a2-ultragpu-4g
$20.2752/hr
~$14800.89/month on-demand in us-central1
The a2-ultragpu-4g is an Accelerator optimized instance with 48 vCPUs and 680 GiB of memory, 4x NVIDIA nvidia-tesla-a100 with up to 32 Gbps bandwidth, starting at $20.275 per hour or $17.900 per hour with spot instances.
Last updated: Jan 14, 2026, 09:17 PM UTC
Pricing Options
CUD
Committed Use Discounts
1 Year
$18.6475/hr
~$13612.71/mo
Save 8%
3 Year
$17.4164/hr
~$12713.94/mo
Save 14%
Spot Pricing
Up to 90% savings - flexible workloads
Spot Price
$17.8999/hr
~$13066.96/mo
Save 12%
a2-ultragpu family
Switch between sizes in us-central1
| Size | vCPU | Mem |
|---|---|---|
| a2-ultragpu-1g | 12 | 170 |
| a2-ultragpu-2g | 24 | 340 |
| a2-ultragpu-4g | 48 | 680 |
| a2-ultragpu-8g | 96 | 1360 |
Instance Specifications
Detailed hardware and capabilities for a2-ultragpu-4g
Processor
Architecturex86_64
vCPUs48
Memory680 GiB
Memory per vCPU14.17 GiB
Network
PerformanceUp to 32 Gbps
Bandwidth32 Gbps
Max Interfaces8
Enhanced Networking
Storage
Local Storage1500 GB
Storage TypeLocal SSD
NVMe Drive
GPU
Count4
ManufacturerNVIDIA
Modelnvidia-tesla-a100
Memory80 GiB
Total GPU Memory320 GiB
Capabilities
Familya2-ultragpu
GenerationCurrent
Burstable
Regional Pricing Comparison
Compare a2-ultragpu-4g (Linux) pricing across all regions - sorted by cost
Current region (us-central1)Other regions
💡 Click any bar to open that region in a new tab