Dynamic Workload Scheduler pricing (original) (raw)
- Dynamic Workload Scheduler (DWS) is a resource management and capacity scheduling platform designed for AI Hypercomputer. Dynamic Workload Scheduler provides a cost optimized consumption option for Google Cloud TPUs and NVIDIA GPUs. Dynamic Workload Scheduler is also integrated in many of your preferred Google Cloud AI/ML services: Compute Engine Managed Instance Groups, Google Kubernetes Engine, Vertex AI, Batch, and more.
How DWS pricing works
DWS prices are variable, but are updated infrequently. You pay the price that is in effect when your instances are running. Currently, the price for A2, A3, A4 VMs, and TPUs are intended to have the same price across each of their supported regions. To compare DWS prices to on-demand and commitment rates, visit the Compute Engine pricing page.
Prices listed are exclusive of applicable taxes and duties, and do not reflect any enterprise negotiated pricing. To view your account’s specific pricing history, you can export your Cloud Billing data to BigQuery.
Pricing components
Dynamic Workload Scheduler Flex-start VMs are charged based on usage, in the same way as on-demand instances. A bill is sent out at the end of each billing cycle, providing a sum of Google Cloud charges. Calendar mode VMs are tied to a reservation in which you pay for the full duration the VMs are reserved for. Prices on this page are listed in U.S. dollars (USD). Disk size and machine type memory are listed in gigabytes (GB), where 1 GB is 2^30 bytes.
Flex-start GPU VM Pricing
| Machine type | GPUs | vCPUs | Memory | Local SSD | DWS price (USD) |
|---|---|---|---|---|---|
| a4-highgpu-8g | 8 | 224 | 3,968 GB | 12,000 GiB | $64.44 / 1 hour |
| a3-ultragpu-8g | 8 | 224 | 2,952 GB | 12,000 GiB | $42.4 / 1 hour |
| a3-megagpu-8g | 8 | 208 | 1,872 GB | 6,000 GiB | $40.32 / 1 hour |
| a3-highgpu-1g | 1 | 26 | 234 GB | 750 GiB | $4.79 / 1 hour |
| a3-highgpu-2g | 2 | 52 | 468 GB | 1,500 GiB | $9.58 / 1 hour |
| a3-highgpu-4g | 4 | 104 | 936 GB | 3,000 GiB | $19.16 / 1 hour |
| a3-highgpu-8g | 8 | 208 | 1,872 GB | 6,000 GiB | $38.32 / 1 hour |
| a2-ultragpu-1g | 1 | 12 | 170 GB | 375 GiB | $2.4 / 1 hour |
| a2-ultragpu-2g | 2 | 24 | 340 GB | 750 GiB | $4.8 / 1 hour |
| a2-ultragpu-4g | 4 | 48 | 680 GB | 1,500 GiB | $9.6 / 1 hour |
| a2-ultragpu-8g | 8 | 96 | 1,360 GB | 3,000 GiB | $19.2 / 1 hour |
| a2-highgpu-1g | 1 | 12 | 85 GB | N/A | $2 / 1 hour |
| a2-highgpu-2g | 2 | 24 | 170 GB | N/A | $4 / 1 hour |
| a2-highgpu-4g | 4 | 48 | 340 GB | N/A | $8 / 1 hour |
| a2-highgpu-8g | 8 | 96 | 680 GB | N/A | $16 / 1 hour |
| a2-megagpu-16g | 16 | 96 | 1,360 GB | N/A | $32 / 1 hour |
| g4-standard-48 | 1 | 48 | 180 GB | 1,500 GiB | $2.25 / 1 hour |
| g4-standard-96 | 2 | 96 | 360 GB | 3,000 GiB | $4.5 / 1 hour |
| g4-standard-192 | 4 | 192 | 720 GB | 6,000 GiB | $9 / 1 hour |
| g4-standard-384 | 8 | 384 | 1,440 GB | 12,000 GiB | $18 / 1 hour |
| All other GPUs | See on-demand pricing |
Flex-start TPU VM Pricing
| Cloud TPU Version | DWS price (USD) |
|---|---|
| TPU v6e | $1.35 / 1 hour |
| TPU v5p | $2.10 / 1 hour |
| TPU v5e | $0.60 / 1 hour |
Price shown is per chip. To learn more about Google Cloud TPU accelerators see the Cloud TPU page.
Flex-start CPU VM Pricing
| Machine Type | vCPUs | Memory | Default egress bandwidth | DWS price (USD) |
|---|---|---|---|---|
| h4d-standard-192 | 192 | 720 GB | Up to 200 Gbps | $5.89 / 1 hour |
| h4d-highmem-192 | 192 | 1,488 GB | Up to 200 Gbps | $6.63 / 1 hour |
| h4d-highmem-192-lssd | 192 | 1,488 GB | Up to 200 Gbps | $7.07 / 1 hour |
Calendar Mode GPU VM Pricing
| Machine Type | GPUs | vCPUs | Memory | Local SSD | DWS price (USD) |
|---|---|---|---|---|---|
| a4-highgpu-8g | 8 | 224 | 3,968 GB | 12,000 GiB | $90.22 / 1 hour |
| a3-ultragpu-8g | 8 | 224 | 2,952 GB | 12,000 GiB | $59.36 / 1 hour |
| a3-megagpu-8g | 8 | 208 | 1,872 GB | 6,000 GiB | $44 / 1 hour |
| a3-highgpu-8g | 8 | 208 | 1,872 GB | 6,000 GiB | $41.6 / 1 hour |
Calendar mode only supports the 8 GPU shapes (e.g. a4-highgpu-8g). To learn about more about Google Cloud GPU accelerators see the Accelerator optimized machine family page.
Calendar Mode TPU VM Pricing
| Cloud TPU Version | DWS price (USD) |
|---|---|
| TPU v6e | $1.89 / 1 hour |
| TPU v5p | $2.94 / 1 hour |
| TPU v5e | $0.84 / 1 hour |
Price shown is per chip. To learn about more about Google Cloud TPU accelerators see the Cloud TPU page.
Calendar Mode CPU VM Pricing
| Machine Type | vCPUs | Memory | Default egress bandwidth | DWS price (USD) |
|---|---|---|---|---|
| h4d-standard-192 | 192 | 720 GB | Up to 200 Gbps | $6.28 / 1 hour |
| h4d-highmem-192 | 192 | 1,488 GB | Up to 200 Gbps | $7.07 / 1 hour |
| h4d-highmem-192-lssd | 192 | 1,488 GB | Up to 200 Gbps | $7.54 / 1 hour |
Take the next step
- Compare DWS list prices to on-demand, committed use discounts
- See DWS documentation to learn more
- Get started with DWS and learn about their provisioning models
- Use the Cloud Billing Catalog API for programmatic access to SKU information.
Request a custom quote
Connect with our sales team to get a custom quote for your organization.