Cloud compute refers to the on-demand provisioning of computing resources (processors, memory, storage, and networking) from cloud providers as virtual machines or bare metal instances. It enables organizations to scale computational capacity dynamically without upfront hardware investment, paying only for what they use. At its core, cloud compute transforms fixed infrastructure costs into variable operational expenses while offering unprecedented flexibility in resource allocation. The 2026 generation of instances — including Graviton4-powered M8g/C8g/R8g, Intel Xeon 6-based M8i/C8i/R8i, and NVIDIA Blackwell P6 families — delivers 20–40% better price-performance than their predecessors, making generation migration one of the highest-ROI optimizations available. The fundamental distinction between instance types — general purpose, compute optimized, memory optimized, storage optimized, and accelerated computing — determines workload performance and cost efficiency, making instance selection the single most important decision in cloud architecture.
What This Cheat Sheet Covers
This topic spans 15 focused tables and 107 indexed concepts, 101 flashcards. Below is a complete table-by-table outline of this topic, spanning foundational concepts through advanced details.
A jump-to index of every table row in this cheat sheet.
An interactive map of every table and concept in this topic.
Table 1: Instance Type Families
Instance families group EC2 configurations by their primary optimization target. Choosing the wrong family — for example, a general purpose instance for a memory-bound database — is one of the most common and expensive cloud misconfigurations. Each family has a corresponding optimal CPU-to-memory ratio that determines which workloads it serves best.
| Family | Example | Description | |
|---|---|---|---|
AWS m8g.xlarge (Graviton4)Azure D-seriesGCP n2-standard-4 | • Balanced CPU-to-memory ratio (1:4) for web servers, small databases, and development environments • latest M8g/M8i/M8a generation in 2026 delivers ~30% better performance over M7 at similar cost | ||
AWS c8g.2xlarge (Graviton4)Azure F-seriesGCP c2-standard-8 | • High CPU-to-memory ratio (1:2) for batch processing, HPC, gaming servers, and CPU-intensive workloads • C8 family launched in 2026 with Intel Xeon 6 and Graviton4 variants | ||
AWS r8g.4xlarge (Graviton4)Azure E-seriesGCP m2-ultramem-208 | • High memory-to-CPU ratio (1:8+) for in-memory databases, real-time analytics, and large caching tiers • R8g offers up to 192 vCPUs and 1,536 GiB with DDR5-5600 memory | ||
AWS i4i.8xlargeAzure L-seriesGCP n2-highmem-80 | • High-performance NVMe SSD storage for NoSQL databases, data warehousing, and log processing requiring sustained IOPS • I8ge adds Graviton4 and third-gen Nitro SSDs in 2026. | ||
AWS p6-b200.48xlarge (8x B200)Azure NC-seriesGCP a3-highgpu-8g | • GPU, FPGA, or custom ML accelerator hardware for training and inference • P6-B200 features NVIDIA Blackwell GPUs with 1,440 GB HBM3e and up to 2x throughput over P5en | ||
AWS c8gn.48xlarger8gn.48xlarge | • Up to 600 Gbps network bandwidth — highest among EC2 instances • Graviton4-based C8gn and R8gn with 6th-generation Nitro Cards for security appliances, proxies, and network-intensive databases | ||
AWS t4g.mediumAzure B-seriesGCP e2-micro | • Baseline CPU with ability to burst using CPU credits • ideal for variable workloads like dev/test servers and microservices that spike infrequently | ||
AWS hpc8a.48xlargeAzure HB-seriesOracle BM.Optimized3.36 | Specialized instances with low-latency EFA/InfiniBand networking and tightly coupled cluster support for CFD, weather modeling, and molecular simulation. |