# P-Speed

May 2026 · Infra & Systems · Rust, Solana CLI, Pinocchio

Benchmarks P-Token (SIMD-0266) against SPL Token on a local validator.

[Source](https://github.com/anmol0b/p-speed)

## Numbers
- -96%: compute units per SPL token transfer
- 255 CU: P-Token transfer, vs 6,323 CU on SPL Token
- 3: operations measured: mint creation, minting, transfer

## Problem
P-Token (SIMD-0266) is a Pinocchio-based rewrite of the SPL Token program. The claim is a large drop in compute, but the published figures are hard to reproduce, and the difference only shows up when you measure the compute units a real instruction consumes on-chain rather than estimating.

## Approach
P-Speed starts a local validator with the P-Token feature gate toggled, runs the same sequence of mint creation, minting, and transfers against SPL Token and against P-Token, and reads compute units from the transaction metadata. Every run leaves verifiable transaction signatures, so a number can be checked against the ledger instead of taken on trust.

## Compute units per transfer
local validator · feature gate on
- SPL Token: 6,323 CU
- P-Token: 255 CU

## What I'd do next
- Cover Token-2022 extensions (transfer hooks, confidential transfers) with the same harness.
- Run the benchmark in CI on every Agave release and publish the series.

[All projects](https://www.anmol0b.xyz/projects)
Anmol Bhardwaj

P-Speed

May 2026 · Infra & Systems · Rust, Solana CLI, Pinocchio

Benchmarks P-Token (SIMD-0266) against SPL Token on a local validator.

Source

Results

compute units per SPL token transfer-96%
P-Token transfer, vs 6,323 CU on SPL Token255 CU
operations measured: mint creation, minting, transfer3

Problem

P-Token (SIMD-0266) is a Pinocchio-based rewrite of the SPL Token program. The claim is a large drop in compute, but the published figures are hard to reproduce, and the difference only shows up when you measure the compute units a real instruction consumes on-chain rather than estimating.

Approach

P-Speed starts a local validator with the P-Token feature gate toggled, runs the same sequence of mint creation, minting, and transfers against SPL Token and against P-Token, and reads compute units from the transaction metadata. Every run leaves verifiable transaction signatures, so a number can be checked against the ledger instead of taken on trust.

Compute units per transfer

local validator · feature gate on, lower is better.

SPL Token6,323 CU
P-Token255 CU

What I'd do next

  • Cover Token-2022 extensions (transfer hooks, confidential transfers) with the same harness.
  • Run the benchmark in CI on every Agave release and publish the series.