← All posts
ThesisMay 22, 2026·6 min

Why perpetuals belong on the GPU

By zKmex Team

Every on-chain perp DEX eventually hits the same wall: matching is serial. You walk the orderbook one resting order at a time, and as volume and markets grow, that single loop becomes the bottleneck that decides how many fills you can commit per block.

A GPU does not walk the book — it processes it. Thousands of resting orders across every market can be evaluated in parallel under a single price-time priority pass. The match goes from an O(n) crawl to a batched kernel that finishes in a fraction of a millisecond.

On Solana this matters twice. The chain is fast enough to commit the result every block, and cheap enough that re-deriving the match to verify it costs almost nothing. The GPU produces the fills; the chain anchors and proves them.

That is the bet behind zKmex: the unit of throughput for on-chain derivatives is not the transaction, it is the parallel match. Open a position in the app and watch a GPU-matched fill land in under a block.

Want to see it run? Open the playground →