Gear & upgrades · Published 2026-09-07 TW live

Gem tag line: damage or cooldown? The maths says 3 to 1

✍️ Alice・愛麗沙

Quick answer

Take damage. Unless you are a support class living off a buff cycle, or you intend to use a gem to replace a cooldown rune, this slot is always the damage line. Three fully engraved lines on an Elite Star Prism give +6.0% damage or −1.95% cooldown. That 6% lands in a multiplicative bucket of its own and is collected on every single hit; that 1.95% has to first add up to one whole extra cast before it converts into anything, and across a three-minute fight it usually does not get there. Converted, the exchange runs at roughly 3 to 1. The effort that actually matters is not picking a line, though — it is checking first that the gem’s tag covers the skills in your rotation. At zero coverage, 2.0% and 0.65% are both zero.

The two numbers, stage by stage

A tag line is one choice slot: engrave damage or engrave cooldown, never both. How many lines each stage gives and what each is worth is laid out in full by Korea’s gem simulator.

StageMono gem (each)Star Prism (all stats / HP)Tag linesDamage per tag line
Stage 1 (green, Advanced)+8———
Stage 2 S (green, Advanced)+15———
Stage 3 (blue, Rare)+2222/1801? not published
Stage 4 S (blue, Rare)+2929/25021.7%
Stage 5 (no prefix, Elite)+3636/32032.0%
Stage 6 SNo mono gem at this stage? not published32.1%
Stage 7 (final)No mono gem at this stage+10 / +180 over Stage 632.2%

Mostly KR player panel transcriptions; the stage-5 all-stats 36 / HP 320 match the Heliodor values Nexon Taiwan prints. Taiwan took Korea's 7/16 gem-alchemy rework on 9/9: mono gems can be crafted up to stage 4 (crude S), and from elite up only Star Prisms can be promoted — elite mono gems you already hold can be converted back into crude gems plus gold.

Three numbers do the work. Stage 5 (Elite): +2.0% damage or −0.65% cooldown per line, three lines. Stage 6 S: 2.1% / 0.7%. Stage 7 (final): 2.2% / 0.75%. The damage-to-cooldown ratio holds near 3 to 1 the whole way up. That is not a coincidence — it is the exchange rate the designers set.

The question is whether that exchange rate is fair.

Converting cooldown into damage

A cooldown line has exactly one route to paying out: letting you press the button more times in the same fight. So valuing it means assuming a fight.

Three Elite lines all engraved cooldown = −1.95%. Take a main skill on a 30-second cooldown and a fight lasting 180 seconds:

  • Before: 180 ÷ 30 = 6 casts.
  • After: the cooldown is 29.415 seconds, so 180 ÷ 29.415 = 6.12 casts.

6.12. You do not cast 0.12 of a skill. To collect that fraction, the cooldown has to come back at exactly the right moment near the end of the fight, and you have to still have time to press it, and that second cannot be spent repositioning or casting something else.

In practice most fights hand you 6 casts and a gain of zero.

Now the damage line. Three of them = +6.0%, landing in segment E of the damage chain — a separate multiplicative bucket, entirely apart from the crowded “damage increase” bucket where skill power, damage entries and resonance all pile up. That 6% is not diluted and does not need to reach a whole number. Every tagged hit you land collects it.

Discrete against continuous. That is the real reason behind 3 to 1.

Korea’s own conclusion, and the exceptions it allows

The Korean community’s phrasing is blunt: damage beats cooldown in almost every case. The reasoning is specific too — the cooldown ceiling cannot convert into more casts, because this game’s skills have wind-up and recovery animations that eat part of whatever you saved. A second guide site lands on the same conclusion.

They leave two exceptions, and neither one belongs to a damage dealer.

One: a support class’s buff cycle. When your skill is putting a buff on the party and that buff’s uptime is stuck at 80%, shaving the cooldown slightly can push uptime to 100%. A damage line cannot buy that, because what the buff multiplies is the whole party.

Two: you were going to run a cooldown rune anyway. If a gem can take over that job, the rune slot is freed for something else. That is trading a slot, not trading a number.

Coverage first, numbers second

This is the most-skipped section on the page and the most expensive one.

A tag line only pays out on skills carrying that tag. A +2.0% Elemental line is +1.6% for a class whose rotation is 80% Elemental-tagged, and +0.2% for a class with a single Elemental button. Same line, eight-fold difference.

The tag sets each class actually runs:

ClassRecommended tag order
SwordsmanSmash > Support > Movement (Season 2 wedge tree; Season 1 was Smash / Multi-hit / Movement)
VanguardSmash damage > Support damage > Support cooldown — the third line is deliberately not a damage line
WarriorHinder > Smash > Movement (Survival in Abyss, Support on a 2.5-line gem)
KnightSmash > Support > Multi-hit (> Summon > Survival > Hinder)
FighterSmash = Movement > Support > Movement cooldown
MonkSupport / Multi-hit / Movement
RogueHinder / Multi-hit / Smash (in Abyss: Hinder / Multi-hit / Summon)
Dual BladeMulti-hit > Movement > Smash = Support, with Hinder last
DancerMulti-hit > Movement > Support
ArcherHinder > Multi-hit = Smash (Korea S2: Multi-hit > Smash = Hinder > Movement)
ArbalistMulti-hit = Elemental > Smash
LongbowmanHinder / Smash / Multi-hit
MagicianElemental > Hinder > Smash = Summon (Multi-hit not recommended)
Fire MageElemental / Multi-hit / Smash
Ice MageElemental / Summon / Survival
Lightning MageElemental > Smash > Support (or Multi-hit; Hinder last)
Dark MageSummon > Hinder > Multi-hit
HealerSupport > Multi-hit > Summon = Support cooldown > Smash
PriestSupport > Smash > Hinder > Multi-hit
BardHinder > Multi-hit > Support (Korea late S1: Hinder > Support > Smash)
MusicianHinder / Multi-hit / Smash

KR community sheet (2026-02). All eight tag names match the skill tags printed on the official TW class pages (強擊, 連擊, 元素, 召喚, 干擾, 輔助, 移動, 生存).

So the correct question is not “is the damage line better than the cooldown line.” It is “this line × this tag’s share of my rotation = what?” Do that multiplication first, then come back and compare 2.0 against 0.65.

Two rules about engravers

An engraver re-rolls the tags, not the values. The official description says only “random” and publishes no distribution — so nobody can tell you how many engravers it takes to roll into Smash, and any claim that quotes a specific count did not come from an official source.

Decide before you promote. Promoting a Star Prism at Elite rarity or above lets you spend 5 engravers to preserve the first-slotted gem’s tag set. If you have already rolled a gem into its ideal tags, do not forget to pay that at promotion — skip it and every re-roll you just did resets to nothing.

What to do now

  • Engrave all three lines for damage, unless you are a support class bottlenecked on buff uptime.
  • Confirm tag coverage before you look at the values. On a low-coverage tag, anything you engrave is waste.
  • Pay the 5 engravers before promoting to preserve the tags. Get the order wrong and you start over.
  • Do not stall stage 5 progress chasing stage 7’s 2.2%. From 2.0 to 2.2 is +0.2 percentage points, or +0.6% across three lines. The same resources spent on sockets you have not filled yet return several times that.

What would overturn this

The conclusion that a cooldown line does not convert rests on a qualitative Korean-community description — that skill wind-up and recovery absorb the shortened time — and not on measured data. Nobody has published a measurement of “x% cooldown reduction buys y% more casts.”

If someone runs a controlled three-minute fixed rotation on a long-cooldown skill and measures the cooldown line’s actual increase in casts as near-linear — that is, −1.95% really buying +1.95% more casts — then 3 to 1 becomes 1 to 1 and this page’s conclusion flips with it.

The per-stage percentages are on firmer ground: they are constants out of the simulator, and stage 7 has two independent sources agreeing. Stage 3’s per-line values are absent from both sources, so nothing is priced onto that row here either.