Crypto's '1 Million Payments a Second' Promise Just Met Cheap Hardware
Adding more computers made things slower — a warning for every fast-payment promise.
NEAR Protocol is a cryptocurrency network that promises to handle a million payments per second by dividing the workload across many computers at once — a technique called sharding. Until now, the only public speed tests ran on rented cloud servers costing roughly $700 per hour, which almost nobody could afford to repeat or check. This team ran the first independent test on plain, affordable hardware: a single machine with 48 processor cores and ordinary disk storage costing a fraction as much.
The results show a clear sweet spot. The researchers tried 1, 2, and up to 24 shards (each shard being one slice of the network doing part of the job). Speed climbed until 8 shards, about 40 percent faster than a single one — then it fell apart. At 24 shards, each individual shard did 23 times less work than it managed alone. In other words, adding more computers created more coordination overhead than extra capacity. Past a point, extra machines were mostly talking to each other instead of doing useful work.
The most surprising finding was accidental. The machine's slow spinning hard drive, which writes only 80 to 100 megabytes per second, was quietly acting like a traffic cop. It slowed down the messages shards send each other just enough to keep everything in sync. When researchers swapped it for a fast memory-based disk, the entire network froze at 16 shards. That suggests the original cloud benchmarks may have been hiding a fragility, because their storage was too fast to expose it.
The takeaway for everyday users is simple. Big blockchain speed claims depend heavily on the exact hardware underneath, and "just add more shards" is not free magic. Independent testing on machines people can actually afford matters, because if crypto payments are ever going to compete with credit cards on speed and cost, they have to work reliably on realistic equipment — not just on a $700-an-hour rental.
- Splitting a crypto network into more pieces stopped helping past 8; by 24 pieces, each did 23 times less work.
- The machine's slow hard drive accidentally prevented crashes — replacing it with fast memory froze the network at 16 pieces.
- Earlier tests ran on $700-an-hour cloud servers; this one ran on one ordinary machine and still hit a wall.
Why It Matters
If crypto payments are to rival credit cards, networks must work on hardware people can actually afford.