GB) or RTX 4090 (24 GB)Threadripper PRO · 256 GB ECC RAM · 2 TB NVMe SSDFull admin access · your setup is saved between sessions1Register, top upFirst $50+ top-up doubled2Upload your projectiRender Drive, free3Start and connectRemote desktop, any OS4Render, shut downBilling stops, per minute✓ Your workstation, just more GPUs: no queue to wait, no farm plugin, no job submission, no settings changed
Your mouse and keyboardServer screen streamed back
Schematic, not to scale
  1. Upload the project folder to iRender Drive before you start the server; uploading is free and does not use rental time.
  2. Start a server with RTX 5090 (32 GB) or RTX 4090 (24 GB) GPUs and connect by remote desktop.
  3. Tick every GPU in Preferences › System, run nvidia-smi, and render a short test range.
  4. Render the sequence one Blender per GPU, then shut down; multi-GPU servers are billed per minute.

Frequently asked questions

1. Why is my GPU usage low when rendering in Blender?

Usually the GPU is not enabled in Preferences › System, the meter shows the 3D engine instead of compute, or the CPU steps of each frame leave the GPU idle.

2. What does GPU utilization in nvidia-smi mean?

NVIDIA defines it as the percent of time over the sample period during which a kernel was running on the GPU. It does not measure how much of the chip is in use.

3. Why does Task Manager show low GPU usage in Cycles?

Its default graph is the 3D engine. Switch a graph to the compute or CUDA engine, or use nvidia-smi.

4. Does Persistent Data speed up animation?

The Manual says it keeps render data in memory for faster re-renders and animation renders, using more memory.

5. Why do more GPUs not make my frame faster?

On light frames, the CPU steps and splitting one frame limit scaling. Render separate frames with one Blender instance per GPU.

Render your Blender projects on iRender

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. You install Blender once; 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 Blender 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.

Accelerate your production pipeline with high-performance IaaS render farm services and dedicated bare-metal workstations powered by up to 8x NVIDIA GeForce RTX 5090 (32GB GDDR7), high-frequency AMD Ryzen™ Threadripper™ PRO processors, 256GB ECC RAM, and dedicated 2TB NVMe SSD storage tailored for your pipeline:
, , ,

Linh Nguyen

Linh Nguyen works in customer support at iRender, helping artists render Blender and Houdini projects on multi-GPU servers. She has written more than 300 articles for the iRender blog, covering Blender, Houdini and practical GPU rendering.
Contact

iRENDER GPU

Render Farm Pricing
Quick Start
Multi-GPU Benchmarks
iRender GPU App
Blog
Documentation
The Render Farm
The Company
Terms & Conditions

INTEGRATIONS

Autodesk Maya
Blender
Cinema 4D
Houdini
Karma XPU
Lumion
KeyShot
Unreal Engine
Twinmotion
Redshift
Octane
And many more…

iRENDER TEAM

Elite support 7 days a week
SATURDAY – SUNDAY: 6:00 AM – 11:59 PM
(UTC+7)
Hotline: (+84) 912-785-500
Skype: iRender Support
Email: [email protected]
Address 1: 68 Circular Road #02-01, 049422, Singapore.
Address 2: No.22 Thanh Cong Street, Hanoi, Vietnam.

Contact
이 페이지는 한국어로도 제공됩니다. 한국어로 보기 →