D&D 5E Is Savage Attacker worth it?

[Mirtek]

I think I understand what you're trying to say. There are diminishing returns for each additional die being rerolled, but each additional die added to the pool still increases the expected value. If the DM allows Savage Attacker to work on Sneak Attack dice, High level Rogues may get the greatest DPR boost from Savage Attacker.

No, I am saying the get the least. It's not like great weapon style that lets you re-roll 1's and 2's, you have to re-roll your entire damage and take either total.

If you roll 11d6 and use savage attacker, you have to re-roll your 5's and 6's together with your 1's and 2's. You can't just re-roll four from the bunch and keep the others, you have to re-roll all and you'll not that likely to end up with a much different result while rolling 11 dice.

Sure, there might happen the rare case of re-rolling your five 1's and end up with a vastly better result, but with 11 dice being rolled and re-rolled that might happens once in 500 rolls while re-rolling a single d6 leads to an increase every third roll

 

[sandvirm]

No, I am saying the get the least. It's not like great weapon style that lets you re-rolle 1's and 2's, you have to re-roll your entire damage and take either total.

If you role 11d6 and use savage attacker, you have to re-roll your 5's and 6's together with your 1's and 2's and you'll not that likely to end up with a much different result.

Sure, there might happen the rare case of re-rolling your five 1's and end up with a vastly better result, but with 11 dice being rolled and re-rolled that might happens once in 500 rolls while re-rolling a single d6 leads to an increase every third role

I understand how Savage Attacker works and that all dice are rerolled. Savage Attacker allows one to pick either sum; it is does not force the second sum for better or worse. The math is quite clear and no matter how many dice are being rolled, the expected value of max ( sum(xdy), sum(xdy) ) will ALWAYS be greater than the expected value of sum(xdy) for any x > 0 and y > 1.

E(sum(1d6)) = 3.5; E(max(sum(1d6), sum(1d6))) = 4.47; Δ = .97
E(sum(2d6)) = 7.0; E(max(sum(2d6), sum(2d6))) = 8.37; Δ = 1.37
E(sum(10d6)) = 35; E(max(sum(10d6), sum(10d6))) = 38.05; Δ = 3.05
E(sum(20d6)) = 70; E(max(sum(20d6), sum(20d6))) = 74.31; Δ = 4.31

The only thing that can be said about adding more dice is that each die, regardless of size, will provide diminishing returns.

Even the application of Savage Attacker is more straight forward for a Rogue ‒ use it every single time Sneak Attack is used.

 

[The Myopic Sniper]

My problem with Savage Attacker is more that the flavor doesn't match the mechanic at least in my conception I would think a feat like that should do. If anything the -5atk/+10dmg mechanic from Great Weapon Master and Sharpshooter probably more mentally lines up with what I would expect it to do but they seem to have used that to model precision.

Maybe the feats should have been more like this.

Savage Attacker - Before you make an attack you can choose to take a -5 penalty to get +10 damage.

Sharpshooter - No disadvantage from attacking at long range. Ignore cover. Once per round you can reroll ranged weapon damage and take the higher total.

Great Weapon Fighter - You can make one extra attack on a crit or if you drop an opponent. Once per round you can reroll melee weapon damage and take the higher total.

It is a bit of a nerf on the two other feats, but it brings things a bit more in line with my expectations of how the feats feel at the table. Would something like this work at tables where those other two feats are banned or is -5/+10 still a problem with certain party compositions?

 

[jgsugden]

DPR is not the tool to evaluate Savage Attacker. Savage Attacker has an impact on average damage per round, but it isn't really about increasing your average damage, it is about reducing the chance you do minimal damage. A flat bonus of +1 to damage shifts the amount of damage you do per strike up by 1 no matter what you roll. When you use Savage Attacker, you get a chance to redo a bad roll.

Is that a meaningful difference? In some games, yes. In others... not so much.

For example, let's say I'm in a game where I'm a 3rd level fighter (not a champion) that has just been told to level up to 4. I have the duelist feature and an 18 strength with a d8 one handed weapon. My current attack bonus and damage is +6 for d8+6. Which is better: +2 strength (taking me to a flat +7 for d8+7) or Savage Attacker (keeping me at +6 for d8+6, but letting me reroll damage once per turn). Most people would jump at the +2 strength.

However, what if I told you that I was going to be facing exclusively one monster in battle after battle - a fodder monster with AC 15 with 11 hps. I would only get one attack per round in almost all circumstances?

Any 2 blows will be a kill and one blow might kill the enemy, regardless of which option I choose.

If I go with the +2 strength, I crit 5% of the time (killing in one blow 95.3125% of the time I crit) and hit 60% of the time (killing in one blow 62.5% of the time). Chance of a kill on one attack: 42.265625%. If I hit and do not kill, the chance of a kill on the second attack is the chance I hit: 65%.

If I go with Savage Attacker, I crit 5% of the time (killing in one blow 95.3125% of the time I crit) and hit 55% of the time (killing in one blow 50% of the time, and rerolling on the other 50% raising the chance of a kill to 75%). Chance of a kill on one attack: 46.015625. If I hit and do not kill, the chance of a kill on the second attack is the chance I hit: 60%.

Net effect: Savage Attacker makes me a better fodder clearer, but if I fail to clear the fodder in one blow the advantage reverts to the flat bonus to hit/damage.

 

[Sword of Spirit]

Net effect: Savage Attacker makes me a better fodder clearer, but if I fail to clear the fodder in one blow the advantage reverts to the flat bonus to hit/damage.

I still don't think it's worth it, but thanks for the evaluation. That makes it easier to see how different variables can actually influence its effectiveness.

 

[CapnZapp]

You can prove anything with statistics.

But what this tells me is that not even statistics can save this weakass feat.

Thanks for your attempt at making it look worthwhile. If that's all you can do, then it truly is excruciatingly worthless.

 

[FrogReaver]

My DPR / my attackers' DPR. Best computed as a party metric.

Quick example: if I'm a fifth level DX 18 fighter with Sharpshooter, a rapier/shield, archery style, and a longbow, I can be either AC 18 doing 10.4 damage to an AC 14 minotaur per turn, or AC 20 doing 6.18 points of damage. For simplicity, let's say I have two identical CH 18 AC 13 warlock buddies who can each do 8.8 damage per round to a minotaur using Eldritch Blast + Hex from range, or 5.9 from melee due to disadvantage. (I have scripts to calculate these values, if you want them.)

There are six possible configurations if we ignore minotaur charge (assume I know about charges and I'm savvy enough to prevent it from happening):

Minotaur attacks me (AC 20 rapier and shield): I take 6.6 DPR, we inflict 6.18 + 8.8 + 8.8 DPR, loss ratio is .27.
Minotaur attacks me (AC 20 rapier and shield), using Reckless attack: I take 11.09 DPR, we inflict 8.17 + 11.7 + 11.7 DPR, loss ratio is .35.
Minotaur attacks me (AC 18 configuration): I take 8.3 DPR, we inflict 5.61 (bow at disadvantage) + 8.8 + 8.8 DPR, loss ratio is .36.
Minotaur attacks me (AC 18 configuration), using Reckless attack: I take 13.13 DPR, we inflict 10.4 + 11.7 + 11.7 DPR, loss ratio is .39.
Minotaur attacks warlock: Warlock takes 12.55 DPR, we inflict 10.4 (using my bow) + 8.8 + 5.9 (warlock at melee disadv) DPR, loss ratio is 0.5.
Minotaur attacks warlock, using Reckless attack: Warlock takes 16.74 DPR, we inflict 15.2 (using my bow) + 11.7 (warlock at adv) + 8.8 DPR 0.47.

Conclusions:
1.) It is advantageous for me to put away my bow and engage the minotaur in melee instead of letting the warlock do it, even though that prevents me from benefitting from Sharpshooter.
2.) It doesn't make sense to Legolas in this situation even if the Minotaur is using Reckless Attack, because while my damage goes up, so does my damage taken, and my loss ratio suffers.
3.) The minotaur should use Reckless attack against me. Against the warlocks it's better not to go Reckless.
4.) Eyeballing the numbers above, I should probably dodge instead of attacking with my rapier. (Checking the numbers, I find that Dodging gives a loss ratio of 0.12 if the minotaur attacks normally, or 0.28 if it uses Reckless attack, so Dodging helps a little but not that much.)

Of course the best loss ratio is infinite: at range, we inflict 28 points of damage per round and take 0 DPR in return. But if I can't arrange that, I should put down my bow and front-line it with my rapier.

Hope that helps,
Max

Interesting statistic. I'm debating it's merits.

It seems a very useful tool to try and determine the best course of action in combat with.

However, the more party hp values differ the less useful the concept is.

For example, a 1 DPR monster with 500ft move speed and 1,000,000 hp would kill everything. Even though by your statistics standards it would have the lowest loss ratio in virtually every scenario.

The point is that your tool is only useful when it comes to comparing things with similar hp values. (Which is a similar flaw in DPR as well, however, DPR doesn't attempt to take into account defensive abilities of a character, so you are at least accounting for defensive attributes).

-----------------------------------------
I would actually anticipate that looking at something like your "loss ratio" divided by "hp ratio" would be an even better metric. Essentially it is the same thing as comparing the number of rounds on average the enemy will last against your DPR vs the number of rounds on average that you will last against the enemies DPR.

 

[FormerlyHemlock]

Interesting statistic. I'm debating it's merits.

It seems a very useful tool to try and determine the best course of action in combat with.

However, the more party hp values differ the less useful the concept is.

For example, a 1 DPR monster with 500ft move speed and 1,000,000 hp would kill everything. Even though by your statistics standards it would have the lowest loss ratio in virtually every scenario.

The point is that your tool is only useful when it comes to comparing things with similar hp values. (Which is a similar flaw in DPR as well, however, DPR doesn't attempt to take into account defensive abilities of a character, so you are at least accounting for defensive attributes).

-----------------------------------------
I would actually anticipate that looking at something like your "loss ratio" divided by "hp ratio" would be an even better metric. Essentially it is the same thing as comparing the number of rounds on average the enemy will last against your DPR vs the number of rounds on average that you will last against the enemies DPR.

It depends really on what decision you're trying to evaluate. Loss ratio is indeed a tool for greedily evaluating tactical decisions. It's by no means a perfect metric (it's greedy). It won't tell you who's going to win the combat, it will only tell you which configurations are more or less advantageous during a combat. I do think it's better than DPR though, with minimal additional complexity (it's just the ratio of DPR sums). If you tried to factor current HP into the calculation you'd need to recompute every round.

To put it differently, loss ratios help you answer the question, "How do I minimize the resource expenditure during this combat?" If you factor current HP in there, it changes to, "Am I going to win or lose this combat?" I confess to being less interested in the latter question than the former, especially since factoring in current HP might lead you to suboptimal resource usage, e.g. preferring doing 150 damage to the barbarian with 200 HP instead of 39 HP to the rogue with 50 HP even though magically healing the rogue for 39 HP is easier between combats than healing the barbarian for 150 damage. (Yes, with HD healing they'd be about the same.)

Anyway, that's what loss ratios are. I prefer them to DPR.