Best 10Gbps VPS in 2026: I Ran iperf3 for 60 Seconds and None of Them Sustained Even 5Gbps

Last updated: March 2026 · By Alex Chen

Every VPS marketing page in 2026 has the same claim: "10Gbps network port." They say it like it means something. Like you are going to spin up a $6/mo cloud instance and saturate a 10-gigabit pipe.

I wanted to know what that claim actually means. So I did something apparently unreasonable: I rented instances from 5 providers advertising 10Gbps, installed iperf3, and ran a 60-second sustained throughput test with 8 parallel TCP streams against multiple endpoints. Not a 5-second burst. Not a speedtest.net screenshot. Sixty continuous seconds of iperf3 -c [target] -P 8 -t 60 -i 10, repeated three times at different hours.

The best result across all cloud VPS instances was 4.7Gbps sustained. Most hovered between 1.8 and 3.2Gbps. One provider that prominently advertises "10Gbps" on its pricing page delivered a consistent 1.2Gbps — which is basically a 1Gbps unmetered connection with extra marketing.

This is not a rant. Shared networking is how cloud economics work, and most workloads never need more than 1Gbps. But if you are reading this page, you probably have a specific reason to want multi-gigabit throughput — media streaming, large file distribution, database replication, CDN origin serving — and you deserve to know what you are actually buying.

Here is what I found, with the numbers to prove it.

The iperf3 Reality Check: Marketing vs. Measured Throughput

Before the individual reviews, let me show you the table that tells the whole story. Every test used iperf3 -P 8 -t 60 against three different endpoints (Serverius NL, Hetzner DE, Vultr-hosted in same region). The "sustained" column is the average of the final 30 seconds — not the initial burst, which every provider is happy for you to screenshot.

Read that "Sustained" column again. Vultr cloud VPS: 3.21Gbps from a "10Gbps port." Contabo: 1.24Gbps after the initial burst faded. The only providers that actually delivered near 10Gbps were running bare metal with dedicated NICs — Vultr at $120/mo and Hetzner at ~$54/mo.

Now let me walk through each provider so you understand why the numbers look the way they do.

#2. Hetzner — 9.1Gbps Sustained For Half the Price of Vultr (But Zero US Datacenters)

iperf3 sustained (dedicated): 9.12 Gbps · iperf3 sustained (cloud): 2.41 Gbps · Dedicated from: ~$54/mo · Locations: Falkenstein, Nuremberg, Helsinki (EU only)

Hetzner's dedicated server line (the AX series) is where 10Gbps actually happens on a budget. The AX42 I tested — AMD Ryzen 7 7700, 64GB RAM, 2x512GB NVMe — costs 49 EUR/mo (~$54) and includes a 10Gbps dedicated NIC with 20TB of included bandwidth. That is 5TB more than Vultr at less than half the price.

My iperf3 results were remarkably stable. Peak at 9.44Gbps, sustained at 9.12Gbps. Almost no variance between 3am and 3pm tests, which tells me the switch infrastructure is not congested. Hetzner runs their own AS (AS24940) with massive peering capacity, and it shows.

The iperf3 timeline that impressed me

$ iperf3 -c speedtest.serverius.net -p 5002 -P 8 -t 60 -i 10
[SUM]  0.0-10.0 sec  11.1 GBytes  9.44 Gbits/sec
[SUM] 10.0-20.0 sec  10.9 GBytes  9.31 Gbits/sec
[SUM] 20.0-30.0 sec  10.8 GBytes  9.22 Gbits/sec
[SUM] 30.0-40.0 sec  10.7 GBytes  9.18 Gbits/sec
[SUM] 40.0-50.0 sec  10.7 GBytes  9.14 Gbits/sec
[SUM] 50.0-60.0 sec  10.6 GBytes  9.12 Gbits/sec
- - - - - - - - - - - - - - - -
[SUM]  0.0-60.0 sec  65.8 GBytes  9.24 Gbits/sec  sender

That slow, gentle decline from 9.44 to 9.12 is TCP congestion control doing exactly what it should. No sudden drops. No throttling. No burst-then-cliff pattern. This is what an honest 10Gbps connection looks like.

Hetzner Cloud (the VPS product) is a completely different machine. The CPX31 I tested ($17/mo) sustained only 2.41Gbps despite the same "10Gbps" marketing on the specs page. The cloud product uses SR-IOV virtualization, which is efficient, but the host NIC is still shared with 30-50 other VMs. The 20TB bandwidth allowance is generous, but you are never going to use it at 2.41Gbps unless you run traffic 24/7 for the entire month.

Who should pick Hetzner: If your traffic is EU-centric or EU-to-EU, the AX42 dedicated server is the best 10Gbps value available anywhere. For US-targeted traffic, the latency tax kills your effective throughput to American end users regardless of how fast the port is. Full Hetzner review →

#3. UpCloud — The Most Consistent Shared 10Gbps I Tested (But You Pay For It)

iperf3 sustained: 3.47 Gbps · Peak: 4.31 Gbps · From: $46/mo (4 CPU, 8GB) · US locations: New York, Chicago, San Jose

UpCloud does not sell bare metal. Every instance is a cloud VPS. So the 10Gbps number on their specs page is always going to be a shared port. What makes UpCloud interesting is not the raw speed — 3.47Gbps sustained is good but not the highest — but the consistency.

Across 6 iperf3 runs over 48 hours (3am, 11am, 3pm on two consecutive days), UpCloud's throughput ranged from 3.21 to 3.74Gbps. That is a 14% variance window. Compare that to Vultr cloud (2.84 to 4.73Gbps, a 40% variance window) or Contabo (0.91 to 2.87Gbps, a 68% variance window). UpCloud's MaxIOPS storage technology gets all the marketing attention, but their network QoS is the real differentiator.

The peak-to-sustained ratio of 1.24:1 is also telling. UpCloud does not appear to use aggressive burst allowances that inflate short speed tests. What you see in the first 10 seconds is close to what you get in the last 10 seconds. That is a sign of honest bandwidth allocation, not marketing-optimized throttling profiles.

The catch

UpCloud is not cheap. The DEV-4xCPU-8GB plan at $46/mo with 5TB bandwidth gives you roughly the same sustained throughput as Vultr's $48/mo cloud instance (3.47 vs 3.21Gbps), but with slightly better consistency. If consistency matters for your workload — video streaming, API serving, real-time data feeds — the UpCloud premium is justified. If you just need burst speed for occasional large transfers, Vultr cloud is marginally cheaper for similar peak performance.

Who should pick UpCloud: Teams running network-sensitive workloads that need predictable throughput, not just peak speed. If your SLA depends on consistent multi-gigabit performance and bare metal is overkill, UpCloud's cloud VPS is the safest bet. Full UpCloud review →

#4. BuyVM — The Provider That Does Not Pretend (937Mbps Unmetered, and That Is the Point)

"We do not market 10Gbps because we sell 1Gbps unmetered and mean it. You get 1Gbps. All the time. No asterisks."

— Paraphrasing BuyVM's actual product positioning, which is refreshingly blunt

iperf3 sustained: 937 Mbps · Peak: 941 Mbps · From: $3.50/mo (1 CPU, 1GB) · US locations: Las Vegas, New York, Miami

BuyVM is on this list as the control group. They do not claim 10Gbps. They sell 1Gbps unmetered, and my iperf3 test measured 937Mbps sustained — 93.7% of advertised, maintained perfectly flat across 60 seconds. The peak-to-sustained ratio was 1.004:1. Basically a flat line. That is what "unmetered" looks like when the provider is not overselling.

Why include a 1Gbps provider on a 10Gbps list? Because 937Mbps unmetered for $3.50/mo transfers more total data than any 10Gbps provider on this page.

The math that matters

If your workload is about volume — backups, file mirrors, VPN traffic, media libraries — BuyVM at $3.50/mo is a better network deal than any 10Gbps provider at any price. The throughput is lower, but it never stops and it never costs extra.

BuyVM also includes DDoS protection via Path.net at no additional cost. The DDoS scrubbing is inline and does not measurably affect normal traffic throughput. I tested with and without an active Path.net filter and saw no difference in iperf3 numbers.

Who should pick BuyVM: Anyone whose real requirement is "move a lot of data cheaply" rather than "move data very fast in short bursts." File mirrors. Plex/Jellyfin servers. VPN endpoints. Offsite backup targets. BuyVM is also the honest answer for anyone tempted by 10Gbps marketing who has never actually measured their peak bandwidth. Full BuyVM review →

#5. Contabo — The Biggest Gap Between Marketing and Reality in This Roundup

iperf3 sustained: 1.24 Gbps · Peak: 2.87 Gbps · From: ~$21/mo (VDS S + 10G add-on) · US locations: St. Louis, New York, Seattle

Contabo offers a 10Gbps NIC add-on for their VDS (Virtual Dedicated Server) plans. It costs an extra $8.49/mo on top of the base VDS price. The VDS S plan with 4 dedicated CPU cores, 12GB RAM, and 100GB NVMe comes to about $21/mo total with the 10Gbps upgrade. On paper, that looks like an incredible deal. In practice, it was the most disappointing result in this roundup.

The burst-then-cliff pattern

$ iperf3 -c speedtest.serverius.net -p 5002 -P 8 -t 60 -i 10
[SUM]  0.0-10.0 sec   3.37 GBytes  2.87 Gbits/sec
[SUM] 10.0-20.0 sec   2.44 GBytes  2.08 Gbits/sec
[SUM] 20.0-30.0 sec   1.89 GBytes  1.61 Gbits/sec
[SUM] 30.0-40.0 sec   1.53 GBytes  1.31 Gbits/sec
[SUM] 40.0-50.0 sec   1.47 GBytes  1.25 Gbits/sec
[SUM] 50.0-60.0 sec   1.44 GBytes  1.24 Gbits/sec
- - - - - - - - - - - - - - - -
[SUM]  0.0-60.0 sec  12.1 GBytes  1.73 Gbits/sec  sender

Look at that decline. 2.87Gbps at the start, 1.24Gbps by the end. That is a 57% drop. The first 10 seconds look decent in a quick speedtest. By 40 seconds, the throughput has stabilized at barely above 1Gbps — which is what you get on Contabo's standard VPS without the 10Gbps add-on.

What is happening here is transparent: Contabo allows a burst that draws from the shared 10Gbps uplink, then the traffic shaper kicks in and throttles your VM back to roughly its fair share. The 10Gbps NIC add-on gives you a faster physical interface, but the bandwidth allocation policy behind it has not changed. You are still sharing the switch with dozens of other VDS customers.

In Contabo's defense

The 32TB monthly bandwidth allowance is enormous. Even at 1.24Gbps sustained, you would need to run traffic 24/7 for roughly 24 days to exhaust it. For workloads that need moderate speed (1-2Gbps) with high total volume, Contabo's VDS + 10G add-on at $21/mo is a legitimate option. Just do not expect the "10Gbps" in the product name to mean 10Gbps sustained throughput. It means 10Gbps port, 10Gbps burst, and roughly 1Gbps steady state.

Who should pick Contabo: Budget-conscious users who need 1-2Gbps burst capability with high transfer volume (32TB) and do not mind that the "10Gbps" label is aspirational. Not appropriate for workloads requiring sustained multi-gigabit throughput. Full Contabo review →

What "10Gbps" Actually Means: A Field Guide to Network Marketing

After running these tests, I have a taxonomy of what providers mean when they say "10Gbps." It is not the same thing across the board.

Tier 1: Dedicated 10Gbps (Vultr Bare Metal, Hetzner Dedicated)

You get a physical NIC on a physical switch port that is yours alone. Sustained throughput is 9.1-9.7Gbps. This is real 10Gbps. It costs $54-$120/mo and comes with bandwidth caps (15-20TB). If you need more transfer volume, overage fees apply.

Tier 2: Shared 10Gbps with decent QoS (Vultr Cloud, UpCloud)

The host machine has a 10Gbps NIC. Your VM accesses it through a virtual switch. The provider does not hard-cap your VM but uses traffic shaping to share fairly among tenants. Sustained throughput is 2.4-3.5Gbps depending on host load. Peak bursts can hit 4-5Gbps briefly. This is what most "10Gbps VPS" products actually deliver.

Tier 3: Shared 10Gbps with aggressive burst-then-throttle (Contabo)

The host has a 10Gbps NIC. Your VM gets a 10Gbps virtual interface. But the bandwidth policy allows a brief burst followed by hard throttling to 1-2Gbps. Short speed tests look impressive. Sustained tests reveal the real allocation. The "10Gbps" in the product name refers to the port hardware, not your usable bandwidth.

Tier 4: Honest 1Gbps (BuyVM)

No 10Gbps claims. 1Gbps port, 1Gbps delivered, unmetered. What you see is what you get. Paradoxically, this is the best value if your metric is "total bytes moved per dollar spent."

How to Choose: Match Your Workload to the Right Tier

How I Tested: Methodology and Why 60 Seconds Matters

Most VPS "speed tests" run for 10 seconds. That is the default iperf3 duration. It is also exactly what providers optimize for, because burst allowances are designed to make short tests look good.

Here is my testing protocol:

1. Tool: iperf3 version 3.16, compiled from source with --enable-tcp-window-size.
2. Command: iperf3 -c [target] -P 8 -t 60 -i 10. Eight parallel TCP streams (critical for saturating high-speed links where single-stream TCP window scaling is the bottleneck), 60-second duration, reporting every 10 seconds to capture the throughput curve over time.
3. Endpoints: Three different iperf3 servers for each test — Serverius (NL), Hetzner (DE), and a self-hosted iperf3 on a Vultr bare metal in the same region. Results reported are the average of the three.
4. Timing: Each test run repeated at 3am, 11am, and 3pm local datacenter time on two consecutive weekdays. Six runs per provider. Reported "sustained" figure is the average of the final 30 seconds across all six runs.
5. Configuration: Default OS network stack settings (Ubuntu 24.04 LTS). No sysctl tuning for TCP buffers, window sizes, or congestion control algorithms. I wanted to measure what you get out of the box, not what you can theoretically achieve after spending a day tuning net.core.rmem_max.

The 60-second duration is non-negotiable for honest testing. Here is why: run iperf3 -t 10 on Contabo's VDS with the 10Gbps add-on and you get ~2.87Gbps. Run iperf3 -t 60 and the last 10 seconds show 1.24Gbps. That 57% drop is invisible in a short test. Every provider on this page showed some throughput decline after 10 seconds. The degree of decline tells you how much of your "10Gbps" is burst and how much is real.

How to Run These Tests on Your Own VPS

Do not trust my numbers. Run the test yourself. Here is the exact procedure:

# Install iperf3
sudo apt update && sudo apt install -y iperf3

# Test 1: Sustained throughput (60 seconds, 8 parallel streams)
iperf3 -c speedtest.serverius.net -p 5002 -P 8 -t 60 -i 10

# Test 2: Same test against a different endpoint
iperf3 -c bouygues.iperf.fr -p 5200 -P 8 -t 60 -i 10

# Test 3: Reverse mode (download instead of upload)
iperf3 -c speedtest.serverius.net -p 5002 -P 8 -t 60 -i 10 -R

# Ongoing monitoring: install vnstat
sudo apt install -y vnstat
# Wait 24 hours, then check:
vnstat -h   # hourly traffic
vnstat -d   # daily traffic

Key things to watch for in your results:

• Compare first 10s to last 10s. A drop greater than 30% indicates burst-then-throttle behavior.
• Run at different times. A 3am test that shows 4Gbps and a 3pm test that shows 1.5Gbps means you are heavily affected by noisy neighbors on the shared host.
• Test upload AND download separately. Some providers cap upload more aggressively than download. Use the -R flag for reverse (download) tests.
• Use multiple endpoints. If you only test against one iperf3 server, a congested intermediate hop could bottleneck the result. Test against at least two geographically different endpoints.

Frequently Asked Questions