Chivalry II launched in June 2021 to a player count that no one had predicted. Torn Banner Studios and publisher Tripwire Interactive had modelled their best-case scenarios — and reality beat all of them, twice over. Within hours of launch, 250,000 players were fighting simultaneously across dedicated servers in 30 locations worldwide. In the first month, nearly one million players joined the battle. To date, 420 million knights have been slain.
None of that would have been possible without infrastructure that could scale in real time. The partnership with Gameye, which began 15 months before launch, was built around one question: what happens if success exceeds your best-case forecast?
The challenge: scaling for an unknown ceiling
Dedicated multiplayer games live or die at launch. Players who can't get into a match in the first hour often don't come back. For a melee combat game like Chivalry II — where the experience depends on low-latency, stable server sessions — this pressure was even higher.
The team's previous infrastructure model made it difficult to scale on demand. Capacity had to be provisioned in advance, which meant paying for idle servers when player counts were low, and risking under-provisioning when they spiked. That model also limited global reach: adding a new region required lead time, coordination, and upfront cost.
What Torn Banner and Tripwire needed was a platform that could expand and contract with actual player behavior — and do it anywhere in the world, without requiring weeks of planning to add a new region.
15 months of preparation
Gameye and the Chivalry II team worked together through a 15-month closed alpha programme. Regular stress tests against the live infrastructure revealed edge cases and performance issues that would have been invisible in a staging environment.
One of those tests caught a critical issue. Reports from the Chivalry II community on Discord during a closed alpha identified a performance problem that would have severely degraded the player experience at scale. The team identified the root cause, shipped a fix, and ran further tests to confirm the issue was resolved — all before a single player bought the game.
That kind of iterative validation — test early, fix fast, test again — was only possible because the infrastructure was already in a production-equivalent state well before launch day.
Launch day: twice the projection
Despite thorough preparation, launch exceeded every forecast. Player demand in the first hour was roughly double the best-case projection. Gameye's orchestration platform responded automatically: sessions were allocated on demand, capacity expanded across providers as concurrency climbed, and no manual intervention was required.
Infrastructure stayed online with no major downtime events throughout launch day. The player experience — the thing that determines whether a multiplayer game retains its audience — was stable from the first match to the last.
Going global on short notice
One of the surprises was geographic. Chivalry II generated significant interest in Japan and Korea, markets that hadn't been prioritised in early capacity planning. Because Gameye is provider-agnostic and can add regions without advance reservation, capacity was provisioned in those locations within hours.
Players in Tokyo and Seoul got low-latency matches. A rigid, single-provider infrastructure model would have either locked those players out entirely or delivered a poor experience — neither of which was acceptable for a game trying to build a global community.
The ability to respond to unexpected demand — geographically or in terms of raw concurrency — is what separates platform-agnostic orchestration from traditional fixed-capacity hosting.
Cost without surprises
Traditional approaches to launch-day readiness mean over-provisioning: buying capacity you might not use, because the cost of under-provisioning is worse. That model works, but it's expensive — and the cost is front-loaded, before you know whether the launch will succeed.
Gameye's per-second billing and automatic scaling meant Torn Banner and Tripwire paid for sessions that ran, not capacity that sat idle. When demand exceeded projections, the platform expanded. When sessions ended, that capacity was released. The result was a launch that handled twice the expected load at a substantially lower infrastructure cost than a traditional approach would have required.
What made it work
Several factors combined to produce a clean launch:
- Long integration runway. Fifteen months of testing meant the integration was mature, not rushed. Issues were found and fixed in alpha, not in front of a live audience.
- Community-driven stress testing. Real players using real clients found a performance issue that internal testing missed. Building the community into the QA process had a direct impact on launch quality.
- Provider-agnostic capacity. The ability to add regions — including Japan and Korea — on short notice without pre-negotiated contracts or reserved instances was essential to meeting demand where it appeared.
- Automatic scaling. No manual intervention was required on launch day. The platform responded to demand in real time, keeping pace with player concurrency without an engineering team on standby to provision servers.
"When we spoke to Gameye, I'll be honest, we weren't sure it could be done."
It was done. 250,000 players on day one. Nearly a million in the first month. Zero downtime. 420 million knights slain — and counting.