[TC] Chasing BIS: 2T16H+3HWF or 4T16H?

[Komma]

"What is BiS for Arcane?"

Most of you have probably seen many duplicates of that question asked on every mage forum and Arcane guide thread. In most up-to-date guides (ie. Voltaa's compendium), two variations are often quoted as BiS:
1. Itemlevel 588, 4T16H with HWF Garrosh Helm
2. Itemlevel 589, 2T16H with HWF Garrosh Helm, TFP Shoulders, Siegecrafter Gloves
The current belief is that 4T16H is recommended for progression which heavily focuses on single target, while 2T16H is believed as "roughly the same on single target, better for multitarget fights and rank chasing".

The goal of this thread is to thoroughly check whether this belief stands the test of analysis.

Setup

Similar to most of my TC posts, the primary method of analysis will be done with Simulationcraft. I have enough confidence in the current state of Simulationcraft (Ver. 548-7) to model fights between 1 to 3 targets, while the AOE sequences are a little less optimized. However, there are very few fights with more than 3 consistent targets, so I feel it still gives a sufficiently complete picture.

As mentioned earlier, there are 2 sets being compared:
1. Itemlevel 588, 4T16H with HWF Garrosh Helm. This profile has 13959 haste and 21557 mastery from gear.
2. Itemlevel 589, 2T16H with HWF Garrosh Helm, TFP Shoulders, Siegecrafter Gloves. This profile has 12424 haste and 23904 mastery from gear.
By default, profiles will be using Frost Armor and Living Bomb. Variations using Mage Armor or NT will be denoted with name suffixes.

Single Target


Single target, Patchwerk fight, 450s fight duration, 100K iterations, DPS Error ~= 120.0

The 4T16H profile leads 2T16H by a small margin of 0.638%. This is unlikely to be noticeable in real raid conditions. Using Mage Armor with 2T16H causes a loss of notable size - this is most likely due to the extreme levels of mastery achieved with the additional mastery pieces. This agrees with the Frost Armor vs Mage Armor analysis from my earlier thread: http://altered-time.com/forum/viewtopic.php?f=3&t=313" target="_blank

The 2T16H profile has several points of interest. Aside from the expected higher share of damage done by AB/ABar, of interest is the haste/mastery reforge chart for it:

Origin has 12424 haste and 23904 mastery. Haste/mastery values in the graph are before amplification by PBoI.

The value of Haste is practically equal to Mastery, up until 14179 haste. This corresponds to the 14242 haste BP, and the only spell that suffers beyond this point should be Arcane Barrage. This is very different from the reforge chart of 4T16H:

Origin has 13959 haste and 21557 mastery. Haste/mastery values in the graph are before amplification by PBoI.

Mastery is clearly stronger than haste with the 4T16H profile, and gunning for 14242 haste is a clear loss. This is significant, as we will find out later.

Three Targets

Before we start, I should explain that Simulationcraft has no simple way of separating multiple mobs. As such, Mage Bombs will always cleave all targets. I'll mention it if it affects our results.

Three targets, Patchwerk fight, 450s fight duration, 100K iterations, DPS Error ~= 175.0

The first test seems to agree with our assumption - 2T16H with LB on multiple targets has a small 0.758% advantage over 4T16H. The damage breakdown shows that explosion damage makes up 10.0% for both 2T16H and 4T16H, which means that LB cleave isn't skewing results for us. Of note is how Mage Armor falls further behind on multiple targets, contrary to popular belief.

Again we can look at the reforge graph for 2T16H:

Origin has 12424 haste and 23904 mastery. Haste/mastery values in the graph are before amplification by PBoI.

Once again, it shows that we can trade the additional mastery for 14242 haste without losing any damage. This is very different from 4T16H, which shows a very erratic graph:

Origin has 13959 haste and 21557 mastery. Haste/mastery values in the graph are before amplification by PBoI.

When talking about 3 target fights, the first one that comes to mind is TFP. Nether Tempest should be used instead of Living Bomb on this fight.

Three targets, Patchwerk fight, 450s fight duration, 100K iterations, DPS error ~= 224.0

"...Wait, what?"

Contrary to expectations, the 4T16H profile more than catches up as we switch to NT, and actually takes a tiny lead just larger than the error bar. Why is the change to NT creating such a huge difference in results?
The first answer that came to my mind was Haste. Nether Tempest does not have a fixed 1 second GCD, unlike Living Bomb. The 2T16H profile has 12424 base haste, which is ~1800 off the 14242 GCD cap with Frost Armor.

To verify this, we look at the reforge plot for 2T16H:

Origin has 12424 haste and 23904 mastery. Haste/mastery values in the graph are before amplification by PBoI.

The reforge plot can essentially be interpreted as "HASTE! MOAR HASTE! MOAR HASTE!!11". The final reforge point with a huge jump is at 15934 haste, which has just achieved the 15832 NT breakpoint. This peak, at 874911 DPS, holds a 1.38% advantage over 4T16H - the biggest difference we've seen so far.

Now let's go back to what we noted earlier:
1. Reforging to 14242 haste with 2T16H produced <0.1% difference on single target, no ~0.0% difference on three target LB fights
2. Looking at the reforge graph on 3 target NT, 14242 haste would provide a ~1% gain.
I think it's safe to conclude that 14242 haste is a much better bargain overall.

2T16H with 14242 Haste

There's not much left to be said at this point: it's just charts and numbers. But first we need a setup: Click here for what I used as the 2T16 14242 profile.


Single target, Patchwerk fight, 450s fight duration, 100K iterations, DPS Error ~= 120.0

Three targets, Patchwerk fight, 450s fight duration, 100K iterations, DPS error ~= 174.0

Three targets, Patchwerk fight, 450s fight duration, 100K iterations, DPS error ~= 224.0

What about shorter fights? Wouldn't higher haste builds hurt more due to GCD capping during Bloodlust?

Single target, Patchwerk fight, 180s fight duration, 100K iterations, DPS Error ~= 212.0

Three targets, Patchwerk fight, 180s fight duration, 100K iterations, DPS error ~= 326.0

Three targets, Patchwerk fight, 180s fight duration, 100K iterations, DPS error ~= 255.0

Changing from 7.5 min to 3 min fights does not change its lead on 3 target NT. On single target it hurts a bit more, but 4T16H reigns supreme on single target anyway.


Conclusions

To summarize what we've found so far:
1. 4T16H has a small advantage on single target fights.
2. 2T16H has a small advantage on 3 target LB fights.
3. 4T16H and 2T16H are exactly even on 3 target NT fights, EXCEPT:
4. Reforging to higher haste gives 2T16H a notable gain on 3 target NT fights, while staying even on 3 target LB.
5. Once again, Frost Armor > Mage Armor in every scenario.

I offer two main conclusions/propositions based on the above. They also work as answers to the original question.

"The 4T16H BiS set is undisputed BiS for single target."

"The 2T16H set is BiS for multi target IF reforged to 14242 haste. 15836 haste is probably BiS for NT fights(ie. TFP/Spoils)."

[voltaa]

Beautiful analysis Komma! Though I'm going to be a downer on the parade still and remind everyone that while the graphs and sims may show things to be a certain way, that in practice things may not always be as they seem in theory. Particularly, I've been skeptical of Simcraft haste values (for all mage specs) for quite some time now (the whole expansion).

Now I sm not saying this is wrong and I don't want people to misconstrue this as me saying that simcraft is in fact wrong, because no one knows just how accurate it is at this point. I just want to say that we should still take this wiyh a grain of salt but test the shit out of it.

[Frosted]

Particularly, I've been skeptical of Simcraft haste values (for all mage specs) for quite some time now (the whole expansion).

What is happening in SimC to make you question this? I'm not saying you're wrong - more so that I'm interested in looking into whatever seems off so we can bring it in line with what you see in game.

We have not begun to really delve into the extreme nity-grity spell by spell verification yet (I think the thought between Komma and I has been to wait until the module for WoD is "done" to do that) but if you have something for us to go on, I'll look into it asap.

[Komma]

Keep in mind that most of these differences are by several thousand DPS margins, in the context of a million or half a million. They're not easily reproduced in raid conditions.

That said, some of the haste stuff is actually done by Asian mages, so I don't think it's pure theory. It's probably a similar case with Frost Armor.

[Cycobi]

It's important to keep in mind that Sims will often overvalue haste.

I'm not disputing FA > MA or anything like that, but it's still an important thing to take into consideration.

Computers don't suffer from latency and don't have to do very quick mental arithmetic to determine whether their current cast will be too long or too short or whatever. Humans do while having to factor latency into account and, as such, Computers can get a much better grip on Haste than we can.

That said, it is very, very good analysis Komma and I'll do what I can to test it out with the 14242 breakpoint next reset and will post results here.

[Komma]

While I'm not sure whether haste is being overvalued, I'd like to clarify a few things regarding latency.

Computers don't suffer from latency and don't have to do very quick mental arithmetic to determine whether their current cast will be too long or too short or whatever. Humans do while having to factor latency into account and, as such, Computers can get a much better grip on Haste than we can.

The simulation does in fact take latency into account: https://code.google.com/p/simulationcra ... ns#Latency" target="_blank

Here is debug output that shows some of the relevant default settings. The default settings roughly correspond to a raider who has sub 100 ping, very good framerates, and plays with a fair bit of button-spamming.

Code: Select all

channel_lag=0.25 channel_lag_stddev=0 gcd_lag=0.15 gcd_lag_stddev=0 queue_lag=0.037 queue_lag_stddev=0 queue_gcd_reduction=0.032 strict_gcd_queue=false default_world_lag=0.1 default_world_lag_stddev=-9.22337e+15 default_aura_delay=0.3 default_aura_delay_stddev=0.05 reaction_time=0.5 aura_delay=0.5

On the other hand, I do agree that the sims have a slightly better grip on haste. Higher haste usually means we make rotational errors more easily, and server lag in heroic 25s is usually a lot worse than the smooth situations noted by these settings.

[Vatti]

I tested the 14242 w/ 2PT16 build tonight. Here are the results:

Hypothesis: I'm a fan of high haste builds which is why I use 4 piece. It is less forgiving when you cast a 5th arcane blast in high haste scenarios because of the additional mana regen from your rune. I'm also a fan of the higher pace just to keep me on my toes and increase my skillcap as a player. I thought, however, that this build wouldn't be as good as the four piece with frost armor due to the lack of the four set. I rate the four set very highly as it is extremely fun to play with/around on some of my more favorite fights in the instance (IJ, Garrosh). Of course I know the higher mastery would be better for more fights in SoO but I stayed 4set/MA for a long time.

Given the amount of time I've been clearing SoO now I looked to the forums for more ideas on how to play and spice up my clears. I'm not a theorycrafter so naturally I sought after theorycrafters to do the dirty work for me. What came out of that search was this thread, and I thought I would give it a shot with low expectations.

Results: Holy. Shit.

This build performs well. Very well. I trashed some of my more prized parses in some environments that weren't favorable with the RNGods. You can find the logs below.

This build fills my need for high pace game play with great emphasis on your fight knowledge to get the most out of the resources given to you. Putting so much emphasis on bomb upkeep and snap-shotting really appeals to me and hopefully over the next couple of weeks I'll be able to improve on that aspect. I have no noticeable gripes with the build but some should surface over the coming weeks.

Ranks: Fallen Protectors (11), Iron Juggernaut (13), Thok (4, but was 2 from last week), Paragons (10), Garrosh (3!!!! I'm coming for you Voltaa)

Conclusion: Try this build if you're 1-2 items off BiS or pulling high ranking parses already and see what you can do with it. Frost Armor is a standard, sure, but if you have the 3 HWF offsets then I invite you to give it a go.

Great work Komma. It's people like you who make these parses possible. Hope your work continues in Warlords as I would be lost at the start of tiers without you math folk.

Logs: http://www.warcraftlogs.com/reports/YBHCvVZRPnp2FcAK" target="_blank
VoD: http://www.twitch.tv/vattitv/c/4736594" target="_blank
Stream: http://www.twitch.tv/vattitv" target="_blank (Follow me~)

[Komma]

Thanks for the compliments! Those are some amazing parses, seriously. Combining this with Cycobi's results, this is starting to look really promising in real raid testing.

I'm actually surprised that you even did that well on IJ - short-duration single target fights is supposedly the weak link for this build.

I noticed that you used LB instead of NT on Dark Shamans - I think NT pulls ahead with how well you stack them, especially given the 1s GCD. And judging from how well the NT fights went, I wonder what would have happened if you used NT and survived the last 2%.

[Cycobi]

I was running ~527K on Iron Jugg before I died, based on our killtime would have had a second AT as well to perk it up slightly.

I'm honestly a massive fan of this build and, like Vatti, can only thank you for pointing me towards it ^_^

[Vatti]

I noticed that you used LB instead of NT on Dark Shamans - I think NT pulls ahead with how well you stack them, especially given the 1s GCD. And judging from how well the NT fights went, I wonder what would have happened if you used NT and survived the last 2%.

Old progress mindset auto-pilot to living bomb. Will try NT next week!

[voltaa]

Ugh, am I gonna have to change my build/BiS list now? JUST WHEN I HIT BIS IT ALL CHANGES!

Rage aside I think I'll try this build for a few weeks and see how it goes. Nothing else to do really, just the same 14 bosses over and over.

[Lionheart]

Ugh, am I gonna have to change my build/BiS list now? JUST WHEN I HIT BIS IT ALL CHANGES!

Rage aside I think I'll try this build for a few weeks and see how it goes. Nothing else to do really, just the same 14 bosses over and over.

Stream or record VoDs if you can.

Also, I watched both cycobis and vattis streams, it's so good to have mage streaming again.

[bazz]

I have the 3 HWF pieces but I only have HWF Hit/Mastery bracers and Crit/Mastery boots. I would have to reforge/gem almost completely haste. I would hit the 50% haste mark and end up with 86% Mastery.

Would this be worth it?

[Trustbucket]

By the way, why wouldn't you use the jugg haste mastery legs for the 2pc build? If you're going to be swapping mastery into haste anyway it wouldn't matter to lose the mastery on hands/ shoulders for increased stats.

[Komma]

By the way, why wouldn't you use the jugg haste mastery legs for the 2pc build? If you're going to be swapping mastery into haste anyway it wouldn't matter to lose the mastery on hands/ shoulders for increased stats.

...

...

... . I think you might be on to something!

The traditional 2T16H has always been about maximizing mastery, which explains the decision of switching tier pieces with haste for off-set pieces with mastery. This also created the annoying issue of having too much hit from the hit/mastery shoulders, and potentially needing to use undesirable int gems.

There are no haste/mastery shoulders or haste/mastery gloves. What would be the third piece?

[Komma]

Went with Garry helm, SC gloves, IJ legs for testing. SC gloves seemed potentially better when you consider how shoulders has ridiculous amounts of haste - but that could also count against it, I suppose.

"Mage_Arcane_2T16_Custom" refers to the set with IJ legs.

Single Target:


Three Target:


Trustbucket, I think your comment might have just given a lot of high end Arcanists a bunch of new frustrations.

[Vatti]

I have the legs Will run it tomorrow.

Could you link an AMR profile with them in? Not sure what stat weights I would run.

[Komma]

http://www.askmrrobot.com/wow/gear/4f16 ... bd0e58dd14" target="_blank is what I tested with...I think.

[Vatti]

Thanks! Will post results tomorrow.

[Swordz]

Absolutely brilliant stuff Komma, thank you for all your time and dedication. Now I have some other pieces to look forward to during farm and test out such a build, which will make farm much more exciting