Regions, routes and ports
Two verified answers about where capacity sits, and no world map of cities we have never had a machine in.
Verified locations
This is the complete list. A region appears here once the first machine is live in it and comes off when the last one leaves — neither one early. A map that only ever gains flags is a map nobody is maintaining.
- Dedicated servers
Three regions, one country
All 46 bare-metal configurations run from Virginia, Missouri or Oregon. Every one of them publishes its port speed on the card; where the operator states no transfer policy alongside it, the card says so and we confirm before you order.
Dedicated (46) - Virtual private servers
Virginia and Oregon
Every VPS tier is offered in both virtual regions: US East and US West. Traffic is unmetered. You choose the region when you order.
VPS (4)
- Missouri
- Dedicated servers
- Virginia
- Dedicated servers · Virtual private servers
- Oregon
- Dedicated servers · Virtual private servers
What the network gives you
A published port on every machine
Every machine states a port, from 300 Mbps at the entry of the range to 1 Gbps at the top of it, and every one of those figures is the operator’s own. Where their record also states an unmetered allowance, the card says so; where it states a width and nothing about volume, the card says that instead and we confirm the transfer policy in writing before you order. Every figure is the width of the link and not a rate you are guaranteed — the section below shows what that means.
Unmetered VPS traffic in standard US regions
Virtual tiers carry unmetered traffic in standard United States regions, at — to — depending on the size you pick.
Always-on DDoS mitigation upstream
Volumetric floods are filtered at the operator’s network edge, before they reach the link your machine sits on — always on, not enabled after the fact. Its scope varies by region, it is best-effort rather than a service level, and it does nothing about an application-layer attack that looks like real traffic. All of which is drawn out below, because a shield badge is not an explanation.
IPv6 where it is published
Most dedicated servers are allocated an IPv6 /64 alongside their IPv4 block, and the card says so where that is the case. The ranges that publish no allocation get the same treatment as the port: we leave the row off rather than assume one, and confirm it with you before you order. Virtual servers carry IPv6 connectivity as standard.
Every machine, at the speed its operator publishes
The port each machine actually sits on
All 46 bare-metal machines publish a port speed, and all 46 figures below are the operator’s own — from 300 Mbps at the entry of the range to 1 Gbps at the top of it. Nothing here is our estimate.
- Anvil 11 Gbps
- Anvil 21 Gbps
- Anvil 31 Gbps
- Anvil 41 Gbps
- Anvil 51 Gbps
- Anvil 61 Gbps
- Anvil 71 Gbps
- Anvil 101 Gbps
- Anvil 111 Gbps
- Anvil 121 Gbps
- Anvil 13500 Mbps
- Anvil 14300 Mbps
- Anvil 151 Gbps
- Anvil 16500 Mbps
- Anvil 171 Gbps
- Anvil 181 Gbps
- Anvil 81 Gbps
- Anvil 9500 Mbps
- Forge 11 Gbps
- Forge 21 Gbps
- Forge 31 Gbps
- Forge 41 Gbps
- Forge 5500 Mbps
- Forge 61 Gbps
- Forge 71 Gbps
- Forge 81 Gbps
- Forge 91 Gbps
- Apex 11 Gbps
- Forge 101 Gbps
- Forge 111 Gbps
- Forge 12500 Mbps
- Forge 131 Gbps
- Forge 141 Gbps
- Apex 21 Gbps
- Apex 31 Gbps
- Apex 41 Gbps
- Apex 51 Gbps
- Apex 61 Gbps
- Apex 71 Gbps
- Apex 101 Gbps
- Apex 111 Gbps
- Apex 121 Gbps
- Apex 81 Gbps
- Apex 91 Gbps
- Apex 131 Gbps
- Apex 141 Gbps
Where the addressing is published — 32 machines — it is 5 usable IPv4 addresses and an IPv6 /64 block. That is the operator’s allocation, not a policy of ours, which is why it is quoted rather than promised.
Every figure here is the width of the link and not a rate you are guaranteed: reaching it end to end also needs the far side to send that fast and the path between to have the capacity free. Where a machine’s record states an unmetered allowance the spec sheet says so; where it states a width and nothing about volume, the spec sheet says that instead, and we confirm the transfer policy in writing before you order rather than guessing at it.
What a port speed actually buys
A gigabit is a ceiling, not a promise
“Up to” means up to
Every port figure on this site — here, on the product cards, and in the comparison table — is the width of the link. It is the most the port can carry, not an amount reserved for you and not a rate you are guaranteed to see. Reaching it end to end also needs the far side to send that fast, the path between to have that capacity free at the time, and the workload not to be waiting on a disk or a core first. Read every number below as the arithmetic best case for the link alone, because that is exactly what it is.
Pick a speed. The bar sweeps in true proportion to the transfer it depicts — a 100 Mbps link takes fifty times as long as a 5 Gbps one because it does — and the figures beside it are the same arithmetic in units you plan a migration in.
Transferring 1 GB
The animation is compressed by the same factor at every speed, so the four bars stay in true proportion to one another. The times printed are real.
All four at once
- 100 Mbps1 min 20 s
- 500 Mbps16 s
- 1 Gbps8 s
- 5 Gbps2 s
The same four links started together on the same gigabyte. Nothing is selected here and nothing needs to be — the gap is the whole argument, and it is the gap you are buying when you move up a step.
Time to move
- Container image1 GB
- 1 min 20 s
- Database dump50 GB
- 1 h 7 min
- Full backup500 GB
- 11 h 7 min
4
4K streams at once
At 25 Mbps each, which is what the large services push for UHD.
12,207
Cached responses a second
A 1 KiB body, and the link taken as the only limit. The machine’s cores are the other ceiling and the product pages quote the smaller of the two.
Time to move
- Container image1 GB
- 16 s
- Database dump50 GB
- 13 min 20 s
- Full backup500 GB
- 2 h 13 min
20
4K streams at once
At 25 Mbps each, which is what the large services push for UHD.
61,035
Cached responses a second
A 1 KiB body, and the link taken as the only limit. The machine’s cores are the other ceiling and the product pages quote the smaller of the two.
Time to move
- Container image1 GB
- 8 s
- Database dump50 GB
- 6 min 40 s
- Full backup500 GB
- 1 h 7 min
40
4K streams at once
At 25 Mbps each, which is what the large services push for UHD.
122,070
Cached responses a second
A 1 KiB body, and the link taken as the only limit. The machine’s cores are the other ceiling and the product pages quote the smaller of the two.
Time to move
- Container image1 GB
- 2 s
- Database dump50 GB
- 1 min 20 s
- Full backup500 GB
- 13 min 20 s
200
4K streams at once
At 25 Mbps each, which is what the large services push for UHD.
610,351
Cached responses a second
A 1 KiB body, and the link taken as the only limit. The machine’s cores are the other ceiling and the product pages quote the smaller of the two.
When something points a botnet at you
Where a flood stops
Volumetric mitigation runs at the operator’s network edge, upstream of the port your machine sits on. It is always on rather than something you enable once an attack has already started, and it is theirs rather than ours — we resell the capacity, we do not run the filter. Step through what a flood actually does.
Nothing is being filtered, and that is correct
Traffic reaches the machine over the port on its card. Detection is watching the flow the whole time, but no filtering is applied to ordinary traffic — mitigation that inspected every packet all the time would be a latency cost you pay every day for an event that happens rarely.
For a few seconds, the junk is on the wire with you
A flood starts and reaches the edge before anything has classified it. Detection is signature-based and fast, but it is not instantaneous, and in that window legitimate requests are queueing behind the attack: slow responses and timeouts rather than a clean outage. Every always-on mitigation in the industry has this window. A page that draws the shield already up is describing a product nobody sells.
The flood is dropped upstream, not at your port
Once the attack is classified, traffic for your address is routed through the operator’s scrubbing infrastructure and the illegitimate part is dropped there — before it reaches the link your machine is on, which is the only place it can usefully be dropped. Your port never sees it, so a 1 Gbps machine is not knocked over by an attack larger than 1 Gbps. Legitimate traffic continues through the same path.
What it does not do
- It does not stop an application-layer attack that looks like real traffic. A thousand well-formed requests a second from a botnet are a thousand well-formed requests; rate limiting and caching in your own stack are what answer that.
- It is not a service level. Mitigation is best-effort at the operator’s discretion, and neither they nor we promise a specific attack size, a time to mitigate or an outcome.
- Its scope varies by region, and by the machine. Ask us about the exact plan you are looking at and we will tell you what the operator states for it rather than what sounds best.
- It is not a backup, and it does not protect you from yourself. The most common cause of a machine being unreachable is still a configuration change, not an attack.
We do not operate this filtering and we do not publish a capacity figure for it. The operator does, the number is theirs, and a figure copied onto this page would be a figure nobody here re-checks when they revise it — which is how a specification becomes wrong quietly. If you need the current published scope for a specific region before you order, ask and we will send you what they state.
On latency numbers
We do not display a live latency widget, because a number generated in your browser against our marketing site tells you nothing useful about your server. If you need a real figure, ask support and we will run a measurement from the actual location — or order a machine, test it, and cancel within the terms if it does not suit you.
Ready when you are
Pick a machine. Get root. Go.
No sales call. No minimum term beyond the month you paid for. Full specifications before you spend anything.
Virtual from $6.99/moBare metal from $35.99/mo