A DPS Simulator

[EasymodeX]

What is the window you are using to calculate each DPS datapoint?

Also, when are you applying TOD and Demolish for both of those? After you get GL3 or as soon as possible?

[KroLeXz]

So out of the classes you have more or less successfully simulated, which is siming the highest?

[pandabearcat]

What is the window you are using to calculate each DPS datapoint?

Also, when are you applying TOD and Demolish for both of those? After you get GL3 or as soon as possible?

Each point is an event, in the case of stacking events, it only does the first event. So Ticks won't register until the next event after.

For PB, demo is applied after BfB/IR but before GL3 (I think it ends up being either GL0 or GL1). ToD is not applied until PB finishes and 3 stacks of GL (still before BfB/IR ends). FR/ID is not used.

For non PB, demo is applied on first GL0.

PS dps is calculated from start of the fight till whatever second.

I could do a rolling window thing but that is equivalent to doing a derivative graph and finding zero, which is what this shows anyway and more intuitively (as in higher is better, etc).

So out of the classes you have more or less successfully simulated, which is siming the highest?

Monks by about 8% higher than dragoons. Bards are 1-2% pure potency ahead of dragoons but their weapon damage is 10% lower, so likely they are lower damage overall.

Note that the bard does not have DE applied. With DE its something like a 6-8% dps increase.

[EasymodeX]

I'm skeptical of the charts. Couple comments:

1. It takes roughly 25s to stack GL3. All attacks after GL3 will have maximum potential DPS. A result of 30-45s is questionable at face value.

2. The length of time that it takes to achieve the DPS or damage target seems to be artificially extended in the second scenario by holding the BFB/IR cooldown until GL stacks are acquired. While this may or may not make sense as a general approach/strategy (I'm actually in favor of using BFB/IR immediately regardless of GL stacks or stacking approach), either way the timing of the external cooldown is artificially skewing the analysis of PB with regards to ramp-up time.

E.g. the longer you wait on big cooldowns, obviously it will take longer to reach maximum DPS. That's a result of when you press BFB, not a result of stacking GL3.

As a result, I recommend removing BFB/IR from the simulation for the comparison.

3. I recommend using some form of rolling window to chart the DPS if possible. The FF14 GCDs are large and the Monk GCDs fluctuate, so this is cumbersome, though. Off the top of my head I can't think of a good method for this. If you're confident in the approach you're taking for measuring/charting the result, then w/e.

Sidenote: after further review, is each datapoint measuring DPS by taking the total damage by that time and dividing by the time elapsed? And, I guess you're marking the average DPS from the full simulation and marking the first point at which your ramp-up hits that DPS threshold? That seems viable enough if so.

[pandabearcat]

Basically the chart represents what you would see if you looked at a realtime Recount parse or a FFXIV parser parse, or realtime dps graph.

What you say is true, after ~20 seconds of rotation you are at GL3 with all buffs accounted for, so you are at "optimal" damage dealing potential. We can count this as ramp up time, so we could say monks' ramp up time is 20-25 seconds.

However, this is discounting the 20 seconds of crap dps you are outputting, aka, vs someone with 0 ramp up time (i'm aware everyone has ramp up time, but lets just say you're versus a bard or a summoner who just bane'd).

So when I say it takes 35-45 seconds to do average dps, I mean how long it would take you w/o PB to well, do what a zero (or low) ramp up time character would do. Aka, this is how long it would take before monks actually become the "kings" of single target dps.

Hopefully I can graph this with multiple classes at some point, to show you when the classes start doing similar damage.

EDIT: about the BfB/IR thing, if I took those out the graphs would never reach optimal dps (at least not in 60 seconds) since you're always playing catch up from starting w/ 0 stacks, so they must be used at some point.

If I used BfB off the start w/o PB, I don't actually know if it would shorten ramp up time or not.

[EasymodeX]

Yeah that's what I figured. That's fine -- but you need to remove BFB/IR from the simulation, or have the second comparison use BFB/IR immediately rather than waiting the ~20s. That delay artificially skews the result against the second test.

[KroLeXz]

Panda where do summoners fall on sims?

[pandabearcat]

Panda where do summoners fall on sims?

Nowhere atm, no casters are currently simmed.

One of the main reasons for is that also BLM and SMN have much lower potency, but much higher weapon damage than physical dps.

Another reason is that noone knows approximate boss resistances.

People are now casting doubt on the valk formulas, so it may turn out all our stat weights are wrong too, so its hard to convert potency to dps.

What is easier to see though, is how all the phys dps stack up to each other, and how the caster dps stack up to each other.

Sometime I might get SMN and BLM numbers, but its holiday season and I have a lot of work to do, so progress has been slow.

Suffice to say though for now, if you play your class well, there is no reason why you can't do any part of coil with any class, with the caveat that you have at least one BLM, because they are so broken resourcewise, it makes everything much, much easier.

An addendum to that is to bring a bard, because that also makes it much easier on your healers (and summoner).

[KroLeXz]

parser are by no means perfect but in a much much better place - summoner seem to be consistently #1 on there by a good amount - and right now logrep2 and ffxiv-app don't track shadowflare or increased dmg from foe of requiem. And it mostly because most encounters in coil for example put you in a good position to bane a target - Turn 5 which I recently downed I could bane scourages/conflags/Snakes/Dreadknights - which by the end of the encounter the effective dmg done on that fight for a summoner is just leaps ahead of other classes.

[pandabearcat]

parser are by no means perfect but in a much much better place - summoner seem to be consistently #1 on there by a good amount

Sure thats fine, but thats not really the primary purpose of the sim either.

The sim wants to know if

• Any class is significantly worse. This has not been the case
• Analyze TP efficiencies and stuff. SO far it turns out all TP using classes run out WAY before 10 minutes full uptime.
• Help people find the best rotation for their class, so intraclass comparisons. See the priority vs rotation tests about halfway through
• Again, the sim isn't here to prove anything, only to provide good evidence of common player fallacies and hopefully debunk anyone's idea that their class is "super underpowered"
• And the reason it was written in the first place, to judge the relative worth of procs and other randomness that is difficult to correctly (with 100% certainty) predict in a mathematical model. That is the primary strength and the difference between models, sims, and parsers, why we use all three.
  
  Parsers are (supposedly) the most accurate prediction of dps in a "real" scenario, but suffer from repetitiveness, lack of trials, and general bias of players who's personal anecdotes always end up with their class being either completely terrible or demigods of dps.
  
  Sims provide great statistical data, can run many hundreds of thousands or millions of trials it takes to prove statistically whether one rotation or class is "better" than another, but suffer from general inflexibility (trying to fix this, but this obviously makes it complicated), and requires actual knowledge of the damage formulas (which none of us actually know right now).
  
  Models provide more simplistic views but are quick to formulate and provide good mathematical evidence one way or another of proving a baseline dps or "theoretical" limit to shoot for. They are good checks for sims, and are good to base things off of because it can easily be determined why they are wrong or right.

This being said, I am REALLY looking forward to actually getting a good BLM sim together, to finally debunk the old BLM vs SMN debate, since it is REALLY hard to say for certain whether a model for BLM damage is really accurate or not.