50 Gbps guaranteed bandwidth

50Gbps Dedicated Servers Sized by Flow Concurrency

A 50Gbps port is filled by the number of flows crossing it, not by the size of any one of them. Five hardware configurations, one guaranteed port.

X-ZoneServers 50 Gbps dedicated servers start at €2799 per month for a guaranteed 50 Gbps port — 5 bare-metal configurations with up to 16.2 PB of monthly transfer across 12 EU and US datacenter locations. The port is effectively unmetered: hold the full committed rate around the clock, with no overage charges and no fair-use throttling inside your tier.

2799
starting price per month
55.98
per Gbps per month
5
configurations
16.2 PB
transfer ceiling per month

Traffic shape, not raw capacity, decides how much of this port you actually collect. A single TCP connection is handled by one receive queue and therefore one CPU core, and one core will not push anywhere near line rate however fast the disks are or however short the network path is. At 25Gbps a small number of well tuned streams can still get most of the way. At 50Gbps that stops being true, and the composition of your traffic becomes the deciding input. Spread the same volume across thousands of simultaneous sessions and the picture inverts: receive-side scaling hashes every flow into its own queue, the queues raise interrupts on separate cores, and the port fills evenly. So the first question to answer here is not how many terabytes you move in a month, it is how many connections you hold open when demand peaks.

What the Port Delivers, and What It Takes to Get There

The port carries 6250 MB/s of payload, which is 16.2 PB over a 30-day month at full utilisation. Reaching that ceiling is arithmetic on sessions rather than on hardware. Ten thousand simultaneous 1080p streams at 5 Mbps get there, and so do 2000 concurrent 4K streams at 25 Mbps, because every viewer is a separate flow with its own queue placement. A single 1 TB archive clears in 2 min 40 s at line rate, but only if the copy is genuinely running at line rate, which means splitting it across parallel streams before the stopwatch means anything at all. The figure worth writing on your capacity plan is therefore not gigabits. It is flows in flight at peak.

16.2 PB
Transfer per month
Sustained volume at full port utilisation across a 30-day month.
10,000
1080p viewers
Simultaneous streams at 5 Mbps each.
2,000
4K viewers
Simultaneous streams at 25 Mbps each.
2 min 40 s
Per terabyte
Wall-clock time to move 1 TB off the server at line rate.

Figures use decimal units (1 Gbps = 125 MB/s) across a 30-day month, at 5 Mbps per 1080p stream and 25 Mbps per 4K stream. They describe what the port can carry, not a guarantee of application throughput: disk, CPU and the far end all apply their own limits.

5 configurations at 50Gbps

The tier opens with the Dual Xeon Gold 6148 at €2799, 40 cores and 80 threads at 2.4GHz, which gives a broad connection pool somewhere to spread its interrupt load. Single EPYC 9254 at €3869 trades breadth for speed with 24 cores at 2.9GHz on DDR5-4800 and ten NVMe slots, a better fit when you have fewer but heavier flows and per-connection throughput is the constraint. Dual EPYC 9254 at €4099 doubles that to 48 cores. Above them sit Dual EPYC 9554 at €4379 with 128 cores and 512GB, and Dual EPYC 9754 at €4939 with 256 cores, 512 threads, 1024GB of DDR5-4800 and 24 NVMe slots. Choose upward when the connection table is enormous and you want queue interrupts landing on cores that are not already busy serving requests.

Dual Xeon Gold 6148

Ultimate Performance

2799
/month
40c / 80t @ 2.4GHz256GB DDR4 @ 2133MHz
6x SSD Slots50Gbps Guaranteed
AI Optimized Routing Included
Order Now

Single EPYC 9254

Entry Level

3869
/month
24c / 48t @ 2.9GHz128GB DDR5 @ 4800MHz
10x NVMe Slots50Gbps Guaranteed
AI Optimized Routing Included
Order Now

Dual EPYC 9254

High Performance

4099
/month
48c / 96t @ 2.9GHz256GB DDR5 @ 4800MHz
10x NVMe Slots50Gbps Guaranteed
AI Optimized Routing Included
Order Now

Dual EPYC 9554

Ultimate Performance

4379
/month
128c / 256t @ 3.1GHz512GB DDR5 @ 4800MHz
24x NVMe Slots50Gbps Guaranteed
AI Optimized Routing Included
Order Now
ULTIMATE

Dual EPYC 9754

Maximum Performance

4939
/month
256c / 512t @ 2.3GHz1024GB DDR5 @ 4800MHz
24x NVMe Slots50Gbps Guaranteed
AI Optimized Routing Included
Order Now

What a 50Gbps server is for

CDN edge and cache nodes

Cache nodes keep tens of thousands of short HTTP connections open at once, which is precisely the traffic shape a wide port wants. Object requests are small, numerous and independent, so the connection count does the distribution work for you.

Live streaming delivery

Every viewer session is an independent flow that lands in its own queue, so audience size spreads itself across cores. The number of people watching at the same moment, and not the encoder count, is what sizes this port.

Multiplayer and real-time backends

Game, voice and messaging fleets hold vast numbers of small chatty connections. Per-session throughput is tiny, but the aggregate arrives already distributed across receive queues, which is the friendliest shape a wide port can be handed.

Backup and object storage endpoints

These behave well when hundreds of clients pull in parallel. The same endpoint fed by one restore job at a time will sit at a fraction of the port, and that is the sizing trap worth checking before you order.

Paying for a Port You Can Actually Fill

The entry configuration works out at €55.98 per guaranteed Gbps. That number only means something if your traffic can reach the top of the port, and reaching it is a property of your workload rather than of the invoice. A service holding thousands of connections in flight converts the whole port into delivered bytes and the figure holds. A service moving the same monthly volume through a dozen long transfers pays for exactly the same port and leaves much of it unused, because those flows cannot be spread wide enough to keep every queue busy. Count the concurrent sessions your application sustains at peak before you commit. Bandwidth here is guaranteed and committed per tier with no overage charges inside the tier, so the sizing risk is under-utilisation, not a surprise bill.

All guaranteed bandwidth tiers

The same hardware range is sold at nine committed port speeds. Compare the entry price of each.

50Gbps dedicated server questions

How many concurrent connections does it take to fill a 50Gbps port?
There is no fixed number, because it depends on what each connection sustains. Divide 50Gbps by the average per-flow rate your application achieves and you have your answer. Sessions that trickle along at a few megabits need tens of thousands of them, while sessions that each pull several hundred megabits need only a few hundred. What matters more than the count is that it comfortably exceeds the number of receive queues on the card, so no queue sits idle while another is saturated.
Why does one large file transfer not fill the port?
Two things cap it. The flow's own congestion window and round-trip time limit how much data can be in flight over a long path, and all of its packets hash to a single receive queue, so one core does the interrupt work for the entire transfer. Neither limit is a policy on our side and neither lifts when the port gets wider. The fix is parallelism: most modern transfer tools open several streams for one job, and those streams hash to different queues.
What are receive-side scaling and multi-queue NICs?
A multi-queue card presents several independent receive queues instead of one. Receive-side scaling hashes each arriving packet by its connection tuple and drops it into one of those queues, and each queue raises interrupts on its own CPU core. Because the hash is per-flow, every packet of a given connection lands in the same queue, so ordering is preserved. The practical consequence is that throughput scales with the number of distinct flows, not with how hard any single one pushes.
Should I size for many small sessions or a few large ones?
Count what your peak actually looks like. Many small sessions is the easy case, because they distribute themselves and you can size on aggregate bandwidth alone. A few large transfers need deliberate work: splitting each job into parallel streams, enlarging socket buffers for the path length, and keeping the handling processes near the cores that serve their queues. If your traffic really is a handful of big flows you cannot parallelise, a narrower tier will deliver the same throughput for less money.
How do I tell whether my traffic is spreading across queues?
Look at the per-queue statistics your operating system exposes for the network interface. If one queue is carrying nearly all the packets while the others sit near zero, your traffic is concentrated in too few flows and adding bandwidth will change nothing. If the counters are broadly even and every core carries a share of the interrupt work, the port is being used as designed, and the remaining limit is genuine capacity.

50 Gbps dedicated servers by location

Same hardware and guaranteed port, deployed from the datacenter closest to your users.

Tell Us Your Peak Concurrency

Send us your peak concurrent session count and your typical per-flow rate. Our own system administrators, on hand 24/7, will work the arithmetic with you and quote a configuration against it. Deployment takes about an hour once the specification is agreed.

Up to 1Tbps DDoS protectionDeployed in about 1 hour99.9% uptime SLA