ARK Server Tuning in 2026: ASE vs ASA Config Flags, RAM Allocation, and Save Performance

ARK: Survival Evolved (ASE) and ARK: Survival Ascended (ASA) look similar in a server list and behave nothing alike on the host. ASA runs on Unreal Engine 5 with Nanite and Lumen, so the same 50 players that fit in 16 GB of RAM on ASE need 32 GB or more on ASA, and the world save file grows from tens of megabytes to hundreds. This guide covers the launch flags, the two config files, RAM sizing per player count, and the save-tuning that decides whether your autosave is a 200 ms blip or a 3-second server freeze.

ASE vs ASA: where the resources actually go

Players	ASE RAM	ASA RAM	ASA (cluster of 3 maps)	Recommended CPU
10	6 GB	12 GB	32 GB	Ryzen 5 5600 / Xeon E-2336
25	10 GB	20 GB	48 GB	Ryzen 7 5800X
50	16 GB	32 GB	64 GB	Ryzen 9 7950X / EPYC 9354P
70+	24 GB	48 GB	96 GB	Ryzen 9 7950X3D / EPYC 9354

Both engines are single-thread-bound on world tick, so clock speed beats core count: a 5.7 GHz Ryzen 9 7950X holds 45–55 tick on a 50-player ASA Island server where a 2.4 GHz 16-core Xeon struggles past 30 tick. Save that headroom by keeping one map per instance rather than stacking three on one machine.

Launch flags that matter

# ARK: Survival Evolved (ASE), Linux
./ShooterGameServer TheIsland?listen?SessionName=MyARK?Port=7777?QueryPort=27015?RCONPort=27020?MaxPlayers=50 \
  -server -log -NoBattlEye -USEALLAVAILABLECORES \
  -clusterid=cluster1 -ClusterDirOverride=/srv/ark/cluster -servergamelog

# ARK: Survival Ascended (ASA), Linux
./ArkAscendedServer TheIsland_WP?listen?SessionName=MyARK?Port=7777?QueryPort=27015?RCONPort=27020?MaxPlayers=50 \
  -server -log -NoBattlEye -NoHangDetection -ServerPlatform=All \
  -clusterid=cluster1 -ClusterDirOverride=/srv/ark/cluster -servergamelog

Flag	ASE	ASA	Effect
-USEALLAVAILABLECORES	Yes	Ignored	Lets ASE worker threads use every core; removes the historical 2-core cap.
-NoHangDetection	Optional	Recommended	Prevents a slow save or map transfer from being misread as a hung process and force-killed.
-preventhibernation	Add-on	Add-on	Keeps dino AI ticking out of range; costs 15–25% more CPU, fixes stasis bugs.
-NoBattlEye	Optional	Optional	Removes anti-cheat overhead (~5% CPU) for private clusters.
-servergamelog	Yes	Yes	Writes kill/tame/join logs used by tribe tools.
-clusterid / -ClusterDirOverride	Yes	Yes	Required for cross-map transfers; point ClusterDirOverride at an NVMe path.
-ForceRespawnDinos	Rare	Rare	One-time flag to wipe wild dinos after a config change; remove it afterwards.

The two config files: what to change

GameUserSettings.ini holds server settings and Game.ini holds gameplay rates. The values below are the ones with real performance impact:

• AutoSavePeriodMinutes=15 in [/Script/ShooterGame.ShooterGameMode] (GameUserSettings) — going below 10 multiplies save I/O; going above 30 risks losing a long rollback window.
• MaxPlayers=50 must match the launch flag, or the query port advertises a cap the world tick cannot sustain.
• RCONEnabled=True plus RCONPort=27020 — the only practical way to broadcast save warnings to players.
• PreventDinoHibernation — the config-side equivalent of -preventhibernation; pick one method, not both.
• TamingSpeedMultiplier and DinoCountMultiplier in Game.ini — wild dino count above 1.3 raises tick cost linearly with no player-visible benefit.

Back up both files before editing: a single malformed line makes the server regenerate defaults and silently drop your rates. Keep them in Git or a timestamped copy under /srv/ark/config-backups/.

Save performance: the part most admins get wrong

ASA world saves are large. A 50-player Island save sits around 90–140 MB, Extinction and Genesis push 180–220 MB, and a cluster of three maps can write 400 MB per autosave cycle. Where that data lands determines how long the world stalls:

Storage	ASA save write	Observed tick stall	Note
SATA SSD (550 MB/s)	100 MB	0.8–2.5 s	Players see a visible freeze every 15 min
NVMe Gen3 (3,000 MB/s)	100 MB	250–500 ms	Acceptable; no packet loss
NVMe Gen4 (6,500 MB/s)	100 MB	80–200 ms	Autosave is effectively invisible
Network/NFS mount	100 MB	3–12 s	Never host ARK saves on remote storage

Put the save directory and ClusterDirOverride on the same NVMe device, and make sure it is not shared with the OS log or Steam update spool. If your host offers only SATA SSD for game servers, prefer smaller maps and stagger autosaves between instances.

Cluster transfers and cross-map performance

Clusters are where tuning breaks down most often, because a transfer touches two maps, two processes, and the shared cluster directory at the same time. Treat the cluster directory as hot storage: local NVMe only, owned by the game user, with at least 20 GB free. A few rules keep transfers reliable:

• Stagger autosaves across maps (15, 25, and 35 minutes) so two instances never write simultaneously to the same device.
• Keep every map on the same machine when possible; cross-host clusters over the network add 20–80 ms per transfer and can corrupt character data if the upload times out mid-write.
• Enable hibernation prevention on transfer-heavy clusters so a target map is already resident when a player uploads a creature or a character.
• Cap upload rates with your provider’s unmetered 1 Gbps port; 50 players uploading tames at once can push 60–100 Mbps sustained, which is nothing on a real port but saturates a 100 Mbps plan.

Watch RAM during transfers specifically: a 50-player cluster typically peaks 20–30% above idle RSS for 30–60 seconds per upload. If you are already at 80% of your container limit, that spike is what triggers the OOM killer, not steady-state play.

Stability habits that survive wipes

• Restart the process weekly during a scheduled window; ARK degrades after 5–7 days of uptime as memory fragments.
• Hibernate-free clusters: if you run multiple maps, cross-server transfers fail when hibernation reclaims the target map, so keep -preventhibernation on transfer-heavy clusters.
• Monitor RSS, not virtual memory: track the process with Prometheus or a simple ps -o rss cron, and alert at 85% of the container limit.
• Verify the save is readable monthly by loading a copy on a spare instance before you need it.

None of this tuning helps if the underlying machine is oversold. Before you rent, review dedicated server specs for game hosting and insist on a 5.0 GHz+ current-generation CPU, NVMe Gen4 storage, and an unmetered 1 Gbps port at minimum. Our dedicated server comparison covers which providers actually deliver that hardware for ARK-class workloads, and it is worth running a trial save-and-reload test during the first week so you can measure autosave stalls yourself.