Summary: Rollups and sidechains scale by moving execution off a congested L1 or onto a separate validator set, but their throughput and cost are ultimately bounded by data-availability choices and settlement rules. Purpose-built real-time settlement platforms (specialized L1s and regulated settlement networks) push the envelope on sub‑second finality and sustained throughput, often at fractions-of-a-cent per transaction or with central‑bank‑grade settlement—changing the total cost, latency, and operational tradeoffs you’ll face.

Can You Explain How Rollups or Sidechains Differ from Purpose-Built Real-Time Settlement Platforms in Terms of Throughput and Cost?

Decision-makers today have three practical paths for high-throughput, low-cost blockchain settlement:

Rollups on Ethereum (optimistic or ZK), which inherit L1 security and pay for L1 data availability (DA).
Sidechains/L1 appchains with their own validators (e.g., BNB Smart Chain, Polygon PoS evolving to validium).
Purpose-built real-time settlement platforms: high-performance L1s optimized for sub‑second finality (e.g., Solana, Aptos, Sui, Sei) and regulated institutional rails (e.g., Fnality; JPMorgan’s Onyx/Kinexys).

Below we unpack how throughput, finality, and fees differ—plus how 2025 upgrades (notably Ethereum’s Pectra) shifted rollup economics.

TL;DR: What changes your throughput and cost most

For rollups, data availability now dominates cost. After Ethereum’s Dencun (EIP‑4844) and Pectra (EIP‑7691), blobs provide cheap, ephemeral DA, doubling target blobspace per block and pushing L2 user fees to near‑pennies. Your real throughput scales with blobspace and your batch/prover pipeline. (eips.ethereum.org)
Sidechains achieve high TPS with low fees but rely on their own validator security and bridges (historically a major source of losses). Finality is fast (seconds), but you don’t inherit Ethereum’s settlement guarantees. (docs.bnbchain.org)
Purpose-built real-time platforms deliver sub‑second finality and massive throughput: Solana commonly sub‑$0.001 per txn with 400–600 ms slots; Move-based L1s (Aptos/Sui) are shipping <1s finality with parallel execution; institutions can use Fnality or JPM’s Onyx for central‑bank‑grade or bank‑grade cash settlement. These change both latency and the all-in operating model. (solana.com)

1) Definitions that matter for cost and throughput

Rollups (Optimistic/ZK): Execute transactions off L1, post state data to Ethereum via calldata or blobs. Optimistic rollups assume correctness unless challenged (typical withdrawal window ~6.4–7 days); ZK rollups post proofs for faster L1 finality. User-facing confirmations arrive quickly, but “hard” L1 settlement depends on challenge/proof windows. (docs.arbitrum.io)
Sidechains/appchains: Separate consensus with their own validator set and bridge. Fast and cheap, but security is not inherited from Ethereum. (docs.polygon.technology)
Purpose-built real-time settlement platforms:
- High-performance public L1s engineered for low latency/high throughput (e.g., Solana’s 400–600 ms slots; Sei’s ~400 ms deterministic finality targets; Aptos/Sui sub‑second pipelines). (solana.com)
- Regulated settlement rails for institutions (e.g., Fnality with settlement finality designation; Onyx/Kinexys settling $2B/day). (fnality.com)

2) Throughput in practice

Rollups: bounded by blobspace and your batching/prover

EIP‑4844 introduced “blob-carrying” transactions (128 KB each), pruned after ~18 days and priced in a separate fee market. Pectra (May 7, 2025) raised the target/max blobs per block from 3/6 to 6/9—boosting aggregate DA throughput and cutting posting costs volatility. That improves effective L2 TPS and lowers user fees when blob demand is below target. (eips.ethereum.org)
Operators see “near-free” blob periods post‑Pectra because average usage often sits below the new target; the aggregate daily capacity rose into multi‑GB/day. This directly translates to more L2 ops per batch and lower per‑op costs. (ethmrc.com)
Example activity: OP Mainnet’s recent daily ops peaked in the low‑millions, with ~17–28 user ops per second (UOPS). Raw throughput can be higher, but practical rates depend on demand, batching cadence, and DA limits. (l2beat.com)

Sidechains: higher TPS, fast blocks

BNB Smart Chain (PoSA) targets ~3s blocks, with “fast finality” around ~6 seconds; TPS in the hundreds is typical, with low, predictable fees due to larger gas limits and fewer bandwidth constraints than Ethereum L1. (docs.bnbchain.org)

Purpose-built real-time L1s: parallelism and pipeline

Solana: ~400–600 ms slot times; fees often around thousandths of a cent. Firedancer/Frankendancer testing demonstrated >1M TPS headroom, while mainnet operates at thousands+ sustained TPS depending on workload. The design emphasizes parallel execution (Sealevel) and high bandwidth networking. (solana.com)
Aptos: Block‑STM parallel execution and new pipeline research (Zaptos/Raptr) push sub‑second latencies at high TPS in testing and continue to reduce end‑to‑end latency in geo‑distributed settings. (theblock.co)
Sui: DAG-based Narwhal/Bullshark with Mysticeti targeting ~250–500 ms commit paths in tests; mainnet records showed multi‑thousand TPS days with sub‑second confirmations, leveraging object‑centric parallelism. (blog.sui.io)
Sei: Aggressively tuned Tendermint (Twin Turbo) targets ~400 ms deterministic finality for EVM developers and CLOB‑style use cases. (docs.sei.io)

Institutional settlement platforms emphasize deterministic, legally final settlement rather than headline TPS:

Fnality (Sterling FnPS) is the first regulated DLT wholesale payment system in the UK, designated for settlement finality and connected to RTGS/CHAPS—built for intraday, real‑time wholesale settlement. (fnality.com)
JPMorgan’s Onyx/Kinexys reports ~$2B/day settled on proprietary rails, now testing tokenized flows and deposit tokens across chains (e.g., Base). (coindesk.com)

3) Cost today (Dec 2025): what you’ll actually pay

Rollups: post‑Dencun/Pectra fees are down materially

Live snapshots (fees fluctuate with blob supply/demand and ETH price):

L2Fees.info (current range): ETH transfer on major L2s around $0.07–$0.19; swaps ~$0.18–$0.57 depending on stack/proofs. These observed user costs reflect both L2 execution and amortized L1 DA. (l2fees.info)
Why: Blobs decouple rollup DA from EVM gas and are pruned after ~18 days, enabling 10–100× lower posting costs than calldata in many periods. Pectra’s 6/9 blob target/max further expanded supply. (blocknative.com)

ZK vs Optimistic cost nuances:

ZK rollups incur proof costs but avoid week‑long withdrawal windows; typical hard finality on Ethereum is hours (e.g., ~3–6+ hours for some stacks), depending on proof aggregation cadence. Linea documents soft finality ~2s, hard finality 6–32 hours (trending down with roadmap). (docs.linea.build)
Optimistic rollups: near‑instant soft confirmations, but withdrawals to L1 wait ~6.4–7 days (configurable). Liquidity networks or “fast withdrawals” can shorten UX at the cost of trust/fees. (docs.arbitrum.io)

Sidechains: low fees, different security model

BNB Smart Chain fees are typically cents with ~3s blocks; costs are stable due to larger gas limits and a small validator set, but security and governance assumptions differ from Ethereum L2s. (docs.bnbchain.org)

Purpose-built real-time L1s: fractions of a cent at web scale

Solana: base fee 5,000 lamports (0.000005 SOL); even with priority fees in busy periods, typical simple tx costs are around $0.0005–$0.001 at $100/SOL. Program hotspots can drive priority fees, but median remains tiny vs. EVM L1. (solana.com)

Regulated settlement rails: fee schedule depends on membership and structure

Fnality (and similar wholesale systems) are designed around central‑bank‑grade settlement assets and participant fee models, not public-gas auctions; the economic value is in intraday liquidity savings and operational/netting efficiencies rather than micro tx fees. (fnality.com)

4) Settlement and finality: what “real-time” means in each model

Optimistic rollups: UX feels instant; economic finality for withdrawals/messages to L1 requires the challenge window (~6.4–7 days default on Arbitrum-type stacks). Apps often use liquidity providers for “fast exits.” (docs.arbitrum.io)
ZK rollups: UX is near‑instant; L1 hard finality governed by proof posting/verification (hours today; vendors are compressing to tens of minutes). Linea documents 6–32 hours hard finality currently, with roadmap to shrink further. (docs.linea.build)
Ethereum bridges (operational practice): Circle’s CCTP uses practical block confirmations; e.g., Base/OP/Arbitrum attestations in ~8 seconds, Solana ~8 seconds, Ethereum ~20 seconds (fast path), reflecting “operational finality” rather than full epoch finality. Plan around these for funds flows. (developers.circle.com)
High-performance L1s: sub‑second deterministic finality targets on Sei; Solana converges within a few slots (hundreds of ms per slot); Aptos/Sui push sub‑second commit paths with parallel pipelines. (docs.sei.io)
Regulated RTS: Fnality is designated for settlement finality in sterling, enabling intraday wholesale DvP/PvP designs. This is “legal finality” with bank‑grade rails, not just probabilistic finality. (fnality.com)

5) Security and risk surface that influence your cost of capital

Rollups inherit Ethereum security for DA and state validation; after 4844/7691, blob supply expanded, lowering fees while preserving the security model. Operator risk remains around centralized sequencers (decentralization efforts are ongoing). (eips.ethereum.org)
Sidechains rely on their validator set and bridges; historically, bridge compromises drove multi‑hundred‑million losses (e.g., Ronin; BSC Token Hub). Insurance, custody controls, and circuit breakers matter. (coindesk.com)
Shared sequencers aimed to reduce centralization across rollups; but operational viability is mixed—note Astria launched mainnet in 2024 and shut down in Dec 2025—so vendor risk and neutrality SLAs matter if you depend on them. (astria.org)
Institutional settlement networks (Fnality/Onyx) are permissioned, with stringent participant controls and cash-on-ledger backed by central bank reserves or bank liabilities—reducing counterparty/bridge risk at the expense of open composability. (fnality.com)

6) Concrete examples and what they cost/throughput

Consumer payments at scale (stablecoins)

Visa: expanded stablecoin settlement support (Ethereum, Solana, now Stellar and Avalanche; USDG, PYUSD, EURC), reflecting production appetite for cheap, fast rails. For USD-retail, Solana enables sub‑cent fees and sub‑second UX at scale. (usa.visa.com)
PYUSD: on Ethereum, Solana, and (2025) Arbitrum—teams can pick UX/security tradeoffs per flow. For near‑instant UX and global reach, Solana/Arbitrum integrations lower per‑tx cost and latency vs. L1‑only. (developer.paypal.com)

High-frequency trading or CLOB settlement

On a rollup: you’re bounded by L1 DA and L2 block cadence. If you need <500 ms determinism for order placement/matching, consider a purpose-built L1 (Sei ~400 ms, Solana sub‑second) or an appchain tuned for a CLOB (e.g., dYdX v4’s Cosmos appchain with ~1s blocks and off‑chain order propagation). (docs.sei.io)
On Solana: parallel runtime and low fees allow fully on-chain orderbooks (Phoenix/Serum lineage) with credible low-latency fills at scale. Fees are a rounding error relative to strategy PnL. (solana.com)

Intraday repo and DvP for institutions

Fnality: live sterling payment system with settlement finality protections; pilots for intraday repo and cross‑chain DvP indicate hours-to-minutes intraday cycles and lower balance sheet costs vs. current rails. For banks, that “cost” win can dwarf per‑tx gas. (reuters.com)
JPM Onyx/Kinexys: reports ~$2B/day settlement and ongoing tokenized asset pilots; this is bank‑operated, not public‑mempool settlement. (coindesk.com)

7) Emerging best practices (late 2025)

For teams choosing rollups:

Go blob‑first: assume cheap blob DA is the norm post‑Pectra; design batchers to opportunistically fill blobs, compress state diffs, and tune posting intervals to market conditions. Monitor blob base fee and aim for near‑target utilization. (blog.ethereum.org)
Model end‑to‑end finality: expose “soft” vs “hard” timelines in your UX (e.g., Base stages: ~200 ms flashblock, ~2 s L2 inclusion, ~2 m batched to L1, ~20 m L1 batch finality), and disclose 7‑day exits if optimistic. Provide on/off‑ramp partners for liquidity. (docs.base.org)
DA optionality: if your cost model is hypersensitive, consider modular DA (e.g., Celestia) for certain chains—pricing has been measured around cents per MB and even fractions of a cent per rollup txn at current parameters—but weigh trust/market maturity and interoperability. (forum.celestia.org)
Shared sequencing carefully: if you depend on shared sequencers for atomicity/neutrality, negotiate SLAs and fallback plans; monitor operator health (Astria’s 2025 shutdown is a cautionary tale). (unchainedcrypto.com)

For sidechains/appchains:

Harden bridges and governance: most catastrophic losses have been bridge‑related; bake in circuit breakers, rate limits, and multi‑party controls. Budget for security reviews as OPEX, not CAPEX. (cointelegraph.com)
Keep finality tight: tune consensus timeouts and gossip settings for your geo; BSC‑style 3s blocks with fast finality are feasible, but validator operations and monitoring are your responsibility. (docs.bnbchain.org)

For purpose-built real-time L1s:

Engineer for hotspots: on Solana, assume priority fees will dominate during peak program congestion; architect with local fee markets and request‑queue design. Fees remain tiny, but latency spikes can be app‑specific. (solanafloor.com)
Parallelize everything: Move/Sui/Aptos pipelines reward object‑centric design and conflict‑free batching; Sei’s pipelined consensus shortens wall‑clock time per state transition. Refactor flows to exploit parallel execution. (docs.sei.io)

For institutions:

If you need legal finality and central bank money claims, evaluate Fnality rails; for bank money settlement at scale with programmability, Onyx/Kinexys and Canton-connected networks are active. Pilot intraday repo/DvP and measure balance sheet savings vs. gas fees. (fnality.com)

8) Decision guide: which path fits your KPI?

Lowest end-user fees with web-scale throughput and sub‑second UX (consumer payments, social, gaming, market‑making UX): Solana or similar high‑performance L1; or ZK rollups where your DA cost model still yields sub‑cent UX. Validate with live fee trackers and latency SLOs. (solana.com)
Ethereum alignment (ecosystem liquidity/composability) with cheap UX and L1 security: launch on an OP/Arbitrum/zkEVM stack and optimize blob usage post‑Pectra. Budget withdrawal UX for optimistic or proof latency for ZK. (blog.ethereum.org)
App-specific exchange/trading requiring deterministic <1s fills: Sei/Solana app design or a tuned appchain (e.g., dYdX model) may beat a general‑purpose rollup on latency. Test p95/p99 end‑to‑end, not just average. (docs.sei.io)
Regulated cash settlement (intraday repo, DvP/PvP, tokenized funds): evaluate Fnality and bank rails, possibly bridging to public chains for distribution. Your main “cost” win is liquidity and capital efficiency, not gas. (fnality.com)

9) Quick math: framing cost per million actions

Rollup with blobs: Per‑op DA cost ≈ (bytes per tx ÷ 128 KB per blob) × (blob base fee) ÷ (txs per batch) + L2 exec gas. Post‑Pectra, blob base fee often sits near minimum due to 6/9 target/max, so your amortization curve improves dramatically as you pack batches. Validate against live L2Fees and your batcher stats. (blog.ethereum.org)
Solana: A million simple transfers at 0.000005 SOL base fee costs ~5,000 SOL lamports, i.e., ~5 SOL in base fees; even with modest priority fees it is orders of magnitude cheaper than L1 EVM. Model program‑specific priority fee spikes for hotspot contracts. (solana.com)
Celestia DA (for modular rollups): forum guidance suggests ~$0.08/MB at current params; for a 256‑byte state diff, DA cost can be on the order of 10^−5 USD per op. Price discovery is ongoing; treat as variable OPEX with monitoring. (forum.celestia.org)

10) What changed in 2025 that you should bake into roadmaps

Ethereum Pectra increased blob throughput (6/9 target/max) and capped worst-case calldata, materially easing rollup DA bottlenecks. Plan for further blob scaling (PeerDAS next). (blog.ethereum.org)
High-performance L1 clients matured (e.g., Firedancer entering validator sets), improving Solana reliability and capacity. Meanwhile, stablecoin heavyweights (Visa, PayPal) expanded multi‑chain settlement footprints—evidence that sub‑second, sub‑cent rails are being used in production payments. (solana.com)
Shared-sequencer experiments advanced but also showed operational fragility (Astria shutdown). Treat them as optional accelerants, not hard dependencies, until networks prove durable. (unchainedcrypto.com)

Bottom line for decision-makers

If you need Ethereum alignment with predictable, low fees: prefer ZK or optimistic rollups and optimize for blobs post‑Pectra. Your throughput will be “enough” for most web apps, and your unit economics trend better as blob supply grows. (blog.ethereum.org)
If you need exchange‑class latency at the edge of human perception: a purpose-built L1 (Solana/Sei/Aptos/Sui) or a tuned appchain will beat general‑purpose L2s on tail latency and determinism today. Model your p95/p99 latencies and sustained TPS under load. (solana.com)
If you need institutional-grade cash settlement and legal finality: regulated DLT platforms like Fnality or bank rails (Onyx/Kinexys) shift the cost equation from gas to liquidity/capital efficiency—this is a different KPI, but it’s “real-time settlement” in the legal sense. (fnality.com)

7Block Labs can help you quantify these tradeoffs with an implementation‑ready model: batch sizes and blob usage curves for rollups; validator/infra sizing and fee‑market behavior on Solana/Sei/Aptos/Sui; and operational/functional testing on institutional rails like Fnality or Canton‑connected networks to achieve the mix of throughput, latency, and cost your roadmap requires. (canton.network)

Notes and sources

EIP‑4844 blobs and Pectra’s blob increase (6/9): official EIPs and EF blog. (eips.ethereum.org)
Live L2 fees: L2Fees.info. (l2fees.info)
Solana fees and slot times: official docs. (solana.com)
High-performance L1 performance research: Sei, Aptos, Sui. (docs.sei.io)
Institutional rails: Fnality designation and Onyx volumes. (fnality.com)
Bridge risk context. (coindesk.com)

If you want, we can plug your actual workload (transactions per user, state diff size, DA policy, target p95 latency) into a one‑page “architecture bill of materials” comparing a blob‑optimized rollup, a Solana deployment, and an institutional RTS integration to select the best-fit path with 12‑month OPEX/latency projections.