Last Update：04/02/2026

What is a GPU-Based Influence map?

Why is TCAT faster than EQS?

Because TCAT uses a GPU-based influence map.

What is an influence map?

An influence map is a data structure that divides the game world into a 2D grid and stores a numeric value in each cell.

The rule that determines what value gets written into each cell is defined by the developer.

<aside> 💡

The value stored in each cell is called influence.

</aside>

Imagine a level containing multiple friendly AI units.

You define the following rule:

“Cells with a higher concentration of friendly AI should have higher influence values.”

Based on this rule, the influence map would look like the example below.

Cells with denser friendly AI clusters store higher values (brighter green indicates higher influence).

Using this influence map, you can implement a wide range of AI behaviors, such as:

• Enemy AI that avoids areas densely populated by friendly units
• Monster AI that charges directly into areas with many friendly units
• Enemy AI that drops meteors on the most crowded friendly positions
• Friendly AI that tries to stay close to one another
• And many more

In other words, an influence map acts as a battlefield situation board for AI decision-making.

Instead of performing complex calculations every frame, AI can make decisions quickly by simply querying the influence map.

What is a GPU-base influence map?

So far, we’ve looked at what an influence map is.

Now let’s talk about what makes it GPU-based.

As mentioned earlier, an influence map is a 2D grid that stores values per cell.

Internally, this is typically represented as a 2D array.

Writing values into every cell of this array requires iterating over the entire grid.

For example, a 750 × 750 influence map contains 562,500 cells.

Iterating over all of these cells on the CPU can quickly become a major performance bottleneck.

To avoid this, TCAT writes influence values on the GPU instead of the CPU.

GPUs are designed for massive parallelism, allowing influence values to be computed and written to many cells simultaneously.

This means that updating a 750 × 750 influence map can be handled in what is effectively a single parallel pass, rather than hundreds of thousands of sequential CPU iterations.

(While it’s not literally a single operation, it is still far faster than a CPU-based approach.)

By combining influence maps with GPU computation, TCAT enables a highly efficient and scalable AI system.

<aside> 💡

For a deeper theoretical discussion of influence maps, see this GDC talk: link

</aside>