Finding the Right Instance

This tutorial walks through the complete workflow for discovering and evaluating EC2 instance types before launching — using truffle find, truffle search, truffle spot, and truffle quotas in sequence.

The workflow

truffle find    →  discover the instance family
truffle search  →  browse sizes and filter by specs
truffle spot    →  check current Spot prices
truffle quotas  →  verify you have headroom
spawn launch    →  launch with confidence

Each step is a separate sub-command. Flags like --min-vcpu only exist on truffle search, not on truffle find.

Example: AMD EPYC Genoa instances

Step 1 — Discover the family with `truffle find`

Use plain language to discover which instance families match a processor or capability:

truffle find "epyc genoa"

Output (abbreviated):

Found 40 instance types across 11 region(s)

┌────────────────┬────────────────┬───────┬──────────────┬──────────────┐
│ Instance Type  │ Region         │ vCPUs │ Memory (GiB) │ Architecture │
├────────────────┼────────────────┼───────┼──────────────┼──────────────┤
│ hpc7a.48xlarge │ us-east-2      │ 96    │ 768.0        │ x86_64       │
│ m7a.medium     │ eu-central-1   │ 1     │ 4.0          │ x86_64       │
│ m7a.large      │ eu-central-1   │ 2     │ 8.0          │ x86_64       │
│ m8a.4xlarge    │ us-east-1      │ 16    │ 64.0         │ x86_64       │
│ c7a.xlarge     │ us-east-1      │ 4     │ 8.0          │ x86_64       │
...

You can see the EPYC Genoa families: m7a, m8a, c7a, r7a, hpc7a. To include spec requirements in the query, put them in the string:

truffle find "epyc genoa 16 cores"
truffle find "epyc genoa 64gb"

find vs search flags

truffle find does not accept --min-vcpu or --min-memory. Include specs in the query string instead. Use truffle search when you want to filter by exact numeric thresholds.

Step 2 — Browse sizes with `truffle search`

Once you know the family, use truffle search to browse all sizes with numeric filters:

truffle search "m8a.*" --min-vcpu 16 --min-memory 64 --skip-azs

Output:

Found 5 instance types across 3 region(s)

┌───────────────┬───────────┬───────┬──────────────┬──────────────┐
│ Instance Type │ Region    │ vCPUs │ Memory (GiB) │ Architecture │
├───────────────┼───────────┼───────┼──────────────┼──────────────┤
│ m8a.4xlarge   │ us-east-1 │ 16    │ 64.0         │ x86_64       │
│ m8a.8xlarge   │ us-east-1 │ 32    │ 128.0        │ x86_64       │
│ m8a.16xlarge  │ us-east-1 │ 64    │ 256.0        │ x86_64       │
...

Common truffle search flags:

--min-vcpu N — minimum vCPU count
--min-memory N — minimum memory in GiB
--architecture arm64|x86_64 — filter by CPU architecture
--show-price — include on-demand pricing
--pick-first — output only the top result (for scripting)

Step 3 — Check Spot prices with `truffle spot`

Before committing, check current Spot prices. Spot can be 60–90% cheaper than on-demand:

truffle spot m8a.4xlarge --sort-by-price --active-only --show-savings

Output (--show-savings adds the On-Demand and Savings columns, using live AWS Price List rates):

┌───────────────┬───────────┬───────────────────┬───────────────┬───────────┬─────────┐
│ Instance Type │ Region    │ Availability Zone │ Spot Price/hr │ On-Demand │ Savings │
├───────────────┼───────────┼───────────────────┼───────────────┼───────────┼─────────┤
│ m8a.4xlarge   │ us-east-1 │ us-east-1a        │ $0.0812       │ $0.7750   │ 89%     │
│               │ us-east-1 │ us-east-1c        │ $0.0834       │ $0.7750   │ 89%     │
│               │ us-west-2 │ us-west-2b        │ $0.0901       │ $0.7750   │ 88%     │

To compare across regions, add --regions:

truffle spot m8a.4xlarge --regions us-east-1,us-east-2,eu-west-1 --sort-by-price

Step 4 — Check your quota with `truffle quotas`

Every AWS account has vCPU quotas per instance family. m8a is in the Standard family (same quota as M, C, R, T instances):

truffle quotas --family Standard --regions us-east-1

Output:

┌──────────┬───────────┬────────────┬───────────┬────────────┬────────┐
│ Family   │ Type      │ Quota      │ In Use    │ Available  │ Status │
├──────────┼───────────┼────────────┼───────────┼────────────┼────────┤
│ Standard │ On-Demand │ 1480 vCPUs │ 153 vCPUs │ 1327 vCPUs │ ✅ OK  │
│ Standard │ Spot      │ 640 vCPUs  │ -         │ 640 vCPUs  │ ✅ OK  │

An m8a.4xlarge has 16 vCPUs. With 1327 available, you can run up to ~82 simultaneous on-demand instances before hitting quota.

If you're near the limit, generate a quota increase request:

truffle quotas --family Standard --regions us-east-1 --request

Step 5 — Launch

spawn launch my-job \
  --instance-type m8a.4xlarge \
  --region us-east-1 \
  --spot \
  --ttl 4h

Or use --pick-first to pipe directly from truffle search:

spawn launch my-job \
  --instance-type $(truffle search "m8a.*" --min-vcpu 16 --pick-first) \
  --spot \
  --ttl 4h

GPU example: NVIDIA H100

The same workflow applies for specialized instances:

# 1. Find H100 instances
truffle find "h100 efa"

# 2. Browse p5 sizes (H100 is in p5 family)
truffle search "p5.*" --skip-azs

# 3. Check Spot prices (p5 is rarely Spot — check on-demand)
truffle spot p5.48xlarge --regions us-east-1,us-west-2

# 4. Check P-family quota (separate from Standard)
truffle quotas --family P --regions us-east-1

# 5. Launch (p5 typically on-demand, not Spot)
spawn launch gpu-job --instance-type p5.48xlarge --ttl 24h

Quick reference: which command for what

Task	Command
"What instances have H100 GPUs?"	`truffle find "h100"`
"What are all c7a sizes?"	`truffle search "c7a.*"`
"c7a instances with ≥32 vCPUs"	`truffle search "c7a.*" --min-vcpu 32`
"Current Spot price for g5.xlarge"	`truffle spot g5.xlarge`
"Do I have quota for more G instances?"	`truffle quotas --family G`
"Check capacity reservations"	`truffle capacity --gpu-only`

Finding the Right Instance ​

The workflow ​

Example: AMD EPYC Genoa instances ​

Step 1 — Discover the family with truffle find ​

Step 2 — Browse sizes with truffle search ​

Step 3 — Check Spot prices with truffle spot ​

Step 4 — Check your quota with truffle quotas ​

Step 5 — Launch ​

GPU example: NVIDIA H100 ​

Quick reference: which command for what ​

See also ​