Cloud provider marketing focuses on price and features, but performance varies significantly across providers, instance types, and regions. Benchmarks give a more objective basis for comparison — with the caveat that synthetic benchmarks don’t always reflect production workload behavior.

Compute Performance

General Purpose Instances (comparable tier: 4 vCPU, 16GB RAM):

The same workload on similar-spec instances often differs 10–30% across providers due to CPU generation, memory bandwidth, and hypervisor overhead.

Key factors:

  • CPU generation — newer processors (Intel Ice Lake, AMD EPYC Milan/Genoa, AWS Graviton3) outperform older generations significantly
  • ARM vs x86 — AWS Graviton3 and GCP Tau T2A (Ampere) deliver 20–40% better price-performance for most workloads vs. x86 equivalents
  • Network-attached vs local storage — local NVMe (available on specific instance families) dramatically outperforms network-attached block storage for IOPS

Storage Benchmarks

Object storage throughput:

  • All major providers deliver multi-Gbps throughput per prefix with parallel requests
  • Latency for small objects (< 1KB): typically 1–5ms for standard classes

Block storage IOPS:

  • AWS io2 Block Express — up to 256,000 IOPS per volume
  • Azure Ultra Disk — up to 400,000 IOPS
  • GCP Extreme Persistent Disk — up to 120,000 IOPS
  • OCI Block Volumes — consistent high IOPS with lower variability than AWS/Azure for database workloads

Network Throughput

Inter-AZ bandwidth within a region is generally 10–25 Gbps per instance for larger instance types. Key differences:

  • GCP’s global network delivers lower latency between regions due to Google’s private fiber backbone (packets avoid the public internet)
  • AWS offers the most options for dedicated connectivity (Direct Connect, Transit Gateway, PrivateLink)
  • OCI’s off-box network virtualization provides more consistent performance with less noisy-neighbor impact

Database Performance

Oracle Cloud OCI benchmarks show lower latency for Oracle database workloads vs. AWS or Azure, attributed to OCI’s network architecture and RDMA-capable interconnects. For open-source databases (MySQL, PostgreSQL), performance differences across providers are smaller.

Benchmark Caveats

  • Results vary by region, instance type, time of day, and workload pattern
  • Noisy neighbor effects are real — shared hardware means performance can vary
  • Run your own benchmarks with production-like workloads before committing
  • CloudHarmony and CloudSpectator publish ongoing third-party benchmarks

Practical Guidance

For most workloads, provider performance differences are smaller than the impact of correct sizing, proper architecture (caching, async processing, connection pooling), and storage tier selection. Optimize architecture before chasing provider benchmarks.

Back to Blog | Choosing a Provider