Google Cloud Platform (GCP) is the best hyperscale performer across all areas of throughput while Microsoft Azure has the best storage systems and Amazon Web Services (AWS) has the lowest network latency.\nThose are the findings of a series of benchmarks performed by the atrociously named Cockroach Labs, maker of a scalable, resilient database called CockroachDB that runs on all three services. The study, part of the company\u2019s third annual Cloud Report, evaluated the performance of AWS, Microsoft Azure, and Google Cloud in online transaction processing (OLTP) applications.\nIn total, 54 machines were assessed and almost 1,000 benchmark runs were conducted to measure CPU, network, storage I\/O, and TPC-C performance, among others.\nWhile the report declared Google Cloud an overall winner, it applied a note of caution: Each cloud provider stood out in different ways. It singled out GCP for overall pejrformance and best throughput on the Derivative TPC-C (OLTP) benchmark; Azure for its disks in storage I\/O; and AWS as the most cost efficient option for OLTP.\n\u201cIt is important to point out that each of the cloud providers showcased overall growth in machine performance since the 2020 report,\u201d the researchers wrote.\nFor having the fastest pricessing rates, Google Cloud won the network throughput benchmark for the third year in a row, with its worst performing machine still beating the best from AWS and Azure. GCP achieved the best single-core CPU performance and delivered the most throughput at every level, including on the OLTP benchmark.\nThe TCP-C benchmarks also found something interesting among the chips AWS is using. It offers instances powered by Intel, AMD, and Amazon\u2019s Graviton2, an Arm derivative. In single-core performance, Intel was a clear winner, while on the 16-core performance benchmark, AWS\u2019s Graviton2 beat everybody.\nGCP\u2019s machines also outperformed Azure and AWS in network and storage I\/O throughput (read and write) and also attained the highest amount of raw throughput (tpm). Despite not having an advanced disk option (extreme-pd) available, GCP machines with general purpose disks (pd-ssds) came in second in cost efficiency ($\/tpm).\nAs far as network latency was concerned, AWS registered the best performance in network latency for three years running. Their top-performing machine\u2019s 99th percentile network latency was 28% lower than Azure and 37% lower than GCP. However, Cockroach noted that there is possible randomness in the physical distance between instances.\nAs far as storage went, Azure had comparable raw throughput (tpm) performance with GCP and AWS, although the differences were razor thin. Azure surpassed AWS in storage I\/O performance with what Microsoft calls \u201cultra disks,\u201d SSDs specialized in high IOPS and low-latency performance.\nWith these ultra disks, their machines led all cloud providers in storage I\/O read IOPS, write IOPS, and write latency. But that performance came at a higher cost. Azure was the least cost-efficient cloud provider in terms of dollars per tpm.\nThe Cockroach report is available for free download, email registration required.