A server running at 64 ticks per second has 15.6 milliseconds to simulate the world, resolve every input and send every update. Miss that budget and the tick runs long, inputs queue, and players feel it. For games whose simulation runs on one thread — Minecraft is the best-documented case — that budget is not enlarged by adding cores, RAM or bandwidth. Yet those are the three things most game hosting plans sell and price.
The tick budget is the capacity number that binds first
A tick is the simulation step. The tick rate sets a hard wall-clock budget per step: the budget is the inverse of the rate.
| Tick rate | Budget per tick | Budget lost versus the row above |
|---|---|---|
| 20 Hz | 50.0 ms | — |
| 30 Hz | 33.3 ms | −16.7 ms |
| 64 Hz | 15.6 ms | −17.7 ms |
| 128 Hz | 7.8 ms | −7.8 ms |
Minecraft’s creator documentation gives “the gameplay tick which is 20 ticks per second” for Bedrock Edition (Microsoft Learn). The other rows are common server configurations, shown for the arithmetic rather than attributed to any title.
Doubling the tick rate does two things at once: it halves the compute budget per step and, if the server sends one update per player per tick, it doubles the packet rate per player. One cost lands on a CPU core, the other on the network path.
PaperMC, which maintains a widely used Minecraft server implementation, states the constraint directly: “A major source of load in the server comes from the tick loop, which uses a single thread,” so “at a certain point, more cores will not give you a performance benefit, so it’s advisable to go for a CPU with high single-threaded performance.” Many other engines also run their main simulation on one thread, but the degree varies by title — check how yours behaves before applying this to it.
For a single-threaded tick loop, the variable that decides a hosting comparison is not core count, RAM or port speed. It is how many milliseconds of the tick budget your busiest tick consumes on one core.
Three markets, three units, no conversion to the tick
Game hosting is sold in at least three units, and none of them converts to the tick budget.
| Tier | Unit sold | Published example | What the unit does not buy |
|---|---|---|---|
| Slot panels | Player slots | Plans named by player count, from a handful of players to several dozen | Any statement about CPU share or tick headroom |
| RAM panels | Gigabytes per 30 days | GPORTAL Minecraft: 2 GB $5.99, 4 GB $13.09, 6 GB $16.64 list (25% off the first term), “∞ Slots” | A core the tick loop does not share |
| Dedicated | A whole machine | OVHcloud GAME-1 2026: $331/month, 8 cores, 64 GB | A guarantee the game is single-thread friendly |
GPORTAL and OVHcloud pages read 17 November 2026; prices excl. tax.
GPORTAL is explicit about the shift away from slots: “At GPORTAL, we don’t charge you by the player,” and the cap “is whatever your server RAM can handle.” That is more honest than a slot tier — and it still prices memory, which lets a world load, not a tick finish.
Prices per gigabyte show how the market sells, not what you need to buy. At list price GPORTAL works out at $3.00, $3.27 and $2.77 per gigabyte per 30 days; OVHcloud’s GAME-1 at $5.17 per gigabyte per month — 1.6 to 1.9 times the panel. A general-purpose dedicated server can invert that: Hetzner’s AX42 is about €1.55 per gigabyte. None of these numbers tells you whether your tick will finish in time; the CPU does.
What a single-threaded tick loop does to the shortlist
OVHcloud lists two 2026 game servers side by side. GAME-1 is an AMD Ryzen 7 9800X3D (4.7 GHz base, 5.2 GHz boost), 8 cores, at $331 a month. GAME-2 is a Ryzen 9 9950X3D (4.3 / 5.7 GHz), 16 cores, at $371. Both carry 64 GB of DDR5.
The upgrade is +12.1% in price for +100% cores, +9.6% boost clock and −8.5% base clock. For a tick loop under sustained load, the cheaper machine has the higher floor and the more expensive one the higher ceiling.
Cache matters as much as clock. AMD lists 96 MB of L3 on the 8-core 9800X3D and 128 MB on the 16-core 9950X3D. On the 9950X3D the stacked 3D V-Cache sits on one of its two eight-core chiplets, so roughly 96 MB of L3 serves one group of cores and 32 MB the other. A tick loop scheduled onto the wrong half of a 16-core game server runs on a materially different machine from the one in the brochure.
Some specifications are simply missing. Hetzner’s AX42 page names its CPU — an “AMD Ryzen 7 PRO 8700GE” — and publishes RAM, storage, uplink and terms, but no clock speed. The GE suffix marks AMD’s low-power variant of that chip, so check its clocks on AMD’s own page before assuming desktop-class single-thread speed. For a purchase decided by single-thread performance, the clock is the specification.
Bandwidth is not the game-hosting network problem
OVHcloud’s Game range ships with “1Gbps unmetered public bandwidth.” For game traffic that is enormous headroom. Assuming one update packet per player per tick, a 64-player server at 64 Hz emits about 4,096 packets per second, and a 200-player server at 30 Hz about 6,000. Real netcode batches and delta-encodes, so treat those as an upper bound. Against a 1 Gbit/s port’s theoretical ceiling of about 1.49 million minimum-size packets per second, that is under half of one per cent.
The port is not sized for your players. It is sized for attacks. OVHcloud reported more than 50 packet-rate attacks above one billion packets per second in 2024, peaking at 1.9 billion (OVHcloud blog, April 2025) — about 1,277 times the packet ceiling of a 1 Gbit/s port. That is why “unmetered 1 Gbps” says nothing about survivability, and why mitigation has to happen upstream of the server. Unmetered, metered and 95th-percentile billing describe three different products; none of them describes packet-rate capacity.
Why game traffic needs its own mitigation profile
Game servers exchange small, frequent packets in title-specific protocols rather than web requests, so generic web filtering is a poor fit. OVHcloud’s Game range advertises protection for “35+ games and protected network protocols”.
RFC 8085, the IETF’s UDP usage guidelines, names the exposure that query and status endpoints create: “Applications that respond to short requests with potentially large responses are a potential vector for amplification attacks, and SHOULD take steps to minimize their potential for being abused as part of a DoS attack.” A public server-list query endpoint is exactly that shape.
The question product pages rarely answer: what happens to a title that is not on the profile list. Ask for the list by title, and for the behaviour on an unlisted protocol, before you ask about scrubbing capacity. See DDoS protection and managed mitigation.
What breaks in production
The panel pins you to one core and you never notice. PaperMC warns that in some panels “a number like 4 doesn’t mean it’ll use 4 cores, but instead that it will use the one core with ID 4.” A server that looks like it uses 6% of a 16-core box may be saturating the only core it was given.
The tick loop lands on the cache-poor chiplet. On a dual-chiplet X3D part the operating system decides unless someone sets CPU affinity. On a dedicated server you can set it yourself; on a panel, ask whether the provider does.
A single long tick cascades. PaperMC’s watchdog fires when the server is “taking very long (10+ seconds) to finish the current tick,” and by then every queued input, chunk save and plugin task is behind it.
Mods and plugins eat the budget. The tick budget is fixed; the work inside it is whatever was installed last.
Mitigation can reroute, and latency moves with it. Diverting traffic through a scrubbing path can change round-trip time — see how anycast routing picks a site.
The query port is an attack surface too. A public status endpoint that answers short unauthenticated requests with larger responses is the amplification shape RFC 8085 warns about.
Measure this before you buy
Run your actual world, mod set and plugin list, not a vanilla instance, and record milliseconds per tick at the 95th and 99th percentile at peak player count.
Divide the p99 figure by the tick budget for your target rate, and track that ratio as the mod list grows.
Establish whether the work is single-threaded, or whether chunk generation, pathfinding and I/O are already offloaded. That decides clock versus core count.
Measure downstream packets per second per player at your target tick rate and multiply by your planned player ceiling.
Ask each provider for base and boost clock in writing, by CPU model, for the machine you will actually receive.
Ask for the DDoS profile list by game title, and the behaviour for a title not on it.
Measure round-trip time from your players’ locations to each candidate data centre before comparing prices.
Which option, when
A slot or RAM panel fits a small private world where the tick budget is never close to full and nobody wants to run a machine. A noisy neighbour on the same host is tolerable when nobody is competing.
A dedicated server fits once p99 tick time uses enough of the budget that a noisy neighbour would push it over, the title is competitive, or player counts make contention certain. You are not buying 64 GB of RAM; you are buying a core the tick loop does not share.
A general-purpose dedicated server beats a game-branded one when you need the clock rather than the game panel — provided the vendor publishes the clock. Compare entry terms too: OVHcloud lists a one-month installation fee on its Game range, waived on commitments of a year or more; Hetzner’s AX42 adds a €49 setup fee on €99 a month, with no minimum term. One-time fees and notice periods are part of the price, as are the line items a dedicated quote leaves out.
Game hosting on INXY.hosting
INXY.hosting offers game hosting on dedicated servers with no setup fee and no minimum term, in the EU and the US — so a title can be benchmarked on real hardware for a month and moved when the mod set outgrows the core. Because we source across providers on a single contract, two vendors’ machines can be tested side by side before anything is committed.
Send us a p99 tick-time trace at peak concurrency and your players’ geography, and we will price the machines that fit it.
FAQ
What should I look for in a dedicated server for game hosting?
For a game with a single-threaded tick loop, single-thread CPU performance first — base and boost clock for the exact CPU model. Then DDoS mitigation with your title on the profile list, then network location relative to your players. Core count, RAM and port speed matter less than those three.
How many cores does a game server need?
Often fewer than you are offered. PaperMC’s documentation says the Minecraft tick loop uses a single thread and that beyond a point more cores give no benefit. Extra cores help chunk loading, backups and additional server instances on the same machine, not the tick itself.
Is unmetered 1 Gbps bandwidth enough for a game server?
For game traffic, comfortably. At one update per player per tick, a 64-player server at 64 ticks per second sends about 4,096 packets per second — well under one per cent of a 1 Gbit/s port’s packet ceiling. The port’s limit matters during attacks, which is why mitigation capacity is the figure to compare.
Why do game servers need special DDoS protection?
Because game traffic is small, frequent packets in title-specific protocols rather than web requests, so generic web filtering fits poorly. Providers with game-specific protection list the titles they cover; ask whether yours is on the list and what happens to a protocol that is not.
Does tick rate affect hosting cost?
Directly. Doubling the tick rate halves the compute budget per step and, at one update per player per tick, doubles the packet rate per player. A 128 Hz configuration needs a faster core and more mitigation headroom than a 64 Hz one serving the same players.
