AWS leads GCP and Azure on latency and cost, but lags on performance, according to a new benchmarking report.

Distributed SQL database firm Cockroach Labs’ third annual report looking into cloud performance of AWS, Microsoft Azure, and Google Cloud Platform (GCP) and runs benchmark tests against CPU Performance, Network Performance, Storage I/O Performance, and Online Transactional Processing (OLTP) Performance.

The report found that GCP delivered the most throughput on network throughput, storage I/O read throughput, storage I/O write throughput, and maximum throughput per minute. GCP’s top-performing machine had 165 percent and 237 percent more throughput than AWS and Azure respectively, while its bottom-performing network throughput machine still performed 62 percent and 105 percent better than AWS’ and Azure’s top-performing network throughput machines.

AWS is cheaper and lower-latency, lagging on performance

– PRNewsfoto/Frost & Sullivan

AWS was found to be the most cost-effective option, while Azure was the least cost-efficient cloud provider in terms of $/throughput per minute. Yet at the same time, AWS demonstrated the lowest performance on seven of the 12 benchmarks, including all storage I/O benchmarks and the single-core CPU benchmark. On latency, AWS saw as low as 40 microseconds, up to 28 percent and 37 percent lower than Azure and GCP respectively.

For the CPUs, Cockroach found Intel processors had the best performance on single-core runs, but faltered on 16-core tests. AWS’ custom-built Graviton2 Processor, which uses a 64-bit ARM architecture, performed better than GCP and Azure, both of which featured AMD processors. In terms of latency, AWS was by far the best performer, with Azure coming second.

The report evaluated 54 machine configurations and conducted nearly 1,000 benchmark test runs. This is the first year GCP has come out on top in performance benchmarks. In the previous edition of the report, GCP offered best performance per dollar while AWS offered the best performance on throughput alone.