Unity VFX & Particles

Retrieval Sources

Key Guidelines

Shuriken Particle System Modules

The Shuriken system is the ParticleSystem component: a fixed pipeline of toggleable modules, each simulated on the CPU (with optional Burst/Job System acceleration — see below) and configured either in the Inspector or through the matching module struct in script. The Main module (always on) sets system-wide defaults — duration, loop, startLifetime, startSpeed, startSize, simulationSpace, maxParticles. Emission controls spawn rate (rateOverTime for continuous emission independent of motion, rateOverDistance for motion-driven emission like tire dust) plus discrete bursts. Shape defines the emission volume/surface (cone, sphere, box, mesh, edge, etc.) and therefore the initial position and direction of new particles. From there, "over Lifetime" and "by Speed" modules (Color, Size, Rotation, Velocity, Limit Velocity, Force, Noise) apply curves or gradients evaluated against a particle's normalized age or current speed — the "varying properties over time" authoring pattern (curve, random-between-two-curves, or random-between-two-constants) is shared across nearly all of them. Every module is independently toggled per system, so a system pays simulation cost only for the modules it has enabled.

using UnityEngine;

public class ConfigureBurstEffect : MonoBehaviour
{
    [SerializeField] private ParticleSystem ps;

    void Start()
    {
        var main = ps.main;
        main.startLifetime = 2f;
        main.startSpeed = 5f;
        main.maxParticles = 500;

        var emission = ps.emission;
        emission.rateOverTime = 0f; // no continuous emission
        emission.SetBursts(new[]
        {
            new ParticleSystem.Burst(0f, 30, 50, cycleCount: 1, repeatInterval: 0f)
        });

        var shape = ps.shape;
        shape.shapeType = ParticleSystemShapeType.Sphere;
        shape.radius = 0.5f;

        var colorOverLifetime = ps.colorOverLifetime;
        colorOverLifetime.enabled = true;
        var grad = new Gradient();
        grad.SetKeys(
            new[] { new GradientColorKey(Color.white, 0f), new GradientColorKey(Color.red, 1f) },
            new[] { new GradientAlphaKey(1f, 0f), new GradientAlphaKey(0f, 1f) });
        colorOverLifetime.color = new ParticleSystem.MinMaxGradient(grad);

        ps.Play();
    }
}

Sub-Emitters & Bursts

Bursts (part of the Emission module) fire a batch of particles at a specific time — or, with cycleCount/repeatInterval, repeatedly — making them the tool for punctual effects like muzzle flashes or explosions rather than driving them through a sustained rateOverTime. Sub Emitters attach a second ParticleSystem to specific lifecycle events of the parent's particles — Birth, Collision, Death, Trigger, or Manual (fired from script) — so a single system can spawn a layered secondary effect, e.g. a smoke trail on birth and a burst of sparks on death. Sub-emitters inherit spatial context from the parent particle they're attached to but are configured as independent ParticleSystem assets/children, so their own modules (Emission, Shape, etc.) apply normally once triggered.

using UnityEngine;

public class ConfigureSubEmitter : MonoBehaviour
{
    [SerializeField] private ParticleSystem parentPs;
    [SerializeField] private ParticleSystem deathBurstPs;

    void Start()
    {
        var subEmitters = parentPs.subEmitters;
        subEmitters.enabled = true;
        // Attach deathBurstPs to fire when parent particles die.
        subEmitters.AddSubEmitter(deathBurstPs, ParticleSystemSubEmitterType.Death,
            ParticleSystemSubEmitterProperties.InheritEverything);
    }
}

Particle Collision & Triggers

The Collision module has two distinct modes with different cost profiles. World mode collides particles against real Colliders in the Scene — flexible (any layer, dynamic obstacles) but the most expensive option, with a quality setting (High/Medium/Low) trading accuracy for CPU cost. Planes mode instead collides against a manually authored list of infinite planes (defined by empty-GameObject transforms) that need no Collider component at all — this is explicitly documented as lower processor overhead and is the right default for simple floors/walls. sendCollisionMessages opts into OnParticleCollision callbacks on scripts for gameplay reactions (spawning decals, damage, sound). The separate Trigger module is cheaper still: it defines a set of collider regions and fires OnParticleTrigger when particles enter/exit/inside/outside them, without running full collision physics (bounce, dampen, lifetime loss) — use it purely to detect particle presence in a volume, not to make particles physically bounce.

using UnityEngine;

public class ParticleCollisionHandler : MonoBehaviour
{
    private ParticleSystem ps;
    private readonly System.Collections.Generic.List<ParticleCollisionEvent> collisionEvents = new();

    void Awake()
    {
        ps = GetComponent<ParticleSystem>();
        var collision = ps.collision;
        collision.enabled = true;
        collision.type = ParticleSystemCollisionType.World;
        collision.mode = ParticleSystemCollisionMode.Collision3D;
        collision.quality = ParticleSystemCollisionQuality.Medium;
        collision.bounce = 0.3f;
        collision.sendCollisionMessages = true;
    }

    void OnParticleCollision(GameObject other)
    {
        int count = ps.GetCollisionEvents(other, collisionEvents);
        for (int i = 0; i < count; i++)
        {
            // e.g. spawn a decal/impact effect at collisionEvents[i].intersection
        }
    }
}

VFX Graph (GPU-Simulated)

VFX Graph is a separate package built around the VisualEffectAsset (a node-graph you author in a dedicated graph editor) and the VisualEffect component that plays an instance of it in a scene. Unlike Shuriken's fixed module pipeline, simulation logic itself is authored as a graph and runs on the GPU via compute shaders, which is what lets it scale to millions of particles versus Shuriken's thousands (per the docs' own comparison table). It only supports the Universal and High Definition Render Pipelines (not the Built-in Render Pipeline), and in HDRP it additionally exposes camera color/depth buffer sampling that Shuriken has no equivalent for. Script interaction is event-driven rather than direct-particle-access: you expose named properties on the graph and drive them from C# via SetFloat/SetVector3/SetTexture/SetGraphicsBuffer/SetGradient, and send named events (with optional attached VFXEventAttribute payload data) via SendEvent instead of reading/writing individual particles.

using UnityEngine;
using UnityEngine.VFX;

public class DriveVfxGraph : MonoBehaviour
{
    [SerializeField] private VisualEffect vfx;
    private static readonly int SpawnRateID = Shader.PropertyToID("SpawnRate");
    private static readonly int ColorID = Shader.PropertyToID("Color");

    void Start()
    {
        vfx.SetFloat(SpawnRateID, 200f);
        vfx.SetVector4(ColorID, Color.cyan);
        vfx.playRate = 1f;
        vfx.Play();
    }

    public void TriggerImpact(Vector3 worldPos)
    {
        var attr = vfx.CreateVFXEventAttribute();
        attr.SetVector3("position", worldPos);
        vfx.SendEvent("OnImpact", attr);
    }
}

Choosing Between Them

The docs' own comparison table (ChoosingYourParticleSystem.html) is the canonical decision reference: Shuriken supports Built-in/URP/HDRP and full C# read/write access to every particle plus physics-engine collision, and is authored via Inspector modules; VFX Graph supports only URP/HDRP, scales to millions of particles via GPU simulation, and is authored via a visual node graph with an event/exposed-property scripting surface rather than per-particle access. Note that on platforms that support compute shaders, Unity explicitly allows using both solutions simultaneously in one project — this isn't an either/or choice at the project level, only per-effect. Default to Shuriken for gameplay-driven effects that need per-particle script logic (collision response, custom job-system behaviors), for projects still on the Built-in Render Pipeline, or for moderate particle counts. Reach for VFX Graph when an effect needs particle counts CPU simulation can't sustain, when you need HDRP camera-buffer sampling (refraction, depth-aware effects), or when the effect is authored primarily by a technical artist working in a graph rather than by a programmer.

Performance / Overdraw

Shuriken's CPU cost scales with maxParticles, active module count, and the Collision module's quality/type setting — profile with the Profiler's CPU timeline, not guesswork. The C# Job System integration (IJobParticleSystem, IJobParticleSystemParallelFor, IJobParticleSystemParallelForBatch) lets custom per-particle behavior run Burst-compiled across worker threads, which the docs call out as strictly faster than the main-thread GetParticles()/SetParticles() pair — prefer it for any per-frame custom particle logic at scale. Independently of CPU sim cost, GPU instancing (Manual/particle-system-optimization.html, Built-in RP) renders particles as instanced Meshes instead of billboards, which is a rendering-side optimization orthogonal to simulation cost. The Lights module is called out explicitly in the docs as high performance cost — "especially in Forward Rendering mode," worse still with shadow-casting lights — which is why it has a dedicated Maximum Lights safety cap to prevent an emission-rate tweak from spawning thousands of real-time lights. VFX Graph cost shows up as GPU time (compute simulation + fill-rate for translucent particle rendering) rather than CPU time, so profile it with the GPU frame-timing tools/Frame Debugger, not the CPU profiler.

using Unity.Collections;
using Unity.Jobs;
using UnityEngine;
using UnityEngine.ParticleSystemJobs;

public struct FadeAndDriftJob : IJobParticleSystemParallelFor
{
    public float deltaTime;

    public void Execute(ParticleSystemJobData jobData, int index)
    {
        var velocities = jobData.velocities;
        velocities[index] = new Vector3(
            velocities[index].x,
            velocities[index].y - 0.5f * deltaTime, // extra gravity, Burst-compiled
            velocities[index].z);
    }
}

public class JobDrivenParticles : MonoBehaviour
{
    [SerializeField] private ParticleSystem ps;

    void LateUpdate()
    {
        new FadeAndDriftJob { deltaTime = Time.deltaTime }
            .Schedule(ps)
            .Complete();
    }
}

Common Mistakes

Quick Reference

Advanced Notes

Mobile GPUs are bandwidth- and fill-rate-constrained far more than desktop/console GPUs, which makes particle-heavy effects one of the fastest ways to tank frame time on mobile even when CPU simulation cost looks cheap in the Profiler. Most particle materials are alpha-blended and unlit-additive, which means every overlapping particle quad is a full-screen-resolution shader invocation stacked on top of whatever's already been drawn — large, overlapping, screen-filling particle systems (explosions, smoke columns, weather) create severe overdraw where the same pixel is shaded many times per frame. On tile-based mobile GPUs this is punished especially hard because blended geometry can't use early-Z/hidden-surface removal the way opaque geometry can, so overdraw cost is close to linear in the number of overlapping translucent layers.

Practical mitigations, in rough order of impact: keep individual particle quads small relative to the screen and cap maxParticles aggressively for mobile-targeted systems, since a handful of huge, screen-filling particles cost far more fill-rate than many small ones covering the same visual area; prefer fewer, larger bursts of short-lived particles over sustained high rateOverTime emission so the number of simultaneously alive (and therefore simultaneously overdrawing) particles stays low; avoid stacking multiple semi-transparent particle systems in the same screen-space region (e.g. layering separate smoke, embers, and glow systems on one explosion) — collapse layers into fewer systems or a single custom shader where possible; disable or heavily cap the Lights module on mobile, since real-time per-particle lights are explicitly documented as high-cost, worse in Forward Rendering, and worst with shadows enabled, all of which are common mobile constraints already; use the World/Planes Collision-mode distinction to avoid physics-query overhead on lower-end CPUs, defaulting to Planes wherever the interaction is against static level geometry; and use the Texture Sheet Animation module or lower-resolution/simpler particle textures rather than driving visual complexity purely through overlapping particle count. For VFX Graph specifically, remember it requires compute shader support — verify target mobile hardware/API actually supports it before committing an effect to it, since a compute-shader-driven graph that silently fails to render on a subset of Android/iOS devices is a common, hard-to-catch regression. Regardless of which system you use, validate overdraw hotspots empirically with the Rendering Debugger's overdraw view (or a platform-specific GPU profiler/frame capture on device) rather than assuming from particle count alone — actual fill-rate cost depends on final on-screen size and blend-layer depth, not raw particle count.