All metrics › Throughput & Block Production

Compute Units per Block

Latest value: 17.8 million CU on 3 November 2026.

30-day average: 21.6 million CU, down 24.9% from a year earlier (28.7 million CU).

Highest daily value since January 2025: 51 million CU on 30 January 2026.

The default network series of the Explorer (total_cu_avg). Download the data (CSV)

Compute units (CUs) consumed per block. CUs measure program execution cost in the Solana runtime, where each CU roughly corresponds to one BPF instruction. Each block has a compute limit: 48M CU until April 2025, then 50M, 60M (July 2025) and 100M (July 2026, SIMD-0286). From August 2026 the limit falls with each slot-time step (SIMD-0525) to 87.5M, 75M, 62.5M and 50M at 200 ms, so compute per second stays near 250M CU and per-block values fall when slots get shorter. Higher values mean denser, more computation-heavy blocks. Vote and Non-vote split each block by transaction type, and Successful and Failed by transaction outcome; the % view shows the share of the type's compute units. These views come from Dune block data.

Open the chart in the Explorer

Area
Throughput & Block Production
Frequency
Daily
Unit
Compute units
Split by
Network, Client, Scheduler, Validator
Source
Firedancer

Monthly averages

MonthCompute Units per Block
November 2026 (to 3 November)15.8 million CU
October 202623.1 million CU
September 202632.8 million CU
August 202640.9 million CU
July 202634.4 million CU
June 202633.9 million CU
May 202632.5 million CU
April 202635.2 million CU
March 202640 million CU
February 202642.1 million CU
January 202635.9 million CU
December 202526.5 million CU

What it shows

Compute units (CUs) consumed per block. One CU corresponds roughly to one BPF instruction. Each block is capped at up to 100M CUs at 400 ms slots (raised from 48M to 50M in April 2025, to 60M in July 2025, and to 100M on 2026-07-29 via SIMD-0286 at epoch 1009, slot 435,888,000). SIMD-0525 then cuts the cap with each slot-time step: 87.5M at 350 ms, 75M at 300 ms, 62.5M at 250 ms and 50M at 200 ms. Higher average CU usage indicates blocks are fuller; sustained near-cap usage suggests network congestion.

How it is computed

Firedancer reports average compute units per block for each validator each day, along with 10th, 50th, and 90th percentile values across that validator's blocks. Network, client, and scheduler figures are a blocks-produced weighted average across all validators in the group. A 7-day rolling average is available at the network level.

Aggregation: Blocks-produced weighted average. Validators that produced more blocks contribute proportionally more.

Things to know

Full method and formula in the Methodology

Related metrics