Skip to main content
TLDR:
  • Every Solana instruction gets a default compute unit (CU) allocation. Non-builtin programs get 200,000 CUs; builtin programs (System, Stake, Vote) get 3,000 CUs.
  • The transaction-level cap is 1,400,000 CUs.
  • Priority fees are calculated from the requested CU limit, not actual usage — overspending on CUs wastes money.
  • Use simulateTransaction to measure actual CU usage, then set SetComputeUnitLimit to the measured value + 10% margin.

​
How compute units work

Every transaction on Solana consumes compute units (CUs). CUs are the Solana equivalent of gas — they measure the computational cost of executing instructions. The runtime enforces a CU budget per transaction; if the transaction exceeds it, execution aborts with ComputationalBudgetExceeded.

​
Default allocation

When you don’t include a SetComputeUnitLimit instruction, the default budget is calculated based on instruction types: 
default_cu_limit = (num_builtin_instructions × 3,000)
                 + (num_non_builtin_instructions × 200,000)
  • Builtin instructions (System Program, Stake, Vote, Compute Budget) — 3,000 CUs each per SIMD-0170
  • Non-builtin instructions (your program, Token Program, any SBF-deployed program) — 200,000 CUs each
  • The result is clamped to the transaction cap of 1,400,000 CUs
A typical transaction with 1 SystemProgram transfer + 1 ComputeBudget instruction + 1 SPL Token instruction gets: (2 × 3,000) + (1 × 200,000) = 206,000 CUs as the default budget.

​
Why defaults are wasteful

Consider a simple SOL transfer with a priority fee:
  • 2 builtin instructions (transfer + SetComputeUnitPrice) = 6,000 CUs allocated
  • But the actual transfer uses ~450 CUs
If you set a priority fee of 10,000 micro-lamports per CU, the fee on 6,000 CUs is 60,000,000 micro-lamports (0.06 lamports). If you set SetComputeUnitLimit to 500 CUs (enough for the transfer + margin), the fee drops to 5,000,000 micro-lamports — a 12x reduction.
The priority fee is determined by the requested compute unit limit, not the actual number of compute units used. If you set a CU limit that is too high or use the default, you pay for unused compute.

​
Get your own node endpoint today

Start for free and get your app to production levels immediately. No credit card required.You can sign up with your GitHub, X, Google, or Microsoft account.

​
Compute Budget Program instructions

The Compute Budget Program (ComputeBudget111111111111111111111111111111) has 4 instructions:
InstructionParameterTypeDescription
SetComputeUnitLimitunitsu32Max CUs the transaction may consume
SetComputeUnitPricemicro_lamportsu64CU price in micro-lamports (priority fee)
RequestHeapFramebytesu32Heap size for each program (32K–256K, multiple of 1024)
SetLoadedAccountsDataSizeLimitbytesu32Max total bytes of account data the transaction may load
Rules:
  • Only one of each instruction variant is allowed per transaction. Duplicates cause a DuplicateInstruction error and the transaction fails.
  • SetComputeUnitLimit accepts any u32 but is clamped to 1,400,000.
  • SetComputeUnitPrice accepts any u64 (0 to u64::MAX).

​
Optimizing your compute budget

The optimal workflow is:
  1. Simulate your transaction to measure actual CU consumption
  2. Add a margin (10–20%) to the simulated value
  3. Set the CU limit with SetComputeUnitLimit
  4. Set the CU price with SetComputeUnitPrice based on recent priority fees

​
Step 1: Simulate and measure

import {
  Connection,
  Keypair,
  SystemProgram,
  TransactionMessage,
  VersionedTransaction,
  ComputeBudgetProgram,
} from "@solana/web3.js";
import bs58 from "bs58";
import "dotenv/config";

const connection = new Connection(process.env.SOLANA_RPC!);
const payer = Keypair.fromSecretKey(bs58.decode(process.env.PRIVATE_KEY!));

async function optimizeComputeBudget() {
  const latestBlockhash = await connection.getLatestBlockhash();

  // Your transaction instructions (without CU limit set yet)
  const instructions = [
    ComputeBudgetProgram.setComputeUnitPrice({ microLamports: 10_000 }),
    SystemProgram.transfer({
      fromPubkey: payer.publicKey,
      toPubkey: payer.publicKey,
      lamports: 1000,
    }),
  ];

  // Build and simulate WITHOUT SetComputeUnitLimit
  const message = new TransactionMessage({
    payerKey: payer.publicKey,
    recentBlockhash: latestBlockhash.blockhash,
    instructions,
  }).compileToV0Message();

  const transaction = new VersionedTransaction(message);
  transaction.sign([payer]);

  const simulation = await connection.simulateTransaction(transaction);
  const unitsConsumed = simulation.value.unitsConsumed || 0;
  console.log("Simulated CUs:", unitsConsumed);

  // Add 10% margin
  const cuLimit = Math.ceil(unitsConsumed * 1.1);
  console.log("Setting CU limit to:", cuLimit);

  // Rebuild with the optimized CU limit
  const optimizedInstructions = [
    ComputeBudgetProgram.setComputeUnitLimit({ units: cuLimit }),
    ComputeBudgetProgram.setComputeUnitPrice({ microLamports: 10_000 }),
    SystemProgram.transfer({
      fromPubkey: payer.publicKey,
      toPubkey: payer.publicKey,
      lamports: 1000,
    }),
  ];

  const optimizedMessage = new TransactionMessage({
    payerKey: payer.publicKey,
    recentBlockhash: latestBlockhash.blockhash,
    instructions: optimizedInstructions,
  }).compileToV0Message();

  const optimizedTx = new VersionedTransaction(optimizedMessage);
  optimizedTx.sign([payer]);

  const sig = await connection.sendTransaction(optimizedTx);
  console.log("Transaction:", sig);
}

optimizeComputeBudget();

​
Block-level limits

The Solana scheduler imposes block-level CU caps that affect how many transactions fit per block:
  • 48,000,000 CUs per block — total compute budget across all transactions
  • 12,000,000 CUs per account per block — write-lock limit per account
These block limits are why tight CU budgets matter: a transaction requesting 200,000 CUs that only uses 500 blocks other transactions from fitting in the same block, reducing overall throughput.

​
Priority fee calculation

The total priority fee for a transaction: 
priority_fee = ceil(SetComputeUnitPrice × SetComputeUnitLimit / 1,000,000)
Where SetComputeUnitPrice is in micro-lamports and the result is in lamports. The division uses ceiling — non-integer results round up. The fee is always based on the requested limit, not actual usage.

​
Additional resources

Last modified on April 16, 2026