EC2 Instance Advisor

Loading instance dataset…

Interactive EC2 Instance Finder

Pick the right EC2 instance
for your workload, in minutes

AWS offers 500+ EC2 instance types across compute, memory, GPU, storage, and general purpose families. This advisor scores every instance on price, CPU, RAM, GPU, and network, weighted by your priorities, and surfaces the best match instantly. Scroll down to explore families, tune your weights, and lock in your pick.

—
Instance Types
—
Sub-families
—
AWS Regions
Snapshot: Mar 9, 2026 · Not live pricing
How to use this advisor (2 minutes)
A quick workflow so you don’t have to guess what to do next
  1. Learn the families — browse workload types and instance families to understand what fits your use case.
  2. Compare specs first — use charts and comparison views to understand tradeoffs before ranking.
  3. Set priorities in Top Match — apply presets and tune weights after you understand the data.
  4. Estimate cost — use the Monthly TCO calculator (compute + EBS + egress) to plan your budget.
  5. Pick a region — change your AWS region anytime via the globe chip in the navbar, or explore regional pricing at the bottom.
Good to know
  • Snapshot-based: prices come from a stored CSV, not live AWS billing.
  • Shortlisting tool: the goal is to narrow options; benchmark before production.
  • Shareable: use the Share this view button in the footer to send a link with your current settings.
Phase 1 Learn Understand workloads & instance families
01

Workload → Family Quick Reference

Not sure where to start? Match your primary bottleneck to a recommended instance family, then continue to explore families and set your priorities.
Workload Type Recommended Family Characteristics
Web & App Servers General Purpose (M/T) Balanced CPU and memory. T-series for burstable dev/test; M-series for steady production traffic.
High-Traffic APIs & HPC Compute Optimized (C) High vCPU-to-memory ratio. Scientific modeling, batch processing, and dedicated game servers.
In-Memory Databases Memory Optimized (R/X) 8:1+ RAM-to-vCPU ratio. Redis, SAP HANA, real-time analytics, large JVM heaps.
NoSQL & Data Warehousing Storage Optimized (I/D) Local NVMe SSDs for millions of low-latency IOPS. MongoDB, Cassandra, high-speed logging.
AI Training & Inference Accelerated (P/G/Trn/Inf) P-series for GPU training; G-series for inference & rendering; trn1/inf2 for AWS ML silicon.
02

Understand Instance Families

EC2 instances are grouped into families, each optimized for a specific hardware profile. Click a card to filter the entire analysis to that family, or keep All to compare across everything. Not sure which? Read the use-cases and examples below, then pick the family closest to your workload.
Tip: choose by bottleneck
If you're unsure, think about what usually limits you first: CPU-bound (builds/encoding/APIs) → Compute, RAM-bound (caches/analytics/JVM) → Memory, IOPS/throughput (datastores/logs) → Storage, CUDA/accelerators → GPU.
03

Family-Level Comparison

A high-level view to sanity-check the recommendation. The bars show typical vCPU count and RAM while the line shows average hourly cost. Use this to understand the broad tradeoffs between families.
Average Specs per Instance Family
Aggregated across all instances in selected region · bars = vCPUs & RAM (GiB) · line = avg cost ($/hr)
Family Highlights
Cheapest, most powerful, and best value instance per family
Phase 2 Decide Understand tradeoffs, then choose your best match
04

Compare Specs & Tradeoffs

Use these charts to understand instance behavior before changing ranking logic. Start with family-level comparison, then switch to instance mode for direct spec tradeoffs. Weight controls and presets are in Your Best Match (Step 05) so decisions come after context.
Instance Spec Comparison
● vCPUs · ● RAM (GiB) · ● Cost ($/hr, right axis)
Compare by: Select families using the pills below
Family Radar — Cheapest vs Most Powerful
Scores normalized [0–1] across 4 dimensions. Solid fill = most powerful · Dashed = cheapest per family.
How to read & Axes

Larger shapes mean stronger overall capability. Overlaps show specific trade-offs.

When comparing families, we plot the Cheapest and Most Powerful instance per family to show the spectrum of capability.

Price ($/hr)
CPU
RAM
Net
GPU
Read First, Decide After
  • Step 04 is for understanding: inspect families, specs, and tradeoffs before tuning ranking logic.
  • No hidden scoring changes here: ranking weights and presets are applied only in Step 05.
  • Then make your call: jump to Top Match when you are ready to prioritize.
05

Your Top Picks

This is the decision step: apply presets, tune weights, and see how your top picks change in real time. Click any pick to compare specs in detail. Use the full table below to explore all candidates.
How to read this section
  • Score: relative ranking within your current region + filters + weights.
  • Why this instance: breakdown + explanation show what contributed most given your weights.
  • Don’t stop at #1: check runners‑up and use the compare table if you need tradeoffs (cost vs. RAM, etc.).
Ranking Controls
Pick a category and apply a preset for instant re-ranking.
Instance Family
Presets
What do these weights do?
Spec axes rank by raw capability: more vCPUs, more RAM = higher score. Price is a separate axis that controls cost sensitivity. Increase CPU to shift toward instances with more vCPUs; increase Price to favor cheaper options.
  • Make one change at a time: adjust 1–2 criteria, then review why the top match changed.
  • Relative matters: “Price High + CPU Low” is meaningful; the exact numbers aren’t.
Click a level to set how important each criterion is for your recommendation
Price Efficiency
High
Favor instances with the lowest hourly cost
Compute (vCPUs)
Med
Favor instances with more vCPUs and better CPU efficiency
Memory (RAM)
Med
Favor instances with more RAM and better memory efficiency
Network Bandwidth
Low
Favor instances with higher bandwidth and better network efficiency
GPU Accelerators
Off
Favor instances with more GPU power and better GPU efficiency
Weight distribution 100%
Price 25% CPU 25% Memory 25% Network 10% GPU 10%
Matched Instances
—
Cheapest in Filter
—
Best Raw Performance
—
Best Value (perf/$)
—
Based on your current weights & filters
Top Picks for You
Probable options ranked by your priorities — pick the one that fits your workload best.
Instance Detail View
vs. top pick
💡
Why These Picks?
Adjust the weight sliders above to see a personalized explanation of why these instances were recommended.
06

Hardware Architecture Reference

CPU vendor, microarchitecture, ISA, hypervisor, and accelerator for each instance family. Use this earlier to validate platform fit before finalizing your top match.
Intel AMD AWS Graviton x86_64 ARM64
Phase 3 Analyze Deep dive charts, full table & cost estimator
07

Visual Analysis

Interactive charts for multi-dimensional exploration. Drag axis ranges to filter, compare metrics, and spot the best-value instances visually.
Parallel Coordinates — Multi-Dimensional Explorer
Each line is an instance flowing across all specs. Find your ideal trade-off visually.
↕ Drag on any axis to filter — only matching instances stay visible
☰ Slide the band up or down to shift your selection range
🎨 Line color = category or sub-family — toggle below
✕ Click "Clear filters" to reset all brushes
Axes:
Color by:
Loading…
↕ Drag on any axis to filter
Instance Comparison
Top 25 instances · price shown per bar · color = category · zoom to inspect
Value Map: Price vs Score
Each dot = one instance · x = hourly cost · y = composite score · top-left = best value · star = #1 recommendation
08

Browse All Candidates

The full ranked list for your current filters. Search by instance name or category. Click any row to open a head-to-head spec and score comparison against the #1 pick. Use / to focus the search field.
Fast ways to shortlist
  • Search by prefix (e.g., m7g, c6i) to focus on one family generation.
  • Click‑compare 2–3 close contenders to see where they differ (RAM, network, price).
  • Use this as a “top N”: you rarely need to read the entire table end‑to‑end.
Top Candidates
09

Monthly TCO Estimator

Compare billing models side-by-side. Factor in EBS storage type, data egress, and fleet size to get a realistic monthly TCO. Compute prices use the selected region snapshot; EBS and egress constants are us-east-1 reference rates — verify on AWS for other regions.
What’s included in this estimate?
  • Compute: hourly instance price × hours × count (pricing snapshot).
  • Storage: EBS $/GB‑month × storage size.
  • Egress: tiered public internet data transfer (rough estimate).
This is for planning and comparisons—not exact billing. AWS invoices can differ due to discounts, free tiers, credits, inter‑AZ traffic, CloudFront, PrivateLink, and more.
Billing Model Prices as of 2025 · Verify on AWS

Defaults to your current #1 recommendation. Change this to estimate monthly TCO for any instance in the ranked list.

Tiered egress: first 10 TB → $0.09/GB  ·  next 40 TB → $0.085/GB  ·  next 100 TB → $0.07/GB  ·  beyond → $0.05/GB

Effective rate: — / hr  ·  Region default: us-east-1  ·  Instance: —
Compute / mo
—
Storage / mo
—
Data Transfer / mo
—
Total TCO / mo
—
Annual estimate
—

Compute price uses the selected region's snapshot. EBS and egress constants are us-east-1 reference rates — actual costs vary by region. Reserved discounts are instance-family weighted averages. Savings Plans reflect 1-year Compute Savings Plan rates. Egress excludes CloudFront, Direct Connect, and inter-region transfers. Always verify on aws.amazon.com/ec2/pricing.

Phase 4 Reference Region pricing & methodology
10

Regional Price Comparison

Compare how your top pick’s on-demand price varies across AWS regions. Use the region selector in the navbar above to switch regions — all scores update instantly.
Regional Price Comparison
Top recommendation’s on-demand price across all available AWS regions
11

Scoring Methodology

How our engine ranks instances. The EC2 Instance Advisor uses Multi-Criteria Decision Analysis (MCDA) — an established operations-research framework — to produce consistent, explainable rankings.
  1. Filter
    Region, category & workload filters narrow the candidate set
  2. Normalize
    Fractional rank maps each metric to [0, 1]
  3. Blend
    Raw capability & value efficiency are mixed per axis
  4. Score & Rank
    Weighted sum → composite score → final ranking

Weighted Scoring Model (WSM)

Each instance is scored on five axes — price, CPU, memory, network, and GPU — then blended into a single composite score. Your slider levels control how much each axis contributes, so the ranking adapts to your workload priorities.
Composite score
score = Σ(wi × scorei)
= (wprice × scoreP) + (wcpu × scoreC) + (wmem × scoreM)
+ (wnet × scoreN) + (wgpu × scoreG)
Normalized weight
wi = slider_valuei / Σ(slider_values)
Sliders at 0 are excluded; remaining weights always sum to 1.0

Rank Normalization

Each metric is normalized to [0, 1] by fractional rank within the current candidate set. Unlike min-max normalization, this distributes scores uniformly regardless of outliers.
n = candidate count rank ∈ [0, n − 1]
Higher is better
score = rank(value) / (n − 1)
Lower is better (price)
score = 1 − rank(value) / (n − 1)
Ties share the average of their rank positions.

Efficiency-Blended Scoring

Each spec axis blends raw capability with value efficiency (spec-per-dollar). Your Price weight controls the blend: high price priority favors cost-efficient instances, low price priority favors raw power. This creates natural within-family progression (e.g. t4g: small → medium → large) and smooth cross-family transitions.
Price axis (independent)
scoreP = inverse rank(price_usd_hour)
Lower $/hr → higher score
Spec axes (blended)
raw = rank(spec)
value = rank(spec / price)
score = priceW × value + (1−priceW) × raw
Price weight controls capability vs. efficiency blend
GPU axis (tier-weighted)
scoreG = blend(rank(gpu×tier), rank(gpu×tier/price))
1K80 · M60 2T4 · V520 3A10G · L4 · V100 4A100 · L40S · Inf2 5H100 · Trn2 6H200

Why rank normalization instead of min-max?

Min-max normalization collapses when outliers dominate. If one $32/hr GPU instance sets the price ceiling, every sub-$1 instance gets a price score of ~1.0 and they become indistinguishable. Rank normalization distributes scores uniformly — each instance gets a distinct, meaningful position regardless of how extreme the raw values are.
Min-Max
t3.micro
0.97
t3.small
0.97
m5.large
0.95
c5.xlarge
0.92
p3.16xl
0.00
Scores clustered — can't differentiate
Rank-Based
t3.micro
1.00
t3.small
0.75
m5.large
0.50
c5.xlarge
0.25
p3.16xl
0.00
Evenly distributed — clear ranking

Limitations to keep in mind

  • Prices are a point-in-time snapshot — they drift from live AWS pricing over time
  • Only Linux on-demand, shared tenancy SKUs are scored
  • Network is a coarse ordinal score (1–5), not actual throughput measurements
  • EBS/egress costs in the TCO calculator use us-east-1 reference rates — actual rates vary by region
  • This is a shortlisting tool — always benchmark your actual workload before committing
12

References

Official AWS sources for instance specs, pricing constants, hardware architecture, and scoring methodology.
Instance Data
Data EC2 Instance Types 500+ instance types — vCPU, memory, storage & network specs per family ↗ Data AWS Bulk Pricing API JSON endpoint powering the on-demand price snapshot (2,571 rows · 30 regions) ↗
Pricing Constants
Pricing EC2 On-Demand Pricing Per-region hourly rates & data-egress tiers used in the TCO estimator ↗ Pricing Amazon EBS Pricing gp3 / gp2 / io1 / io2 / st1 / sc1 rates ($/GB-month) — us-east-1 defaults ↗ Pricing Reserved Instances Pricing 1-Year No-Upfront multipliers (34–36% off) used per instance family prefix ↗ Pricing Compute Savings Plans Flexible 1-Year commitment discount (~20% off on-demand) applied in cost estimator ↗
Hardware Architecture
Arch AWS Graviton Graviton2 / 3 / 3E / 4 — ARM64 ISA, Nitro, perf benchmarks for t4g, m6g–m8g, c6g–c7g, r6g–r7g ↗ Arch AWS Nitro System Nitro v1–v5 hypervisor generations mapped to each instance family ↗ Arch Intel on AWS Xeon Skylake → Cascade Lake → Ice Lake → Sapphire Rapids generations on EC2 ↗ Arch AMD EPYC on EC2 EPYC Rome / Milan / Genoa (Zen 2–4) — c6a, m6a, r6a, c7a, m7a, r7a, t3a, g5, g6 ↗ Arch NVIDIA on AWS T4 / A10G / L4 / V100 / A100 GPUs in g4dn, g5, g6, p3, p4d families ↗ Arch AWS Trainium & Inferentia Custom ML accelerator specs for trn1 (Trainium) and inf2 (Inferentia2) ↗
Methodology
Method Right Sizing Whitepaper MCDA-aligned CPU/memory thresholds, upsize/downsize/family-change decision framework ↗ Method AWS Well-Architected Cost Optimization pillar — right-sizing, reserved capacity & workload profiling ↗ Method AWS Compute Blog Graviton benchmarks (20–40% savings vs x86) & attribute-based instance selection ↗