ARK: Survival Ascended with 10-20 mods - why your CPU graph sits at 150-400%
You opened the panel, saw the CPU sustained at 186%, and assumed something was broken. On a heavily modded ARK: Survival Ascended server, that number is the engine using the cores it has. ARK: SA is heavily multi-threaded; with a real mod stack (CurseForge structures, dinos, QoL) and a busy map, sustained 150-400% is the steady state, not a fault.
This page is the operator reference for what makes ARK: SA expensive at runtime, which design choices in your mod list pile on the most cost, and where the realistic mitigations are.
Why ARK: SA is expensive before you add a single mod
Three structural reasons:
- Wine/Proton. ARK: Survival Ascended ships a Windows binary; Wildcard has not released a native Linux dedicated server. The Supercraft node runs
ArkAscendedServer.exeunder a Wine compatibility layer. This adds noticeable baseline CPU compared to a hypothetical native build - every syscall takes an extra hop through Wine's translation. It is not catastrophic, but it does not disappear. - UE5 server tick. UE5's server-authoritative simulation runs more work per tick than UE4 did. World partition streaming, the lighting baseline, the larger entity radius all add per-tick cost even on a vanilla map.
- Map ambient load. Map complexity matters: TheIsland is the cheapest baseline; Ragnarok is moderate; Valguero_WP and Astraeos are heavier - both pack more biomes, more wild creatures, more terrain features. The default per-tick cost can vary by 30-50% just by map choice.
Stack those three together and a clean ARK: SA server is already in the 120-180% range with a single player connected. Add mods to that and the steady-state moves up.
Mod cost tiers, ranked
Not all mods are equal in CPU cost. Rough ordering from heavy to light:
| Tier | Mod type | Why it costs | Examples |
|---|---|---|---|
| Heavy | Structure overhauls (S+ equivalents, advanced building) | Adds many tickable structure types; AOE structure-decay calculations widen | Structures Plus, Super Structures, Awesome Teleporters |
| Heavy | AI / dino overhauls (Primal Fear, Eternal) | New AI behaviour trees; more wild-dino ticks; expanded breeding state machines | Primal Fear, Eternal Evolved, Pugnacia |
| Medium | Map-extending mods (large new biomes, dungeons) | Larger streaming footprint; more landscape actors | Castles Keeps and Forts, biome expansion mods |
| Medium | Economy / quest mods | Persistent state machines per player; periodic tick scans | ARK Additions: Quest, market mods |
| Light | QoL improvements (stack mods, dye mods, hairstyle mods) | Mostly inventory-time work, not per-tick | Awesome Spyglass+, Dino Storage v2 |
| Light | Cosmetic / skin / UI mods | Client-side mostly; very low server impact | Hair mods, skin packs, UI replacements |
Seventeen mods including two heavy structure overhauls is much more expensive than seventeen mods of the cosmetic-and-QoL variety. If you are at 200%+ sustained and want to come down without losing gameplay content, dropping any one tier-Heavy mod usually pays back more than dropping several tier-Light ones.
WinLiveMaxPlayers has a per-tick cost
The WinLiveMaxPlayers setting in GameUserSettings.ini raises the slot cap. Each slot above the default carries a small per-tick cost in the connection processing and inventory replication paths. The cost is real but modest - going from 32 to 70 is not 2× the CPU, more like 15-25% added. Still, if you are running close to your plan ceiling and have no players using the extra slots, lowering this knob is one of the cheapest wins.
The map difficulty axis
| Map | Baseline cost (CPU% with no mods, no players) |
|---|---|
| TheIsland | ~80-100% |
| The Center | ~90-110% |
| Ragnarok | ~110-130% |
| Scorched Earth | ~100-120% |
| Aberration | ~120-140% |
| Valguero_WP | ~130-150% |
| Astraeos | ~130-160% |
These are rough averages across a population of running servers; your numbers will move with the specific build's published patch level. The shape is what matters: TheIsland is the cheapest, Astraeos and Valguero_WP are at the top of the spread.
Diagnostic commands
How much memory is the server actually using?
pmap $(pidof ArkAscendedServer.exe) | tail -1The last line shows total mapped memory. If this number grows steadily over a session (e.g. 12 GB → 18 GB → 22 GB over a few hours of play), one of your mods is leaking and you should restart on a schedule until you identify which one. A stable line that sits at 18 GB and stays there is normal for a heavily modded server; growth without a corresponding gain in active world state is the red flag.
Which thread is the busy one?
top -H -p $(pidof ArkAscendedServer.exe)The thread with the highest CPU is the main simulation thread. If it is pinned at 100% (of a single core), no amount of plan upgrade will help - the bottleneck is single-threaded on that one thread. Mitigation is the same as for any single-thread-bound game: drop entity counts, drop AI mods, drop active dino population.
Are we actually being throttled?
cat /sys/fs/cgroup/system.slice/u<port>.slice/cpu.statnr_throttled close to nr_periods means the kernel is pinching the process at every measurement window. That is when a plan upgrade pays back. nr_throttled = 0 means your plan is not the ceiling regardless of how high the percentage reads.
Mitigation playbook
If you are uncomfortable with the CPU graph and want to bring it down without dropping play features your community likes:
- Identify and drop the heaviest single mod first. A structure overhaul or an AI overhaul mod usually carries more cost than five QoL mods combined.
- Trim WinLiveMaxPlayers to match your active community size. Sixteen slots used out of seventy is paying for sixty empty slot ticks.
- Switch to a lighter map. Astraeos to TheIsland is the largest single drop possible without changing mod content.
- Lower the structure-decay scan radius and frequency. The S+ family of mods exposes config knobs that change how often decay runs and over how big an area.
- Cap dino-spawn levels and counts. Lowering
DifficultyOffsetreduces wild-dino tick cost without changing player progression. - Restart on a schedule. A 12-hour cycle clears memory drift from leaky mods.
- Upgrade the plan. Only if the throttling check above confirms you are at the cgroup ceiling. If the kernel is not throttling, the bottleneck is somewhere other than CPU shares; an upgrade does not help.
When to actually worry
The honest red flags on a modded ARK: SA server, in order:
- Sustained CPU above 80% of your plan's total core ceiling for 30 minutes, with throttling above 5% of recent windows. (See the cross-game CPU explainer for the math.)
- RAM creeping upward over a play session without a corresponding gain in world content.
- Server tick FPS readable from the console below 15.
- RCON
ListPlayerstaking more than 1-2 seconds to respond.
Two or more of these together is the time to act. One on its own is usually a momentary blip - wait through one autosave cycle and re-check.
What does not help
- Disabling mods you do not actually use. If a mod loads but never ticks (purely cosmetic, no per-tick code), removing it changes almost nothing.
- Reducing slot count if no one was using the extra slots beyond a small amount - see WinLiveMaxPlayers above.
- Switching to a "lighter" mod that does similar things. Heavy structure mods are heavy because of the gameplay they add; alternative mods doing the same gameplay carry similar cost.
- Adding more cores when the bottleneck is single-thread. Verify with
top -Hbefore paying for the upgrade.
Related guides
- ASA self-host hardware sizing
- ASA FPS optimization basics
- ARK UE5 performance upgrade notes
- Workshop mod broke after update
- Cross-game: what CPU 200%+ actually means
Need an ARK: SA host whose Doctor tab tells you when CPU is really a problem vs. when it is just a modded server doing what it should? See ARK: SA plans. Every tier ships with daily backups, automatic mod sync, and panel-managed multi-map clusters.