Standard_E16ds_v5 on Databricks
Standard_E16ds_v5 is a memory optimized Azure worker type with 16 vCPUs and 128 GB of memory. On Databricks it uses 5.5 DBUs an hour, so 1 node on Jobs Compute in East US costs $2.80 an hour: $1.65 for the DBUs and $1.15 to Azure for the machine.
Azure · Memory optimized · Edsv5
Standard_E16ds_v5 at a glance
- vCPUs
- 16
- Memory
- 128 GB
- Own SSD
- 600 GB
- DBUs an hour
- 5.5
- Classic compute
- With Photon
- 13.75
- DBUs an hour, Jobs Compute
- 1 node an hour
- $2.80
- Jobs Compute, East US
1 node an hour on each kind of compute
In East US on the Premium plan. Databricks charges the DBUs, Azure the machine. A cluster pays this for every worker and for the driver.
| Compute | DBUs an hour | A DBU | Databricks | Machine | An hour |
|---|---|---|---|---|---|
| Jobs Compute | 5.5 | $0.30 | $1.65 | $1.15 | $2.80 |
| Jobs Compute + Photon | 13.75 | $0.30 | $4.13 | $1.15 | $5.28 |
| All-Purpose Compute | 5.5 | $0.55 | $3.03 | $1.15 | $4.18 |
| All-Purpose Compute + Photon | 11 | $0.55 | $6.05 | $1.15 | $7.20 |
| Lakeflow pipelines Core | 5.5 | $0.30 | $1.65 | $1.15 | $2.80 |
| Lakeflow pipelines Pro | 5.5 | $0.38 | $2.09 | $1.15 | $3.24 |
| Lakeflow pipelines Advanced | 5.5 | $0.54 | $2.97 | $1.15 | $4.12 |
In each region
1 node an hour on Jobs Compute, Premium, cheapest first.
| Region | Machine | An hour |
|---|---|---|
| East US | $1.15 | $2.80 |
| East US 2 | $1.15 | $2.80 |
| West US 2 | $1.15 | $2.80 |
| West US 3 | $1.15 | $2.80 |
| Central India | $1.21 | $2.86 |
| Sweden Central | $1.22 | $2.87 |
| Canada Central | $1.26 | $2.91 |
| North Europe | $1.28 | $2.93 |
| Central US | $1.30 | $2.95 |
| UK South | $1.35 | $3.00 |
| France Central | $1.35 | $3.00 |
| West Europe | $1.38 | $3.03 |
| Germany West Central | $1.38 | $3.03 |
| Southeast Asia | $1.39 | $3.04 |
| Japan East | $1.39 | $3.04 |
| Australia East | $1.39 | $3.04 |
| Brazil South | $1.82 | $3.47 |
Other memory optimized sizes
- Standard_E4ds_v54 vCPUs, 32 GB, 1.5 DBUs an hour
- Standard_E8ds_v58 vCPUs, 64 GB, 2.75 DBUs an hour
- Standard_E16ds_v516 vCPUs, 128 GB, 5.5 DBUs an hour
- Standard_E32ds_v532 vCPUs, 256 GB, 11 DBUs an hour
- Standard_E48ds_v548 vCPUs, 384 GB, 16.5 DBUs an hour
- Standard_E64ds_v564 vCPUs, 512 GB, 22 DBUs an hour
- Standard_E96ds_v596 vCPUs, 672 GB, 33 DBUs an hour
Questions
- How many DBUs does Standard_E16ds_v5 use on Databricks?
- Standard_E16ds_v5 uses 5.5 DBUs an hour on Databricks classic compute without Photon. With Photon it uses 13.75 DBUs an hour on Jobs Compute and 11 on All-Purpose Compute, as Databricks' price list gives them.
- How much does Standard_E16ds_v5 cost an hour on Databricks?
- In East US, 1 Standard_E16ds_v5 node on Jobs Compute costs $2.80 an hour: $1.65 to Databricks for 5.5 DBUs at $0.30 each, and $1.15 to Azure for the machine. On All-Purpose Compute the same node costs $4.18 an hour, and on Lakeflow pipelines Core $2.80. A cluster pays this for every worker and for the driver.
- Does Photon cost more on Standard_E16ds_v5?
- Yes, by the hour: with Photon, 1 Standard_E16ds_v5 node on Jobs Compute in East US costs $5.28 an hour instead of $2.80, because it uses 13.75 DBUs an hour instead of 5.5. It saves money only when the job finishes fast enough to make up the difference.
- Which region is cheapest for Standard_E16ds_v5 on Databricks?
- East US, at $2.80 an hour for 1 node on Jobs Compute, against $3.47 in Brazil South, the dearest of the 17 regions listed here that offer it. Databricks charges the same for a DBU of Jobs Compute in every Azure region listed, so the gap is the machine.
- What is the same size as Standard_E16ds_v5 on the other clouds?
- The nearest worker types of the same kind are r6id.4xlarge on AWS and n2-highmem-16 on Google Cloud. Each cloud gives its own DBUs an hour for its types, so compare what a cluster costs in the calculator, not the DBUs alone.
- What is the next size up or down from Standard_E16ds_v5?
- One size down is Standard_E8ds_v5, at 2.75 DBUs an hour; one size up is Standard_E32ds_v5, at 11. Within a family, DBUs and the machine price grow with the vCPUs, so 2 nodes of a size cost about what 1 node of twice the size costs.
Sources