The mac-m3ultra.metal Instance Has Been Added to EC2, So I Checked Its Specs and Pricing Using the Pricing API
This page has been translated by machine translation. View original
Introduction
In May 2026, mac-m3ultra.metal, an EC2 Mac instance equipped with the Apple M3 Ultra, was added. It is an instance type that exceeds the previous top-end mac-m4max.metal in CPU core count and memory.
Here is a comparison of the main specs.
| Item | mac-m4max.metal | mac-m3ultra.metal (NEW) |
|---|---|---|
| Processor | Apple M4 Max | Apple M3 Ultra |
| Clock | Up to 4.5 GHz | 4 GHz |
| vCPU | 16 | 28 |
| Memory | 128 GiB | 256 GiB |
| EBS bandwidth | 8 Gbps | 8 Gbps |
| Network | 10 Gbps | 10 Gbps |
| DH price | $6.25/hr | $12.50/hr |
The corresponding hardware is the Apple Mac Studio (M3 Ultra).
I will not allocate or launch an actual machine, and will check only within the scope of the information available from the Pricing API.
Verification
Verification environment
| Item | Value |
|---|---|
| Region | us-east-1 (N. Virginia) |
| API used | AWS Pricing API (pricing:GetProducts) |
| Hands-on verification | None (no Dedicated Host allocation or instance launch) |
| Retrieval date | July 14, 2026 |
Retrieving information on mac-m3ultra.metal with the Pricing API
First, retrieve the product attributes of mac-m3ultra.metal. Because EC2 Mac instances run only on Dedicated Hosts, specify Host for the tenancy. Elements of PriceList are returned as JSON strings, so the attribute portion is extracted with jq.
aws pricing get-products \
--region us-east-1 \
--service-code AmazonEC2 \
--filters \
'Type=TERM_MATCH,Field=instanceType,Value=mac-m3ultra.metal' \
'Type=TERM_MATCH,Field=location,Value=US East (N. Virginia)' \
'Type=TERM_MATCH,Field=tenancy,Value=Host' \
'Type=TERM_MATCH,Field=operatingSystem,Value=macOS' \
--output json \
| jq -r '.PriceList[] | fromjson | .product.attributes
| {instanceType, physicalProcessor, clockSpeed, vcpu, memory,
dedicatedEbsThroughput, networkPerformance, storage, tenancy}'
{
"instanceType": "mac-m3ultra.metal",
"physicalProcessor": "Apple M3 Ultra",
"clockSpeed": "4 GHz",
"vcpu": "28",
"memory": "256 GiB",
"dedicatedEbsThroughput": "8 Gbps",
"networkPerformance": "10 Gigabit",
"storage": "EBS only",
"tenancy": "Host"
}
The vCPU count of 28 and memory of 256 GiB are double those of the Apple M3 Max (14 vCPUs, 128 GiB memory). The M3 Ultra is said to be a configuration that joins two M3 Max dies with UltraFusion, and these values are thought to reflect that.
The OnDemand Dedicated Host price was retrieved as follows.
aws pricing get-products \
--region us-east-1 \
--service-code AmazonEC2 \
--filters \
'Type=TERM_MATCH,Field=instanceType,Value=mac-m3ultra.metal' \
'Type=TERM_MATCH,Field=location,Value=US East (N. Virginia)' \
'Type=TERM_MATCH,Field=tenancy,Value=Host' \
'Type=TERM_MATCH,Field=operatingSystem,Value=macOS' \
--output json \
| jq -r '.PriceList[] | fromjson | .terms.OnDemand[].priceDimensions[]
| "\(.pricePerUnit.USD) USD / \(.unit)"'
It was $12.50 per hour.
12.5000000000 USD / Hrs
Pricing API response (excerpt of attributes and price)
{
"product": {
"productFamily": "Dedicated Host",
"attributes": {
"instanceType": "mac-m3ultra.metal",
"physicalProcessor": "Apple M3 Ultra",
"clockSpeed": "4 GHz",
"vcpu": "28",
"memory": "256 GiB",
"dedicatedEbsThroughput": "8 Gbps",
"networkPerformance": "10 Gigabit",
"storage": "EBS only",
"tenancy": "Host",
"operatingSystem": "macOS",
"location": "US East (N. Virginia)",
"locationType": "AWS Region",
"capacitystatus": "Used",
"servicecode": "AmazonEC2"
}
},
"terms": {
"OnDemand": {
"...": {
"priceDimensions": {
"...": {
"unit": "Hrs",
"description": "USD 12.5 per On Demand macOS mac-m3ultra.metal Dedicated Host Hour",
"pricePerUnit": {
"USD": "12.5000000000"
}
}
}
}
}
}
}
Comparing all Mac instance types
Compared with mac-m4max.metal, mac-m3ultra.metal has double the memory and double the price, with vCPUs going from 16 to 28. The following is a list comparing the nine Mac instance types listed on the official AWS page as of July 14, 2026, retrieved from the us-east-1 Pricing API.
| Instance type | Processor | Clock | vCPU | Memory | EBS bandwidth | DH price ($/hr) |
|---|---|---|---|---|---|---|
| mac1.metal | Intel Core i7-8700B | 3.2 GHz | 12 | 32 GiB | 1,750 Mbps | $1.083 |
| mac2.metal | Apple M1 | 3.2 GHz | 8 | 16 GiB | 1,000 Mbps | $0.65 |
| mac2-m1ultra.metal | Apple M1 Ultra | 3.2 GHz | 20 | 128 GiB | 8 Gbps | $5.00 |
| mac2-m2.metal | Apple M2 | 3.49 GHz | 8 | 24 GiB | 8 Gbps | $0.878 |
| mac2-m2pro.metal | Apple M2 Pro | 3.5 GHz | 12 | 32 GiB | 8 Gbps | $1.56 |
| mac-m4.metal | Apple M4 | 4.4 GHz | 10 | 24 GiB | 8 Gbps | $1.23 |
| mac-m4pro.metal | Apple M4 Pro | 4.5 GHz | 14 | 48 GiB | 8 Gbps | $1.97 |
| mac-m4max.metal | Apple M4 Max | Up to 4.5 GHz | 16 | 128 GiB | 8 Gbps | $6.25 |
| mac-m3ultra.metal | Apple M3 Ultra | 4 GHz | 28 | 256 GiB | 8 Gbps | $12.50 |
Note that for mac1.metal and mac2.metal, there were discrepancies between the Pricing API's dedicatedEbsThroughput and the EBS bandwidth on the official page. The Pricing API returns 1,750 Mbps and 1,000 Mbps respectively, but the official page lists 8 Gbps for both. Please check the official page for the accurate EBS bandwidth values.
Command to retrieve all 9 types at once
INSTANCE_TYPES=(
mac1.metal
mac2.metal
mac2-m1ultra.metal
mac2-m2.metal
mac2-m2pro.metal
mac-m4.metal
mac-m4pro.metal
mac-m4max.metal
mac-m3ultra.metal
)
for instance_type in "${INSTANCE_TYPES[@]}"; do
aws pricing get-products \
--region us-east-1 \
--service-code AmazonEC2 \
--filters \
"Type=TERM_MATCH,Field=instanceType,Value=${instance_type}" \
'Type=TERM_MATCH,Field=location,Value=US East (N. Virginia)' \
'Type=TERM_MATCH,Field=tenancy,Value=Host' \
'Type=TERM_MATCH,Field=operatingSystem,Value=macOS' \
--output json
done | jq -s '[.[] | .PriceList[] | fromjson | {
instanceType: .product.attributes.instanceType,
processor: .product.attributes.physicalProcessor,
clockSpeed: .product.attributes.clockSpeed,
vcpu: .product.attributes.vcpu,
memory: .product.attributes.memory,
ebsThroughput: .product.attributes.dedicatedEbsThroughput,
price: (.terms.OnDemand[].priceDimensions[].pricePerUnit.USD)
}]'
Pricing API results for all 9 types (JSON)
[
{
"instanceType": "mac1.metal",
"processor": "Intel Core i7-8700B",
"clockSpeed": "3.2 GHz",
"vcpu": "12",
"memory": "32 GiB",
"ebsThroughput": "1,750 Mbps",
"price": "1.083"
},
{
"instanceType": "mac2.metal",
"processor": "Apple M1",
"clockSpeed": "3.2 GHz",
"vcpu": "8",
"memory": "16 GiB",
"ebsThroughput": "1,000 Mbps",
"price": "0.65"
},
{
"instanceType": "mac2-m1ultra.metal",
"processor": "Apple M1 Ultra",
"clockSpeed": "3.2 GHz",
"vcpu": "20",
"memory": "128 GiB",
"ebsThroughput": "8 Gbps",
"price": "5.00"
},
{
"instanceType": "mac2-m2.metal",
"processor": "Apple M2",
"clockSpeed": "3.49 GHz",
"vcpu": "8",
"memory": "24 GiB",
"ebsThroughput": "8 Gbps",
"price": "0.878"
},
{
"instanceType": "mac2-m2pro.metal",
"processor": "Apple M2 Pro",
"clockSpeed": "3.5 GHz",
"vcpu": "12",
"memory": "32 GiB",
"ebsThroughput": "8 Gbps",
"price": "1.56"
},
{
"instanceType": "mac-m4.metal",
"processor": "Apple M4",
"clockSpeed": "4.4 GHz",
"vcpu": "10",
"memory": "24 GiB",
"ebsThroughput": "8 Gbps",
"price": "1.23"
},
{
"instanceType": "mac-m4pro.metal",
"processor": "Apple M4 Pro",
"clockSpeed": "4.5 GHz",
"vcpu": "14",
"memory": "48 GiB",
"ebsThroughput": "8 Gbps",
"price": "1.97"
},
{
"instanceType": "mac-m4max.metal",
"processor": "Apple M4 Max",
"clockSpeed": "Up to 4.5 GHz",
"vcpu": "16",
"memory": "128 GiB",
"ebsThroughput": "8 Gbps",
"price": "6.25"
},
{
"instanceType": "mac-m3ultra.metal",
"processor": "Apple M3 Ultra",
"clockSpeed": "4 GHz",
"vcpu": "28",
"memory": "256 GiB",
"ebsThroughput": "8 Gbps",
"price": "12.50"
}
]
Reference: Instance Specs on the official page
The Pricing API cannot retrieve the GPU core count, Neural Engine core count, or the P/E core breakdown. These can be supplemented from the Instance Specs table on the official page. The following is what was confirmed on that page (confirmed on July 14, 2026, quoted for reference).
| Instance | Physical cores | CPU configuration | Memory (GB) | GPU | Neural Engine |
|---|---|---|---|---|---|
| mac-m3ultra.metal | 28 | P20 + E8 | 256 | 60 | 32 |
| mac-m4max.metal | 16 | P12 + E4 | 128 | 40 | 16 |
| mac-m4pro.metal | 14 | P10 + E4 | 48 | 20 | 16 |
| mac-m4.metal | 10 | P4 + E6 | 24 | 10 | 16 |
| mac2-m2pro.metal | 12 | P8 + E4 | 32 | 19 | 16 |
| mac2-m2.metal | 8 | P4 + E4 | 24 | 10 | 16 |
| mac2-m1ultra.metal | 20 | P16 + E4 | 128 | 64 | 32 |
| mac2.metal | 8 | P4 + E4 | 16 | 8 | 16 |
| mac1.metal | 6 | 6 (Intel) | 32 | - | - |
mac-m3ultra.metal has 60 GPU cores and 32 Neural Engine cores. However, looking at GPU core count alone, mac2-m1ultra.metal with 64 cores is higher.
Summary
mac-m3ultra.metal offers the largest vCPU count (28) and memory (256 GiB) of any EC2 Mac instance to date. Its price is also the highest at $12.50/hr, so when using it, please consider the minimum 24-hour charge ($300 or more).
Reference links
