How many players will your modpack hold?
Pick your processor, memory and modpack. You get how many players it held in real load tests, what each player costs the tick, what gave out first and which mods ate the time. Every number shows how many runs it came from.
Your processor and memory
None of it is required. The more you say, the nearer the runs we can show you, and each one says how near it actually is.
Your modpack
Both CurseForge and Modrinth at once. Once you pick one we read its mod list and hold it against what people have measured.
What people have measured 0 runs in all
There is nothing here yet
Nobody has sent us a run yet. Running a server? One command and you would be the first. How
How we measure
Test players are spread across the map, so each one loads its own chunks the way real players do. The tick is profiled per mod. Runs from different measurement versions and scenarios are never averaged together, and a number we don't have shows as "not measured", never zero.
A median from four runs is a rumour. Every number here carries the count it came from, and where there are not enough we say so.
Questions
How much RAM does a modded server need?
Pick your pack and machine above. You get what gave out first in real runs, whether that was memory or something else.
Where do the numbers come from?
From servers running STYGION Keystone whose owners chose to share a load test. Each run says which machine, which pack and which JVM flags.
Why does it say "not measured"?
Because a zero would read as the best possible result. We'd rather show nothing than a number nobody measured.
Measure your own server
Run a server? /keystone test 20 adds your numbers. No server name, no address, nothing that names anybody.
/keystone test 20 Nobody types a pack name: a set is known by the mods it actually loaded. Keystone also
measures five minutes of ordinary play once a day: how long the tick took, how many people
were on, and which mod the time was spent in. Nothing that names the server goes with it,
not even a random id, so we cannot tell two samples from one machine apart from two
machines. /keystone set bench.share-live false turns it off, and emptying bench.send-to stops anything being sent at all.