Single-Thread vs. Multi-Thread Speed Test

What It Means

Single-thread and multi-thread refer to two fundamentally different speed test methodologies that measure different aspects of a connection and can produce dramatically different results on the same ISP plan. A single-thread test opens one TCP connection and measures maximum sustained throughput on that single connection, this is more representative of typical consumer activity like loading a single webpage, downloading a file from one server, or streaming a single Netflix video. Most real-world applications use a small number of parallel connections. A multi-thread test opens 4 to 16 simultaneous TCP connections and measures aggregate throughput, this saturates the ISP link and reports the maximum possible throughput the connection can deliver. Multi-thread is more representative of heavy activity like torrenting, multi-tab web browsing with simultaneous downloads, or speed-testing benchmarks. Multi-thread results are typically 20 to 50% higher than single-thread on cable and DSL connections due to TCP slow-start and BDP (bandwidth-delay product) effects that prevent a single TCP stream from fully utilizing a high-bandwidth, high-latency path. On fiber connections with very low latency, single-thread and multi-thread results are typically within 5 to 15% because TCP can saturate the link even on a single stream. Tool methodology: Ookla Speedtest uses multi-thread by default (8 to 16 streams), M-Lab NDT uses single-thread which better reflects real-world use, the FCC Speed Test uses multi-thread. Neither approach is "correct," they answer different questions. A large gap between single-thread and multi-thread results (over 40%) suggests significant buffering or congestion on a single-stream path.

Frequently Asked Questions