a2-megagpu-16g specs and pricing | GCP

GCP a2-megagpu-16g - Accelerator Optimized: 16 NVIDIA Tesla A100 GPUs, 96 vCPUs, 1360GB RAM. Available in 3 regions starting from $40689.84 per month. A -86% cheaper alternative is available.

Instance Type	a2-megagpu-16g
Details	Optimized for accelerated high performance computing workloads. a – Accelerator optimized 2 – Generation megagpu – Large amound of GPU memory - 80x RAM/GPU ratio 16g – Number of GPUs
Instance Family	Accelerator optimized
Description	Accelerator Optimized: 16 NVIDIA Tesla A100 GPUs, 96 vCPUs, 1360GB RAM
vCPUs	96
Memory	1360 GiB
CPU Architecture	x86_64
Processor	Intel Cascade Lake
Coremark Score	1258852
Shared CPU	no
GPU	yes
GPUs	16
GPU Name	nvidia-tesla-a100
GPU Total RAM (GiB)	640
Sustained Discounts (SUDs)	no

Sole Tenancy	yes
Nested Virtualization	yes
Custom Shapes	no
Confidential Compute	no
Maximum Persistent Disks	128
Maximum Persistent Disks Size	524288 GB
Disk Interface Type	SCSI (PD, Local SSD) NVMe (Local SSD)
Local SSD	yes
Max Local SSD	3000 GB
Extreme Persistent Disks	no
Hyperdisk Extreme	no
Hyperdisk Throughput	no
Network Interfaces	gVNIC, VirtIO-Net

Regional Availability

North America
California / us-west2
Iowa / us-central1
Mexico / northamerica-south1
Nevada / us-west4
Ohio / us-east5
Ontario / northamerica-northeast2
Oregon / us-west1
Québec / northamerica-northeast1
South Carolina / us-east1
Texas / us-south1
Utah / us-west3
Virginia / us-east4

Europe
Belgium / europe-west1
England / europe-west2
Finland / europe-north1
France / europe-west9
Germany / europe-west10
Germany / europe-west3
Italy / europe-west12
Italy / europe-west8
Netherlands / europe-west4
Poland / europe-central2
Spain / europe-southwest1
Sweden / europe-north2
Switzerland / europe-west6

APAC
Australia / australia-southeast1
Australia / australia-southeast2
Hong Kong / asia-east2
India / asia-south1
India / asia-south2
Indonesia / asia-southeast2
Japan / asia-northeast1
Japan / asia-northeast2
Singapore / asia-southeast1
South Korea / asia-northeast3
Taiwan / asia-east1

Middle East
Israel / me-west1
Qatar / me-central1
Saudi Arabia / me-central2

Africa
South Africa / africa-south1

South America
Brazil / southamerica-east1
Chile / southamerica-west1

Regional Prices

US Dollar ($)

Per Hour

On Demand

Geography	Region Name	Region	Savings Options	Instance Price
North America	Iowa	us-central1	compare	55.7395
Europe	Netherlands	europe-west4	compare	56.6266
APAC	Singapore	asia-southeast1	compare	60.4762

Best AI models you can run on this instance

Top open-weight chat models ranked by Intelligence Index that fit in a2-megagpu-16g's 640 GB of GPU memory, assuming the model is sharded across all 16 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
DeepSeek V4 Flash	DeepSeek	40.3#28	158B	—	379 GB
MiniMax M2.7	MiniMax	38.1#42	229B	—	549 GB
Qwen3.6 27B	Alibaba	37.1#51	27B	—	65 GB
7 more models fit on a2-megagpu-16g 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)	Instance Price	Price Diff %
g2-standard-96	Iowa (us-central1)	96	384	8.0033	-86%
a2-highgpu-8g	Iowa (us-central1)	96	680	29.3871	-47%
a2-ultragpu-8g	Iowa (us-central1)	96	1360	40.2216	-28%
a3-edgegpu-8g	Iowa (us-central1)	208	1872	87.8325	58%
a3-edgegpu-8g-nolssd	Iowa (us-central1)	208	1872	87.8325	58%