Asking a Favor - Vault Frame / Breakpoint Testing

Demon Hunter
1

A while back I did some asking and digging around on how Vault’s frames are calculated. I’m fairly certain I reconstructed the formula from the information I could find. However, I’m not able to test it in-game reliably, so I am asking if someone is able to do so. Below is the formula, as well as the frames and movement speed % breakpoints that would be necessary to attain those frames if this formula were correct.

Formula:
Frames/Vault = 4+CEILING(17.999996185303*(35/(24*(1+MS%))))

  • MS% refers only to increased movement speed. I.e. it starts at 0%, not 100% with no bonuses. For this formula, +100% would be represented as 1, not 100.
  • The 24 comes from your base movement speed in yards / second.
  • The 4 frames are 2 frames at the start and 2 at the end of the animation that do not appear to scale and appear to be the frames where you are vulnerable to damage, while the 17.999996185303 should be scaled with movement speed as indicated and should also be the invulnerability frames.

Expected / Predicted Values:

Frames/Vault Vaults/Second MS% Breakpoint
5 12.00 2,525.00%
6 10.00 1,212.50%
7 8.57 775.00%
8 7.50 556.25%
9 6.67 425.00%
10 6.00 337.50%
11 5.45 275.00%
12 5.00 228.12%
13 4.62 191.67%
14 4.29 162.50%
15 4.00 138.64%
16 3.75 118.74%
17 3.53 101.92%
18 3.33 87.50%
19 3.16 75.00%
20 3.00 64.06%
21 2.86 54.41%
22 2.73 45.83%
23 2.61 38.16%
24 2.50 31.25%
25 2.40 25.00%
26 2.31 19.32%
27 2.22 14.13%
28 2.14 9.37%
29 2.07 5.00%
30 2.00 0.96%
31 1.94 -2.78%*

*The negative percent here just means you could be slowed down 2.78% and still Vault at this speed, but further than that would begin to drop your frames.

4 Likes
2

Holy crap! Thanks for doing this.

I’d done some testing with it before to try and optimize my Sprinter/Speed Racer setup. Did enough to verify that it scales with movement speed and not attack speed (in the past there was a subset of speedrunners claiming it scaled with attack speed and I was like “I’m pretty sure that’s not true…”). Didn’t go any further then that because it seemed like a lot of work.

Not going to promise anything, but I’ve done some breakpoint verification on wizard skills (particularly for teleport, when I was optimizing my wizard Sprinter/Speed Racer setup). I’ll try and check this when I get a chance. I won’t verify every breakpoint, but I’ll try and check at least a few different ones, using paragon to get just under then just over the breakpoints to see if the animation frames change. That should at least give us an idea if they seem accurate or not.

3 Likes
3

Nice work on the tabulated data. I have to ask where the formula comes from and why does it include the very strange number 17.999996185303? Wouldn’t it make more sense for this to be just 18? If it’s empirically calculated, maybe the formula could make more sense with a floor function instead.

As for testing, it is difficult to test frame data in D3 since you can’t slow down the game for a frame-by-frame analysis (using tool to do just that would likely be against the ToS too). The best option one has is to do a video recording and hope that the frame rate and internet connection on your computer is very stable.

Edit: upon closer inspection, the number 17.999996185303 appears to be very close to 18 - 2^(-18). I’m not sure if this is where the number was built from but single or double precision numbers are more accurate than that.

4

That’s some excellent work, Childe!

Do you know what the actual maximum MS% achievable in the game is? 2525% and 1212.5% both seem impossible…

I’m not that well versed on a lot of DH mechanics these days, but it occurred to me that you might be able to extract some useful information by taking a larger sample size of vaults and using the Action Shot rune and Bane of the Stricken.

In other words, it’s going to be pretty tough to distinguish a “12 frame Vault” from an “11 frame Vault” on an individual basis, but if you vault for 2 straight minutes around an enemy, and then check the number of Stricken stacks, you may be able to discern some of those nuances a little better.

That said, I have no idea if there’s any weirdness in the way that Action Shot stacks Stricken. I suppose you’d have to test that first, to be sure.

I guess the individual Vault animations are pretty clear, so if you review some video carefully, you can just count them up (again, over a longer timeframe). Probably no need to resort to any of this Stricken nonsense…

5

Yeah I’m working on it now, it’s actually not that bad. It’s a little hard to tell where the bookend frames start and end (the 2+2 Childe mentioned), but the variable frames in the middle are super easy because you change colour (blue for Tumble). So I’ve just been counting the ‘blue frames’.

Lower BPs look good so far, going to check some of the upper ones now.

6

Awesome! Let me know if it aligns.

This comes from a file. I don’t think I can really say more due to weirdly restrictive forum rules. It doesn’t make much practical difference, as it basically just means a few values differ by a hundredth of a percentage point, which would come into play exceedingly rarely.

Yeah, that’s why I can’t test reliably myself.

I think you can get into the +500%s from available buffs (Pylon, Warzechian, Rechel’s Ring, Tactical Advantage, etc.). The 2525% is just the result of the formula (if the formula is correct) and would represent a theoretical hard cap, since you cannot reduce below 1 frame with movement speed, and the 4 other frames cannot be reduced at all.

7

Oh, I forgot to ask, there’s no difference between male and female DH right?

I remember Barbs had different leap speeds based on gender at one point before it was patched!

8

It’s been months since I actually dug into the files, so I don’t recall offhand, but as best I recall they were the same values in both. If anyone wants to validate, you’ll want to search for “*backflip”

9

That’s what I’m trying to do. I fished for an east coast server game for lower latency. Dialed my settings way down to try and lock in the frame rate.

It’s definitely not perfect, I get a little variation. So I do frame-by-frame playback of like 10-15 vaults at each point I’m testing, then look to see what the consistent minimum frame length is.

Was working okay at the lower movement speeds, but it seems to be breaking down at higher movement speed. My methodology might be flawed. Not sure if the ‘blue frames’ actually match up with the animation. Or I’m just not getting steady enough latency/frame rates.

EDIT: redacted results, just realized I made a mistake in my testing. Will update later.

EDIT2: Okay, fixed what I messed up before, but still getting weird results.

I give up.

(just counting the middle animation, not the bookend frames, so add 4)

MS% Expected Observed
0 27 27
1 26 26
5 25 25/26?
5.5 25 25
9 25 25
9.5 24 24
14 24 24
14.5 23 23
19 23 23
19.5 22 22
60 17 16/15?
115 13 13
120 12 12/11?
220 9 8
230 8 8/9?

Was getting a lot of variation at exactly 5.0% MS. Not sure what was going on there.

For the most part the lower BPs seemed to match up okay. It got wonky at the higher BPs though. No idea what was going on at 60% MS, was consistently getting 1 or 2 frames shorter than expected.

220% vs. 230% MS was also very strange. At 220%, I was consistently getting 8 frames, 1 fewer than expected. Then at 230% MS, I was getting a mix of 8 and 9 frames, which doesn’t even make sense. It should either be the same frames or lower, not higher… Was getting a lot more variation at 230% MS than I was at 220% MS.

I think my methodology is flawed. I’ll leave this here in case it’s useful though.

2 Likes
10

Unfortunately I’m not certain what would cause these oddities. That said, it looks like overall the testing tracks very close to the formula even in the couple spots where there are anomalies currently. That’s an exciting development for at least initial results. (Possibly others with reliable connection / low latency could test as well.)

Two questions:

  • what sources of MS% were you using for the 60, 120, 220 and 230? I don’t think this would be the culprit, but I figure it’s worth exploring.
  • how many milliseconds was your latency? I wonder if that might also relate to the variability at certain increments, i.e. if it is syncing up, causing more command queue issues.
1 Like
11

0-25% was all paragon.
60%: Tactical Advantage, 0 paragon.
115%: Tactical Advantage, Krelm’s Belt, Shadow Power: Shadow Glide, 60+25+30 = 115.
120%: Tactical Advantage, Krelm’s Belt, Shadow Power: Shadow Glide, 5% paragon. 60+25+30+5 = 120
220%: Tactical Advantage, Krelm’s Belt, Shadow Power: Shadow Glide, Smoke Screen: Displacement, 5% paragon. 60+25+30+100+5 = 220.
230: Tactical Advantage, Krelm’s Belt, Shadow Power: Shadow Glide, Smoke Screen: Displacement, 15% paragon. 60+25+30+100+15 = 230.

For ones with Tactical Advantage, I would string a series of vaults together in a row, ignoring the first vault, and the rest close enough together that Tactical Advantage didn’t drop.

For Shadow Power, I just used Shadow 4 so I cast it once and then it was up forever.

Krelm’s should just stay up.

Smoke Screen I would cast it and then quickly vault a couple times afterwards. I made sure for any of the ones I counted for that I had it active for the entire vault animation.

I was thinking that too. Couple places it’s off, but only by a frame, for the most part tracking pretty well. Good enough for my Sprinter/Speed Racer optimization, which was my primary interest in this.

Yeah I wondered that too. There was definitely some instances where it seemed like my local instance lagged behind by a frame and then caught back up. I tried to avoid counting any like that. I deliberately fished for an east-coast game to bring my latency down. I’m typically 40-50 ms for east coast server games and ~90-100 ms for west coast server games. Any time I checked I was pretty consistently around 45-50 ms in the game I was testing in.

3 Likes
12

Are you considering attack speed in your tests?

13

Tinne can of course weigh in as to whether or not AS% is varying between those tests, but attack speed should not be affecting Vault, contrary to what some people have said. Some skills have “Attack Speed: Attacks_Per_Second_Total.” Vault is not one of them. Vault has “Attack Speed: 1,” meaning the portion of its animation that in theory would scale with attack speed multiplies by 1 instead of by your attacks per second.

14

I kept attack speed constant. I did switch gear partway through the tests to put 4p Shadow on to keep Shadow Power up, and I caught myself because I was going to toss on my Shadow gloves (have attacks speed) but I just used a different piece so that my attack speed was constant.

I’d actually done some testing in the past with Vault and it seemed like it didn’t scale with attack speed at all to me. Between that and what Childe is reporting it seems like attack speed shouldn’t be a factor?

15

That seems to be the case with consecutive vaulting. There’s a distinct improvement in responsiveness/feel when you switch between vault and any APS based attack (Multishot). (typical play in speed rifts + grifts) – due to the frames being occupied by the Multishot animation.

Use a skill like Cluster Arrow with Manticore equipped – try vault + shoot + vault + shoot. Then switch to Multishot with Yang’s and you’ll see the immediate benefit of the AS.

I know that’s not what this post is about. Perhaps… soon we’ll consider that nice table the OP made and this: figuring out how to match the Vaults per Second to the Attacks per Second and how it might help make the game play feel more fluid for Vault-based speed builds. Thus helping us decide how much movespeed we really need if we use Vault instead of Smoke Screen.

I’m looking at this breakpoint specifically:

17 3.53 101.92%

You need 7% AS on the Bow, Rings, Gloves and 20% AS on the Quiver.

When you cube Echoing Fury instead of Dawn, you’ll want even more movement speed. Which will help quicken animations of skills not buffed by Yang’s Recurved.

Side topic: Does Vengeance affect vault – since this skill creates attacks as you vault. From observation, it doesn’t seem to impact it, but maybe there is some proof out there?

16

Oh yeah I absolutely agree there.

Also a noticable difference between vault + Multishot (w/ Yang’s) vs. vault + Evasive Fire (at least without Hunter’s Wrath).

That’s a neat idea.

1 Like
17

Would dual wielding change these numbers with the alternating shots?

18

No.

Additionally, Action Shot does not alternate your weapons. It uses the one currently queued. This creates an interesting situation where you can actually roll damage off of one of your weapons in favor of another affix as long as you properly queue your weapons before starting.

19

Thanks for that info

20

Neeeeerrrrrrrdddddsss!

This is great, I haven’t seen this much work in a lonnng time

2 Likes