NODE95

An RPC (Remote Procedure Call) node is the gateway between your application and the Solana blockchain. Every balance check, transaction submission, and on-chain data query routes through an RPC endpoint. The speed and reliability of that node directly determines how fast your transactions confirm, how accurately your app reads chain state, and whether your orders land during high-congestion windows.

Every 15 minutes, NODE95 fires two concurrent requests against each provider's Solana mainnet endpoint — getSlot (confirmed) to measure the node's view of the chain tip, and getLatestBlockhash to simulate a real transaction preparation call. Both calls are timed together, and the total round-trip duration is recorded. Over time, these samples are grouped into 5-minute buckets and the P95 latency is calculated per bucket — meaning the value shown is the latency that 95% of samples in that window fell under. Slot lag is measured by comparing each provider's confirmed slot against the network's finalized slot at the same moment. All benchmarked providers are tested against their standard shared node endpoints — not dedicated or enterprise tiers.

Shared nodes are the entry point that the vast majority of developers use. Dedicated and enterprise nodes offer significantly better performance, but at costs starting from hundreds to thousands of dollars per month — out of reach for most teams evaluating providers. Benchmarking shared endpoints gives a realistic picture of what you get when you sign up, not a best-case scenario that requires a premium commitment. Where dedicated nodes are the only option (Triton One), that provider is marked as specs-only in the comparison table.

P95 (95th percentile) latency is the response time that 95% of your requests fall under. An average can look healthy at 150ms while hiding the fact that 5% of requests are timing out at 2,000ms. For MEV bots and high-frequency trading strategies, those outliers represent missed blocks and lost profit. P95 exposes the tail behaviour that averages conceal.

Slot lag measures how many confirmed slots behind the RPC node's view of the chain is compared to the actual network tip. Solana produces approximately 2.5 slots per second. A lag of +4 slots means your application is operating on data that is roughly 1.6 seconds stale — enough to cause failed transactions and wasted compute fees for any application reading live prices or account state.

Compute Units are a standardised measure of the computational cost of an RPC call. Pay-as-you-go pricing means you only pay for what you use — structurally efficient for variable-traffic applications. Fixed monthly tiers force you to pre-purchase capacity, paying for peak load at all times even when your usage is a fraction of the cap.

Stake-weighted Quality of Service is a Solana protocol mechanism that allocates leader bandwidth proportionally to the stake backing a connection. During high-load periods, a transaction submitted through a staked RPC provider has a higher probability of landing in the next block — regardless of priority fees.

Multi-region failover routes your requests across geographically distributed nodes so that a regional outage or congestion event doesn't degrade your entire connection. When one node or cluster underperforms, traffic redistributes automatically to a healthy alternative. Providers with decentralised or multi-operator architectures handle this more gracefully than those running a single centralised node pool, since independent operators are unlikely to fail simultaneously. Check each provider's infrastructure documentation to understand whether failover is within their own data centres or across genuinely independent operators.