I went through a Unity interview at a mid-size studio a while back that I thought I was ready for. I’d practiced the classic engine questions, reviewed the game loop, read through the Unity docs on physics. Then they asked me to explain why two colliders were phasing through each other at high velocity, and I fumbled it. Discrete vs. continuous collision detection. I knew the concept but I hadn’t thought through the real-world symptom.
That’s the thing about game dev interviews. They’re not testing whether you can define things. They’re testing whether you’ve actually shipped a game that broke and had to be fixed.
Here’s what comes up, organized by the areas studios care about most.
The game loop and engine fundamentals
Every interviewer at a studio starts here. The game loop question is really asking: “prove you understand the foundation of all interactive software.”
The standard loop structure is input, update, render. That’s not wrong, but it’s incomplete. A good answer explains fixed vs. variable timesteps, why physics updates typically run at a fixed rate (like 50 or 60 Hz) independent of frame rate, and what delta time is actually doing in the variable timestep case.
Delta time keeps object movement frame-rate independent. A character moving at 5 units per second should cover the same distance whether the frame takes 16ms or 33ms. Multiply speed by delta time, not by 1. This sounds obvious until you see a junior dev’s project where everything speeds up on a 144Hz monitor.
Follow-up questions I’ve seen:
- What’s the difference between
UpdateandFixedUpdatein Unity? (Variable vs. fixed timestep, physics goes in FixedUpdate.) - When would you use
LateUpdate? (Camera follow logic, anything that needs to run after all object positions are settled.) - What happens if FixedUpdate runs too slowly? (Spiral of death. The physics falls behind and the game hitches.)
Physics and math questions
Studios vary a lot here depending on the genre. A physics simulation studio might spend 40 minutes on this. A mobile casual game studio might ask one or two questions and move on.
The collision detection question I fumbled is worth elaborating on. Discrete collision detection checks for overlap at each frame boundary. If an object moves fast enough in a single frame to pass completely through a thin wall, no collision is detected. Continuous collision detection sweeps the object’s path between frames. It’s more expensive, so you typically enable it only for fast-moving objects like bullets or thrown items.
Other questions that come up regularly:
- What’s a Rigidbody and when do you use kinematic mode? Rigidbody hands control to the physics engine. Kinematic mode gives it back to you, via code or animation, while still letting the object participate in collision detection. Platforms you move programmatically should be kinematic.
- Explain quaternions vs. Euler angles. Euler angles are intuitive but suffer from gimbal lock. Quaternions avoid gimbal lock and interpolate cleanly with SLERP. Interviewers don’t necessarily expect you to derive quaternion math, but they want to know you understand why they exist.
- How does raycasting work and what do you use it for? Shooting a ray from a point in a direction, checking what it hits first. Common uses: line of sight checks, shooting projectiles, detecting ground beneath a character.
- How would you implement A* pathfinding? The algorithm itself, open and closed sets, the heuristic function. Interviewers sometimes ask what happens when the heuristic is inadmissible (overestimates) and whether the algorithm still finds the optimal path. (It doesn’t, though it still finds a path.)
Performance and optimization
This is where senior candidates separate from mid-level ones. Anyone can learn the rules. Seniors can tell you which rules matter for which platform and why.
The questions I hear most in this area:
- What is draw call batching and why does it matter? Each draw call has CPU overhead. Combining meshes that share a material reduces draw calls. Static batching bakes combined meshes at build time. Dynamic batching happens at runtime for small meshes. GPU instancing handles many identical objects efficiently.
- Explain object pooling. Instantiating and destroying objects constantly is expensive because of garbage collection. Pooling pre-allocates a set of objects and reuses them. Bullets are the canonical example. You almost always want this for anything that spawns frequently.
- How do you profile a game that’s dropping frames? The answer has to start with “open the profiler, don’t guess.” Common causes are too many draw calls, physics running too slowly, garbage collection spikes, or a hot path in game logic running every frame unnecessarily.
- What’s LOD and when would you implement it? Level of Detail swaps in lower-poly meshes as objects get farther from the camera. It’s a straightforward win on open-world or large-scene games. The follow-up is usually about LOD group settings in Unity or the LOD system in Unreal.
I don’t have data on how mobile-specific questions compare to console-specific ones across studios. My honest impression from talking to people who’ve gone through both is that mobile interviews lean harder on memory budgets and battery usage, while console interviews lean harder on GPU pipeline and render passes. That’s probably not universal.
Rendering and graphics
You don’t need to be a graphics programmer to get through a gameplay engineering interview. But you need enough rendering literacy to not sound lost when the conversation touches shaders or the rendering pipeline.
- What’s the difference between forward and deferred rendering? Forward renders each object once per light. Deferred decouples geometry and lighting passes, making it efficient for many lights. Deferred has higher memory cost and struggles with transparency and MSAA.
- What is a shader and what does it do? A program that runs on the GPU. Vertex shaders transform geometry. Fragment shaders determine pixel color. Compute shaders handle general-purpose GPU work. Interviewers at studios doing any custom visual work will expect you to have written at least a basic shader.
- What is occlusion culling? Not rendering objects that are completely hidden behind other geometry. Unity has built-in baked occlusion culling. Unreal has a hardware occlusion query system. This is a meaningful optimization in dense scenes.
Multiplayer and networked games
Not every studio interview covers multiplayer. If the role is on a multiplayer title, you should expect at least 20 minutes here, and I’d say it’s the hardest section because the concepts are genuinely tricky.
The core problem in networked games is latency. Players are in different locations. Their game states are always slightly out of sync. The challenge is hiding that from the player.
Client-side prediction lets the local player see instant feedback for their own inputs instead of waiting for the server round-trip. The server then either confirms the prediction or sends a correction. If a correction arrives, the client has to reconcile its local state with the authoritative server state, which is called server reconciliation.
Lag compensation handles the situation where a player aims at a target on their screen, but by the time the hit message reaches the server, the target has already moved. The server rewinds to the state that existed when the player fired and checks the hit against that rewound state.
These concepts come from a Valve networking article that’s been required reading in game dev communities for years. If you’re prepping for a multiplayer role, read it.
The Stack Overflow Developer Survey 2024 shows that game developers are among the most likely to work with C++ and C#, which tracks with Unity and Unreal being the dominant engines. Engine-specific fluency matters a lot in studio interviews. Pick one to go deep on. Shallow knowledge of both is almost always worse than real depth in one.
What separates the candidates who get offers? In my experience, it’s not that they know more facts. It’s that they talk about their own games. Real projects with real bugs they had to fix. That’s hard to fabricate, and interviewers know it.