RPC error InstructionError: Computational budget exceeded

Last updated: November 1, 2026

At a glance

Cause

Every transaction has a compute unit limit. Without an explicit instruction the default is 200,000 per non-builtin instruction and 3,000 per builtin, up to a maximum of 1,400,000. The VM reports the moment it runs out as exceeded CUs meter at BPF instruction, and the runtime reports the result as Computational budget exceeded. Both are the same event. A limit that was fine yesterday can fail today if the path through a program changes, for example a swap routed through more pools.

Solution

1. Find the instruction in the logs: the last Program ... invoke before the failure names the program that ran out.
2. Run a simulation and read unitsConsumed.
3. Add a SetComputeUnitLimit instruction with that number plus a margin of about 10%.
4. Do not set the limit far above what you use: your priority fee is the price multiplied by the limit you request, not by what you consume.

Example

{
  "err": {
    "InstructionError": [
      1,
      "ComputationalBudgetExceeded"
    ]
  },
  "logs": [
    "Program ... failed: exceeded CUs meter at BPF instruction"
  ]
}