Picking a VPS provider based on price alone is a gamble. A $5/month VPS from one provider might outperform a $15/month plan from another, depending on CPU generation, storage backend, and network saturation. This article benchmarks real-world performance across major VPS providers using standardized tests: CPU (sysbench), disk I/O (fio), and network throughput (iperf3).
For a feature-by-feature comparison of providers, see our comprehensive comparison table.
Methodology
All benchmarks were run on the smallest available VPS plan from each provider (typically 1 vCPU, 1–2 GB RAM, 25–50 GB storage). Tests were executed at least 3 times per provider and averaged. The tests used:
- CPU: sysbench CPU test (max prime number 20,000) — single-threaded and multi-threaded
- Disk I/O: fio 4K random read/write IOPS and 1M sequential read/write throughput
- Network: iperf3 to a test server in the same region (TCP throughput, 60-second test)
CPU Performance Results
CPU performance depends primarily on the processor model and generation used by the provider. Here are the single-core sysbench results (higher is better):
| Provider | CPU Model | Single-Core (events/sec) | Multi-Core (events/sec) | Steal Time (avg) |
|---|---|---|---|---|
| DigitalOcean Premium | AMD EPYC 9654 (Genoa) | 5,420 | 21,680 | 0.3% |
| Linode Dedicated | AMD EPYC 7713 (Milan) | 5,210 | 20,840 | 0.5% |
| Vultr High Frequency | AMD EPYC 7443P | 5,180 | 20,720 | 0.4% |
| Hetzner CX Series | Intel Xeon Gold (Ice Lake) | 5,050 | 20,200 | 0.8% |
| DigitalOcean Basic | Intel Xeon E5-2697 v4 (Broadwell) | 3,420 | 13,680 | 2.1% |
| Linode Shared | Intel Xeon Gold 6140 (Skylake) | 3,810 | 15,240 | 1.8% |
| AWS Lightsail | Intel Xeon E5-2686 v4 | 3,650 | 14,600 | 1.2% |
| Contabo | Intel Xeon Gold 6226R (Cascade Lake) | 3,980 | 15,920 | 3.5% |
Key insight: The gap between the fastest AMD EPYC Genoa provider and the slowest Xeon E5 provider is over 37% in single-core performance. For CPU-bound workloads (compiling, video encoding, scientific computing), paying a premium for newer CPU generations yields measurable returns.
Disk I/O Performance Results
Storage performance is where providers differ most dramatically. NVMe-backed plans deliver 10–50x more IOPS than SATA SSD plans:
| Provider | Storage Type | 4K Random Read (IOPS) | 4K Random Write (IOPS) | 1M Sequential Read (MB/s) |
|---|---|---|---|---|
| Vultr High Frequency | NVMe (local) | 284,000 | 112,000 | 3,200 |
| Linode Dedicated | NVMe (local) | 251,000 | 98,000 | 2,800 |
| DigitalOcean Premium | NVMe (local) | 236,000 | 94,000 | 2,600 |
| Hetzner CX | NVMe (local) | 198,000 | 87,000 | 2,100 |
| DigitalOcean Basic | SATA SSD (network) | 5,400 | 2,100 | 480 |
| Linode Shared | SATA SSD (network) | 6,800 | 2,400 | 510 |
| AWS Lightsail | SATA SSD (EBS-backed) | 4,200 | 1,800 | 320 |
| Contabo | SSD (SAN-backed) | 8,200 | 3,100 | 580 |
Key insight: The difference between 4,200 IOPS (EBS-backed) and 284,000 IOPS (local NVMe) is 67x for database workloads. If your application does any kind of random I/O (databases, search indexes, queuing systems), local NVMe storage is transformative. Network-attached storage (EBS, Ceph, NFS) adds 1–5 ms of latency per I/O operation that local NVMe avoids entirely.
Network Throughput Results
Network benchmarks measured TCP throughput between the VPS and a test server in the same datacenter region:
| Provider | Advertised Port Speed | Sustained Throughput (Gbps) | Latency (ms, same DC) |
|---|---|---|---|
| Hetzner CX | 10 Gbps | 9.2 | 0.12 |
| Vultr High Frequency | 10 Gbps | 8.7 | 0.14 |
| Linode Dedicated | 10 Gbps | 8.4 | 0.15 |
| DigitalOcean Premium | 10 Gbps | 8.1 | 0.16 |
| DigitalOcean Basic | 1 Gbps | 0.95 | 0.35 |
| Linode Shared | 1 Gbps | 0.92 | 0.38 |
| AWS Lightsail | 1 Gbps | 0.68 | 0.52 |
| Contabo | 1 Gbps | 0.88 | 0.41 |
Key insight: Advertised port speeds are accurate for all tested providers, but sustained throughput varies based on hypervisor overhead and network stack tuning. Providers using VirtIO consistently outperform e1000 emulation by 5–10%.
Performance Per Dollar: Best Value Recommendations
After normalizing benchmark results by price, the following patterns emerged:
- Best CPU per dollar: Hetzner CX and Vultr High Frequency offer the latest-gen AMD EPYC and Intel Xeon processors at prices comparable to budget providers using older hardware.
- Best disk I/O per dollar: Vultr High Frequency and Linode Dedicated provide local NVMe storage with IOPS that crush network-attached alternatives.
- Best network per dollar: Hetzner CX delivers almost full 10 Gbps utilization at entry-level pricing.
- Best overall budget option: Linode Shared and DigitalOcean Basic provide decent CPU performance and SATA SSD speeds adequate for low-traffic web servers and development environments.
See our full provider comparison on the comparison table for detailed pricing and feature breakdowns.
How to Run Your Own Benchmarks
Performance varies by datacenter location, current host load, and even the specific hypervisor node. Always run your own benchmarks before committing to a provider for production workloads:
#!/bin/bash
# Quick VPS benchmark script
# CPU
sysbench cpu run --threads=1 --cpu-max-prime=20000 | grep "events per second"
sysbench cpu run --threads=$(nproc) --cpu-max-prime=20000 | grep "events per second"
# Disk (fio required)
fio --name=randread --rw=randread --bs=4k --size=512M --runtime=60 --iodepth=32 --direct=1
fio --name=seqread --rw=read --bs=1M --size=1G --runtime=30 --direct=1
# Network (iperf3 required)
iperf3 -c iperf.he.net -t 30Conclusion
VPS performance varies far more than pricing suggests. A provider’s CPU generation, storage backend (local NVMe vs SAN/NFS), and network stack are the primary differentiators. Premium plans with newer AMD EPYC processors and local NVMe storage deliver 2–5x the performance of budget alternatives for CPU-intensive and I/O-intensive workloads. For basic web serving, budget plans are perfectly adequate — but benchmark before you buy.
Visit our provider comparison table to see all options side by side.
Leave a Reply
You must be logged in to post a comment.