g4dn.metal


Amazon EC2 instance g4dn.metal with 96 vCPUs, 384 GiB RAM and 8 x NVIDIA T4 16 GiB. Available in 22 regions starting from $5711.52 per month. A -38% cheaper alternative is available.
Instance Typeg4dn.metal
Instance FamilyGPU instance
DetailsMachine Learning (ML), 3D visualizations, graphics-heavy remote workstations, 3D rendering, application streaming, video encoding, and server-side graphics tasks.

g – GPU (Graphics Processing Unit)
4 – Generation
d – Instance store volume
n – Network and EBS optimized
metal – Bare metal
vCPUs96
Memory384 GiB
CPU Architecturex86_64
ProcessorIntel Xeon Family
Clock Speed2.5 GHz
Threads Per Core2
Accelerators
GPUyes
GPU NameT4
GPU ManufacturerNVIDIA
GPUs8
GPU Memory16 GiB
GPU Total Memory128 GiB
Inference Acceleratorno
Accelerator Memory128 GiB
FPGAno
Instances
Sizesg4dn.xlarge, g4dn.2xlarge, g4dn.4xlarge, g4dn.8xlarge, g4dn.12xlarge, g4dn.16xlarge, g4dn.metal
Free Tierno
Burstableno
Hibernationno
Instance Storage2 x 900 GB NVMe SSD
EBS (Elastic Block Store)
EBS Optimizeddefault
Encryptionsupported
NVMerequired
Dedicated Throughput14000 MBps
Optimized Baseline Bandwidth19000 Mbps
Optimized Baseline Throughput2375 Mbps
Optimized Baseline Performance80000 IOPS
Optimized Max Performance80000 IOPS
Optimized Max Bandwidth19000 Mbps
Optimized Max Throughput2375 Mbps
Networking
Networking Performance100 Gigabit
Enhanced Networkingyes
Max Interfaces15
Max Cards1
IPv4 Addresses per Interface50
IPv6 Addresses per Interface50
IPv6 Supportyes
ENA Supportrequired
EFA Supportyes
EFA Max Interfaces1
Encryption in Transit Supportyes
ENA SRD Supportno

Regional Availability

US
GovCloud (US-East) / us-gov-east-1
GovCloud (US-West) / us-gov-west-1
US East (Ohio) / us-east-2
US East (Virginia) / us-east-1
US West (N. California) / us-west-1
US West (Oregon) / us-west-2
Central America
Mexico (Central) / mx-central-1
Canada
Canada (Central) / ca-central-1
Canada West (Calgary) / ca-west-1
Europe
Europe (Ireland) / eu-west-1
Europe (London) / eu-west-2
Europe (Frankfurt) / eu-central-1
Europe (Milan) / eu-south-1
Europe (Paris) / eu-west-3
Europe (Spain) / eu-south-2
Europe (Stockholm) / eu-north-1
Europe (Zurich) / eu-central-2
Africa
Africa (Cape Town) / af-south-1
South America
South America (São Paulo) / sa-east-1
Asia Pacific
Asia Pacific (Hong Kong) / ap-east-1
Asia Pacific (Hyderabad) / ap-south-2
Asia Pacific (Jakarta) / ap-southeast-3
Asia Pacific (Malaysia) / ap-southeast-5
Asia Pacific (Melbourne) / ap-southeast-4
Asia Pacific (Mumbai) / ap-south-1
Asia Pacific (New Zealand) / ap-southeast-6
Asia Pacific (Osaka) / ap-northeast-3
Asia Pacific (Seoul) / ap-northeast-2
Asia Pacific (Singapore) / ap-southeast-1
Asia Pacific (Sydney) / ap-southeast-2
Asia Pacific (Taipei) / ap-east-2
Asia Pacific (Thailand) / ap-southeast-7
Asia Pacific (Tokyo) / ap-northeast-1
Middle East
Israel (Tel Aviv) / il-central-1
Middle East (Bahrain) / me-south-1
Middle East (UAE) / me-central-1

Regional Prices

US Dollar ($)
Per Hour
Linux
Shared Tenancy
On Demand
Geography
Region Name
Region
Savings Options
Instance Price
USUS East (Virginia)us-east-1compare7.8240
USUS East (Ohio)us-east-2compare7.8240
USUS West (Oregon)us-west-2compare7.8240
HiddenHiddenhidden-1compare8.3000
HiddenHiddenhidden-2compare8.6120
HiddenHiddenhidden-3compare8.6870
HiddenHiddenhidden-4compare8.7240
HiddenHiddenhidden-5compare9.1480
HiddenHiddenhidden-6compare9.1540
HiddenHiddenhidden-7compare9.1600
HiddenHiddenhidden-8compare9.3890
HiddenHiddenhidden-9compare9.5960
HiddenHiddenhidden-10compare9.6240
HiddenHiddenhidden-11compare9.7800
HiddenHiddenhidden-12compare9.8620
HiddenHiddenhidden-13compare10.1740
HiddenHiddenhidden-14compare10.3810
HiddenHiddenhidden-15compare10.5620
HiddenHiddenhidden-16compare10.5620
HiddenHiddenhidden-17compare10.9480
HiddenHiddenhidden-18compare12.0480
HiddenHiddenhidden-19compare13.2980
Activate subscription to see all regions and unlock all features  

Best AI models you can run on this instance

Top open-weight chat models ranked by Intelligence Index that fit in g4dn.metal's 128 GB of GPU memory, assuming the model is sharded across all 8 GPUs. VRAM is estimated from the parameter count at the selected quantization, so treat it as guidance rather than a guarantee.

Chat Models
FP16 (full precision)
Model
Creator
Intelligence
Total params
Active params
~VRAM
Qwen3.6 27BAlibaba logoAlibaba37.1#5427B—65 GB
Qwen3.5 27BAlibaba logoAlibaba33.8#7727B—65 GB
Qwen3.6 35B A3BAlibaba logoAlibaba31.6#9535B3B84 GB
7 more models fit on g4dn.metal
Unlock the full ranked list and FP8 / INT4 quantization with a CloudPrice subscription.

Similar Alternative Instances

Find Similar Instances In:
Instance Type
Region
vCPUs
Memory (GiB)
GPU Name
Instance Price
Price Diff %
m7i.metal-24xlUS East (Virginia) (us-east-1)963844.8384-38%
m5n.metalUS East (Virginia) (us-east-1)963845.7120-27%
m5dn.metalUS East (Virginia) (us-east-1)963846.5280-17%
m8azn.metal-24xlUS East (Virginia) (us-east-1)963849.910127%
m8gb.metal-24xlUS East (Virginia) (us-east-1)9638413.968079%
Select your cookie preferences
We use cookies and similar tools to enhance your experience, provide our services, and make improvements. Approved third parties also use these tools to certain site features. More details available here Cookie Policy