High-Performing and Elastic Compute

Bundled into the existing AWS Certified Solutions Architect – Associate premium course — no separate purchase.

Included in this chapter:

Family	Profile	Workload fit	Graviton variant
M (m6i, m7g)	Balanced CPU/RAM	General-purpose web / app servers	m6g / m7g
C (c6i, c7g)	Compute-optimized (high CPU/RAM ratio)	Batch processing, gaming, scientific	c6g / c7g
R (r6i, r7g)	Memory-optimized	In-memory DBs, large JVMs, real-time analytics	r6g / r7g
X (x2idn, x2iedn)	Extra-large memory	SAP HANA, large Apache Spark	—
I (i4i)	High local NVMe IOPS	NoSQL DBs (Cassandra, MongoDB, etc.)	—
D (d3)	Dense HDD storage	Data warehousing, MapReduce	—
P (p5)	GPU (NVIDIA H100)	ML training, HPC	—
G (g5)	GPU (cheaper than P)	ML inference, graphics rendering	—
Inf2 / Trn1n	ML accelerator (AWS chips)	ML inference (Inf2) / training (Trn1n)	ARM-based
T (t3, t4g)	Burstable CPU credits	Cheap, bursty workloads	t4g
Hpc (Hpc6id, Hpc7a)	HPC + EFA networking	Tightly-coupled HPC clusters	—

Cheat sheet

Match the instance family to the workload bottleneck
Compute spectrum: EC2 → ECS/EKS → Fargate → Lambda
Lambda max: 15 min, 10 GB memory, 10 GB ephemeral storage
Lambda provisioned concurrency eliminates cold starts (at a cost)
Placement groups: Cluster, Spread, Partition — each for a different goal
EFA (Elastic Fabric Adapter) bypasses the kernel — only on HPC + ML
Burstable T-family CPU credits — accumulated + spent per second
Lambda SnapStart cuts cold starts ~10× for supported runtimes
Spot interruption notice = 2 minutes; Rebalance recommendation earlier
ECS Fargate capacity provider: base guarantees minimum; weights split the rest
ECS Service Auto Scaling target tracking: asymmetric cooldowns for fast scale-out, slow scale-in
EKS Cluster Autoscaler priority expander controls which node group scales first
ASG lifecycle hooks pause instance launch / terminate for setup / cleanup

Unlock with Premium — includes all practice exams and the complete study guide.

Cheat sheet