Cloud Provider Benchmarks
The tracked providers are AWS, PortvMind Cloud, Google Cloud, Microsoft Azure, DigitalOcean, Vultr and Hetzner. Every figure links to its source and every method is written out in full. Values derived from specifications are labelled as derived, not as measurements.
Benchmarks
How we benchmark cloud virtual machines: CPU throughput, memory bandwidth, boot time and price-performance across AWS, PortvMind Cloud, Google Cloud, Azure and more.
Compare NVIDIA H100, H200 and L40S cloud GPUs for LLM inference: specifications, memory capacity models, provider availability, list prices and verified runs.
How we benchmark cloud block storage volumes with fio: random 4K IOPS, sequential throughput, latency percentiles, snapshot support and pricing by provider.
Benchmarking S3-compatible object storage across PortvMind, AWS S3, DigitalOcean Spaces, Vultr and Hetzner: PUT/GET throughput, latency and API coverage.
How we measure cloud networking: VM-to-VM bandwidth with iperf3, latency to Istanbul and Frankfurt, and load balancer, floating IP and NAT capabilities.
Comparing managed Kubernetes services, including PortvMind VKE, EKS, AKS, GKE, DOKS and Vultr VKE, on cluster creation time, autoscaling and integrations.
A qualitative comparison of documented cloud security capabilities: security groups, private networks, audit logging, identity attribution and certifications.
How cloud audit logging services record who did what: scope, identity, request metadata, search, retention and export for vTrail, CloudTrail, Azure and GCP.
Normalised cloud pricing: implied per-vCPU and per-GiB hourly prices, per-GPU-hour list prices, billing granularity and the sources behind every figure.
GPU & AI
Model fit, KV-cache capacity and decode ceilings for H100, H200 and L40S, across Qwen, Llama, DeepSeek and MiniMax models.
Hopper
Hopper-generation data-centre GPU with 80 GB of HBM3 (SXM) and FP8 Transformer Engine support; the reference accelerator for LLM training and high-throughput inference.
Hopper
Hopper-generation GPU with 141 GB of HBM3e at 4.8 TB/s — the same Tensor Core compute as H100 SXM with 1.76× the memory and 1.43× the bandwidth.
Ada Lovelace
Ada Lovelace universal data-centre GPU with 48 GB of GDDR6 at 864 GB/s and FP8 support — a lower-cost option for inference of small and mid-size models, graphics and video.
Research
Pricing
How tracked cloud providers bill compute by the second, minute or hour, with the minimum charges and stop rules that change what short jobs cost.
Pricing
How tracked cloud providers state currency and tax treatment, and how to compare their published prices without a hidden conversion or a tax mismatch.
Storage
How to point S3 tools and libraries at Google Cloud Storage with HMAC keys and the XML API, what gcloud can do with S3 buckets, and what to test before you rely on it.
Storage
Where S3-compatible object storage is located in Türkiye, what the AWS Istanbul Local Zone adds, and which providers have no Türkiye location in the sources checked.
Provider discovery
AWS, PortvMind Cloud, Google Cloud, Azure and the other tracked providers when a workload needs an AWS alternative in Türkiye.
Cloud security & auditing
CloudTrail's equivalents on Google Cloud, Azure, OpenStack-based clouds and smaller providers, compared on scope, identity, retention, export and integrity.
Providers
Compare
Rankings
Reports
CPB-2026-003 · v1.0
CPB-2026-006 · v1.0
CPB-2026-001 · v1.0
CPB-2026-002 · v1.0
CPB-2026-004 · v1.0
CPB-2026-005 · v1.0