Render Still Noisy With High Samples: Why Adding More Samples Stops Working

Render Still Noisy With High Samples: Why It Happens

Adding samples does help, but with sharply diminishing returns. The textbook Physically Based Rendering shows that Monte Carlo error shrinks in proportion to one over the square root of the sample count. In practice that means going from 256 to 1,024 samples halves the noise, while going from 256 to 512 removes only about 29 percent. If your render is still noisy with high samples, one of the causes below is usually responsible.

Fix 1: Find Which Noise You Have Before Adding Samples

Autodesk’s Arnold tutorial maps each kind of noise to the setting that controls it: alpha channel noise to Camera (AA), indirect diffuse to Diffuse, direct specular and shadows to Light, and indirect specular to Specular, with Transmission, SSS and Volume handled separately. It recommends rendering AOVs to find the source. The same documentation explains that the real ray count is the square of the setting, so Camera (AA) 4 means 16 rays per pixel, and that raising Camera (AA) multiplies the cost of every other sample type. Camera (AA) is the only fix for motion blur and depth of field noise, so raise it for those and raise the other types individually for everything else.

Do this before you accept a render still noisy with high samples as normal. Cycles has no equivalent table in the pages we read. Our suggestion, not a documented rule, is to compare the direct and indirect passes of one frame to see which one carries the grain before you change anything.

Fix 2: Clamp Fireflies Instead of Adding Samples

Fireflies are isolated bright pixels caused by rare, very bright light paths, and they average out slowly. Blender’s Light Paths documentation lists Clamp Direct, which limits what a not-yet-bounced ray sample can add to a pixel, and Clamp Indirect, which does the same for rays that bounced several times. A value of 0.0 disables clamping. The manual does not give a recommended number, so our suggestion is to test values on one frame and raise the clamp until the highlights recover. Fireflies are one common reason for a render still noisy with high samples that does not improve.

Autodesk’s Arnold documentation warns that clamping affects the final dynamic range of the render. It also names other firefly sources: lights placed very close to geometry, where a Light Decay Filter with a very low Near Start value helps, non-energy-conserving shaders, and geometry outside the view such as objects behind the camera.

Fix 3: Remove Paths That Are Hard to Sample

In Cycles, Blender’s manual says Filter Glossy blurs glossy reflections after blurry bounces, reducing noise at some cost to accuracy, and suggests 1.0 as a starting value. The same page notes that reflective and refractive caustics can be switched off because they are often too noisy. For scenes with many lights, the 4.0 sampling page describes the Light Tree as a way to sample lights more effectively using distance and estimated intensity, and Light Threshold as a way to terminate samples from lights below a contribution threshold. Light Threshold trades some accuracy for speed, so check the result before adopting it. Hard-to-sample paths are a second reason for a render still noisy with high samples.

Fix 4: Use Adaptive Sampling With a Noise Target

A render still noisy with high samples often needs a target instead of more samples. Set a quality target and let the engine spend samples where they are needed.

• Cycles: Noise Threshold is the error level at which a pixel stops receiving samples. The manual gives typical values from 0.1 down to 0.001, and a value of 0 lets Cycles estimate one from the sample count. Min Samples sets the minimum before adaptive sampling applies, and the manual says you can combine a noise target with a high sample count.
• Redshift: Min and Max Samples set the range of primary rays per pixel, and Redshift adds rays only where noise remains. Maxon documents the Adaptive Error Threshold with a default of 0.01 and about 0.003 suggested for production, where a lower value shoots more rays. With Automatic Sampling on, the threshold is the only control that affects sample counts.
• Arnold: Autodesk documents that adaptive sampling gives more camera samples to higher-variance pixels when it is enabled, but the page we read does not name the threshold parameters, so check your version’s help.

Redshift also has an option to automatically reduce samples of other effects, which divides secondary effect sample counts by the primary count. The documentation does not recommend it. If your global illumination or reflections stay grainy after raising primary samples, check whether that option is on.

Fix 5: Check the Seed in Animation, Then Denoise

In animation, a render still noisy with high samples looks worse because the grain moves from frame to frame. In Cycles, the Seed value gives different noise patterns, and the Animated Seed option changes the seed for each frame. The manual says enabling it is a good idea for animations because a varying noise pattern is less noticeable than a fixed one. To see the effect on your own scene, render two neighboring frames both ways and compare them. For temporal grain problems in Cycles, see our guide to Cycles animation temporal flicker and noise. Once the noise is reduced as far as your budget allows, a denoiser handles the rest, and we explain how to avoid losing detail in our guide to a denoiser losing detail in animation. If edges rather than grain are the problem, see jagged edges in animation renders.

Before you send a render still noisy with high samples to a farm, fix the causes above. If the sample counts you need push frame times beyond what your machine can deliver, compare options in our top 10 render farms comparison. For the wider set of quality problems, see the guide to fixing render noise and flicker in animation.

What We Could Not Verify

We found no official recommendation for a specific clamp value, no official Cycles sample count for any scene type, and no readable current V-Ray documentation. Arnold’s adaptive threshold parameter names and defaults were not confirmed on the pages we read. Treat any number you see in forum posts about a render still noisy with high samples as a starting point for your own test render.

Sources

• Blender Manual 4.0, Cycles Light Paths: docs.blender.org (checked 2 November 2026; current manual versions could not be read)
• Blender Manual 4.0, Cycles Sampling: docs.blender.org (checked 2 November 2026)
• Blender Manual 3.6, Cycles Sampling: docs.blender.org (checked 2 November 2026)
• Autodesk, Arnold for Cinema 4D, Removing Noise: help.autodesk.com (checked 2 November 2026)
• Autodesk, Arnold for Cinema 4D, What Is Sampling: help.autodesk.com (checked 2 November 2026)
• Autodesk, Arnold for 3ds Max, Sampling and Ray Depth: help.autodesk.com (checked 2 November 2026)
• Maxon, Redshift for Cinema 4D, Unified Sampling: help.maxon.net (checked 2 November 2026)
• Pharr, Jakob, Humphreys, Physically Based Rendering, 3rd edition, Monte Carlo Integration (chapter overview; the inverse-square-root error rate is the standard Monte Carlo result): pbr-book.org

FAQ