Comprehensive performance benchmarks across every Compute Engine VM family. CPU, memory, network, storage, and GPU -- measured and compared.
Balanced compute, memory, and networking for a wide range of workloads. From web servers to development environments, these VMs deliver predictable performance at competitive price points.
Geekbench 6 -- higher is better
Geekbench 6 -- higher is better
Relative strengths across categories (normalized)
| Series | vCPUs | Memory | Network |
|---|---|---|---|
| E2 | 2-32 | 1-128 GB | Up to 16 Gbps |
| N2 | 2-128 | 2-864 GB | Up to 100 Gbps |
| N2D | 2-224 | 2-896 GB | Up to 100 Gbps |
| Tau T2D | 1-60 | 4-240 GB | Up to 32 Gbps |
Highest per-core performance on Google Cloud. Built for compute-intensive workloads that need maximum single-thread speed and sustained clock frequencies.
Geekbench 6 -- higher is better
Geekbench 6 -- higher is better
Relative strengths across categories (normalized)
| Series | vCPUs | Memory | Network |
|---|---|---|---|
| C3 | 4-176 | 8-1408 GB | Up to 100 Gbps |
| C3D | 4-360 | 8-2880 GB | Up to 100 Gbps |
| H3 | 88 | 352 GB | 200 Gbps |
Designed for workloads that need large amounts of memory relative to vCPUs. Ideal for in-memory databases, real-time analytics, and large caching layers.
Relative strengths across categories (normalized)
| Series | vCPUs | Memory | Network |
|---|---|---|---|
| M3 | 32-128 | 960-3904 GB | Up to 100 Gbps |
Purpose-built for machine learning training, inference, and GPU-accelerated workloads. Powered by NVIDIA GPUs with high-bandwidth interconnects.
Relative strengths across categories (normalized)
| Series | vCPUs | Memory | Network |
|---|---|---|---|
| A3 | 208 | 1872 GB | Up to 1800 Gbps (GPU-to-GPU) |
| G2 | 4-96 | 16-384 GB | Up to 100 Gbps |
Optimized for workloads that require high storage throughput and IOPS. Local SSDs deliver ultra-low latency for database engines and data-intensive applications.
Relative strengths across categories (normalized)
| Series | vCPUs | Memory | Network | Storage IOPS |
|---|---|---|---|---|
| Z3 | 88-176 | 704-1408 GB | Up to 100 Gbps | Up to 2.4M read IOPS |
Synthetic benchmarks tell part of the story. Here's how GCP VMs handle actual workloads -- from web serving to ML training.
Static file serving, API throughput, and request handling performance for web applications.
Top performer
C3 -- 82,000 req/s
Transactional throughput, query performance, and in-memory operations for database workloads.
Top performer
C3 -- 420,000 ops/s
Model inference speed for LLMs, image classification, and generative AI workloads.
Top performer
A3 -- 4,200 images/s
Training throughput for deep learning models using GPU-accelerated compute.
Top performer
A3 -- 48,000 tokens/s
Video transcoding and encoding performance for media processing pipelines.
Top performer
G2 -- 240 fps
High-performance computing benchmarks including LINPACK and CFD simulations.
Top performer
H3 -- 2,150 GFLOPS
Performance per dollar -- which VM gives you the most bang for your buck? Higher ratio means better value.
Select your workload type below. We'll recommend the best VM series based on performance data and price-performance analysis.
Web servers and API backends need fast single-core performance, consistent network throughput, and efficient handling of concurrent connections. The right VM depends on whether you're serving static content, running a dynamic API, or both.
Maximum throughput and lowest latency
Highest single-core performance on GCP translates directly to the fastest request handling. The 4th-gen Intel Xeon Scalable processors deliver 82K Nginx req/s -- 21% more than the next best option.
Production web servers with good price-performance
Strong performance at a lower price point than C3. 68K Nginx req/s covers most production web workloads comfortably. Up to 100 Gbps network bandwidth for traffic-heavy sites.
Cost-optimized scale-out deployments
Best price-per-request for steady-state workloads. Tau VMs are optimized for scale-out architectures where you run many instances behind a load balancer.
E2 VMs offer the lowest cost for development, staging, and low-traffic production sites. Performance is adequate for most sites under 1000 concurrent users.
Peak performance across the entire Compute Engine lineup.
Sort by any metric to find the best VM series for your workload. Click column headers to sort.
Series | Family | Single-Core ↓ | Multi-Core | Network (Gbps) | Memory BW (GB/s) |
|---|---|---|---|---|---|
| C3 | Compute Optimized | 1,850 | 14,200 | 100 | 45.8 |
| H3 | Compute Optimized | 1,720 | 22,400 | 200 | 58.2 |
| Z3 | Storage Optimized | 1,700 | 20,100 | 100 | 48.6 |
| C3D | Compute Optimized | 1,680 | 18,500 | 100 | 52.3 |
| G2 | Accelerator Optimized | 1,540 | 9,800 | 100 | -- |
| N2 | General Purpose | 1,520 | 8,750 | 100 | 32.4 |
| M3 | Memory Optimized | 1,480 | 11,600 | 100 | 68.4 |
| N2D | General Purpose | 1,410 | 12,800 | 100 | 38.6 |
| Tau T2D | General Purpose | 1,350 | 9,200 | 32 | -- |
| E2 | General Purpose | 1,180 | 4,520 | 16 | 18.2 |
| A3 | Accelerator Optimized | -- | -- | 1,800 | -- |