G2

Accelerator Optimized

Purpose-built for machine learning training, inference, and GPU-accelerated workloads. Powered by NVIDIA GPUs with high-bandwidth interconnects.

Specifications

vCPUs

4-96

Memory

16-384 GB

Network

Up to 100 Gbps

Category	Metric	Result	Source
gpu	GPU Count (L4)	8 GPUs	GCP Documentation
gpu	GPU Memory (Total)	192 GB GDDR6	GCP Documentation
gpu	FP8 Performance	1.98 PFLOPS	NVIDIA L4 Specs
gpu	Inference Latency (Llama 2 7B)	23.8 ms/token	Community Benchmarks
gpu	Training Throughput (ResNet-50)	2,400 images/s	Community Benchmarks
cpu	Geekbench 6 Single-Core	1,540 points	Geekbench Browser
cpu	Geekbench 6 Multi-Core	9,800 points	Geekbench Browser
network	Max Bandwidth	100 Gbps	GCP Documentation

Best result per workload category (mixed units -- see table below for details)

Workload	Metric	Machine Type	Result	Notes
ML Inference	Llama 2 7B Tokens/sec	g2-standard-8	42 tokens/s	Llama 2 7B, FP16, single L4 GPU, vLLM
ML Inference	ResNet-50 Images/sec	g2-standard-8	850 images/s	ResNet-50, batch size 64, FP16, single L4 GPU
ML Inference	Stable Diffusion Images/min	g2-standard-8	8.5 images/min	Stable Diffusion XL, 1024x1024, 30 steps, single L4 GPU
ML Training	GPT-2 Training Throughput	g2-standard-96	5,400 tokens/s	GPT-2 medium, 8x L4, DeepSpeed ZeRO-2, FP16
ML Training	ResNet-50 Training Images/sec	g2-standard-96	2,400 images/s	ResNet-50, 8x L4, mixed precision, PyTorch DDP
Video Encoding	FFmpeg H.265 Encoding (NVENC)	g2-standard-8	240 fps	1080p source, NVENC hardware encoding, single L4 GPU

On-demand pricing in us-central1. Spot and committed-use discounts shown for comparison.

Machine Type	vCPUs	Memory	On-Demand/hr	Spot/hr	1yr CUD/hr	3yr CUD/hr
g2-standard-4	4	16 GB	$0.784	$0.235	$0.494	$0.353
g2-standard-8	8	32 GB	$0.918	$0.275	$0.578	$0.413
g2-standard-24	24	96 GB	$1.853	$0.556	$1.167	$0.834
g2-standard-96	96	384 GB	$10.416	$3.125	$6.562	$4.687

208 vCPUs

1872 GB

Up to 1800 Gbps (GPU-to-GPU)