Google Cloud G4 machine types
G4 is a gpu & accelerated Compute Engine family with 7 machine types, from 6 – 384 vCPUs and 22 GiB – 1,440 GiB of memory with NVIDIA RTX PRO 6000 Blackwell accelerators, running on AMD Turin. Linux on demand starts at $0.5596 per hour.
gpu & acceleratedx86-64
- sizes
- 7
- vCPUs
- 6 – 384
- memory
- 22 GiB – 1,440 GiB
- accelerator
- RTX PRO 6000 Blackwell
- on-demand from
- $0.5596
per hour · 23 regions
G4 sizes and prices
Linux, hourly, Iowa (other regions on each size's page)
| Size | vCPU | Memory | Accelerator | Local storage | On-demand | Monthly | Spot | 1-yr |
|---|---|---|---|---|---|---|---|---|
| g4-standard-6 | 6 | 22 GiB | 1 × NVIDIA RTX PRO 6000 Blackwell | – | $0.5596 | $408 | $0.2091 | $0.3861 |
| g4-standard-12 | 12 | 45 GiB | 1 × NVIDIA RTX PRO 6000 Blackwell | – | $1.125 | $821 | $0.4202 | $0.7763 |
| g4-standard-24 | 24 | 90 GiB | 1 × NVIDIA RTX PRO 6000 Blackwell | – | $2.25 | $1,642 | $0.8404 | $1.553 |
| g4-standard-48 | 48 | 180 GiB | 1 × NVIDIA RTX PRO 6000 Blackwell | – | $4.50 | $3,285 | $1.681 | $3.105 |
| g4-standard-96 | 96 | 360 GiB | 2 × NVIDIA RTX PRO 6000 Blackwell | – | $9.00 | $6,570 | $3.362 | $6.21 |
| g4-standard-192 | 192 | 720 GiB | 4 × NVIDIA RTX PRO 6000 Blackwell | – | $18.00 | $13,140 | $6.723 | $12.42 |
| g4-standard-384 | 384 | 1,440 GiB | 8 × NVIDIA RTX PRO 6000 Blackwell | – | $36.00 | $26,280 | $13.45 | $24.84 |
G4 equivalents on other clouds
closest match to g4-standard-48 (48 vCPU, 180 GiB)
AWSg7e.8xlargeG7e family · 32 vCPU · 256 GiB · 1 × NVIDIA RTX PRO 6000 Blackwell$5.268/hr · US East (N. Virginia)G7e family →
Other gpu & accelerated families on Google Cloud
G4 FAQ
What sizes does the Google Cloud G4 family have?
G4 has 7 machine types, from g4-standard-6 (6 vCPU, 22 GiB) to g4-standard-384 (384 vCPU, 1,440 GiB). They run on AMD Turin.
What is the cheapest G4 machine type?
g4-standard-6 at $0.5596 per hour (Linux, on demand) in Iowa, about $408 per month.
What is G4 best for?
G4 is a gpu & accelerated family. GPUs and AI accelerators for training, inference, rendering and video.