We benchmarked every Valkey version. It was the wrong question.

We ran BullMQ against three Valkey versions on three different AWS instance types. Nine combinations total, 50,000 jobs each, five runs per combination.

After all that testing, the version difference on any given platform was 5 to 12%. Noticeable, sure, but not transformative.

Then we looked at the hardware axis.

On the same Valkey version, switching from Intel to Graviton doubled throughput. Not 10%, not 20%. A full 2x, and sometimes more.

We had been tuning the wrong knob. The Valkey version is a fine tuning parameter. The instance type is the coarse one. If you are optimizing BullMQ throughput, pick your hardware first. Then worry about which Valkey to run on it.