Maybe the first post needs a "tools" section, which would mention Rawr vs Simulationcraft, and how people can stop questioning the latter every other day.

Speaking of wearing mage armor, did blue ever post anything regarding a new Mage Armor Glyph? As it stands with meditation and mage armor each being 50%, the glyph becomes absolutely worthless as exceeding 100% regen grants no additional effect.

Mage and molten armor were close for arcane before, and it was more of a flavor/playstyle decision, but with full mage armor arcane mages taking a 25% regen hit, Glyphed Molten (@6% crit with solid arcane gear) seems to be far and away the better armor with any reasonable values of spirit for any build.

Edit: Wanted to come back and provide the math behind a couple of these numbers when I had time.

The 25% regen hit is coming from the Spirit based regen cut of 40%. Currently with mage armor glyphed and AM, you've got 80% regen. 3.1 will be capped at 100% of the remaining 60% of spirit regen, for a net effect of 80% -> 60% normalized to 75% of 3.0 regen.

I've updated SimulationCraft Mage module with a zero execution time action "choose_rotation". This action examines regen rates, dps-vs-dpm consumption rates, and target "time to die". Based upon that data it will set the current rotation to "dps" or "dpm".

All standard mage spells now have two new conditionals "dps" and "dpm".

Consider the following action priority list:
Under these specifications, an AB 3-stack will only occur when in a "dps" rotation or when under Arcane Power.

This has allowed me to equip Molten Armor in all the Mage profiles, resulting in less than 0.4% "wait time" (ie: 4sec out of every 1000sec the Mage is standing there waiting for mana to regen). In addition, I also report the amount of time spent in "dps" mode vs "dpm" mode.

This has certainly (and dramatically) improved the DPS output of my Arcane Mage actors.

Unfortunately, despite being in a "dps" rotation upwards of 80% of the time, I'm still finding that Arcane/Frost still lags Arcane/Fire by ~6% in 3.1 mechanics (using 3.0.9 BIS). Is this too far off? Perhaps I need help choosing the "dps" vs "dpm" actions. Perhaps my itemization is not good enough. I don't know....

(It was actually closer until I realized that the Fire Mage was holding off Hot Streak procs while waiting for the Pyro DoT to clear.)

Regarding simcraft, is there a place where you calculate the additional spellpower from int for arcane mages? I'm not overly familiar with the code so I may have missed it, but if it's not being added you're short-changing arcane a couple hundred spellpower.

One improvement you could make to the gear set would be to move the dragons eye from the helm to the pants, and put a red gem in the helm (giving up the 2 mp5 socket bonus). The chest orange gem might also be better off as a spellpower gem since dropping the 6 hit socket bonus will still leave you above 17% hit after talents/buffs.

Additionally, Arcane missiles provides 5 stacks of illustration per cast, although the increase from modeling that is probably fairly minor.

edit: If I'm reading this correctly, it looks like arcane using the arcane barrage glyph over molten armor.

Rawr shows that under certain cooldowns (mostly AP+something) you should spam AB but use MBarr procs at 3 stacks, and then for the rest of the time mix AB3-AM and AB3-ABarr in a way that will use exactly all the remaining mana (though the rotations are very similar). Of course you'd have to determine based on fight duration under what cooldowns exactly to spam AB, which is why rawr handles arcane so much better. Anyway if you're not at least spamming AB during heavy cooldown stacking you'll probably get a decent DPS loss just from that. AB3-AM or even AB3-ABarr just doesn't use our mana fast enough. The reason for using MBarr at 3 stacks when AB spamming is less for the DPS (the difference is measly) - it's mostly for the massive mana savings. Spam AB mindlessly and you'll find yourself with a lot less mana to spam AB at other points in the fight.

Definitely using Molten Armor now......

I'll experiment more with your ideas. Thanks for the tips!

The code does stack AB-Buff and Arcane Power additively. For "convenience" they are separated in the code, but I'm actually adding a multiplication against the "base" cost in both places.

I don't believe the simulation should be in "dps" rotation 80% of the time. More likely the other way around. Rawr is suggesting a mana-saving rotation of AB3AM for 240 seconds of a 5 min fight. It only switches to high-dps ABspam3MBAM in conjunction with IV/AP/Gem cooldowns. Therefore, high-dps mode should only be of the order of 20% of the time. This is largely because high-dps mode is so ridiculously expensive in mana that there is little point gimping your dps just to add a few seconds more of high-dps mode.

For an example, Rawr suggests for my own mage the following rotations
AB3AM: 196.28 sec (5004.65 dps, 136.32 mps, 3017.14 tps)
ABSpam3MBAM: 0.68 sec (5418.16 dps, 361.59 mps, 3265.61 tps)
Icy Veins+AB3AM: 6.21 sec (5906.01 dps, 188.46 mps, 3560.53 tps)
Icy Veins+Arcane Power+ABSpam3MBAM: 3.79 sec (7658.45 dps, 489.18 mps, 4612.40 tps)
Icy Veins+Arcane Power+Mana Gem Effect+ABSpam3MBAM: 26.21 sec (8056.71 dps, 489.18 mps, 4851.36 tps)
Icy Veins+Mana Gem Effect+AB3AM: 3.79 sec (6222.47 dps, 188.46 mps, 3750.41 tps)
Arcane Power+ABSpam3MBAM: 30.00 sec (6501.79 dps, 392.16 mps, 3915.79 tps)
Mana Gem Effect+AB3AM: 15.00 sec (5272.81 dps, 136.32 mps, 3178.03 tps)

Some of those aren't really full rotations because cooldowns overlap, but I've bolded some of the mana per second consumed on the more important rotations. You can see from the first two lines that the high-dps rotation on it's own is only 8% more dps than the mana-saving rotation, yet AB3AM spends only 136 mps, while ABspam3MBAM spends 362 mps. So, according to Rawr, that high-dps rotation is only worth doing in conjuction with cooldowns, and specifically arcane power. You'll notice that it switches back to AB3AM when there's no arcane power, even if icy veins or mana gem effect are still up.

So now we've identified arcane power as the trigger for high-dps mode, what is really the difference in spell selection? Well both rotations cast 3 x AB, and the mana-saving mode clears that stack with AM regardless of missile barrage procs. So it's only in high-dps mode where Rawr suggests to wait for missile barrage, keep spamming AB until you get it. This is something for which your simulator is ideally suited.

Some pseudocode:
1. Cast AB x 3
2. If mana-saving mode or missile barrage proc? Cast AM.
3. Cast another AB, goto 2.

Of course, since you mentioned your simulator is having trouble with running out of mana, you can also add the ABAM rotation, which is almost mana-neutral (but very low dps - 4.1kdps on my mage). This could be used when mana is less than 2k and there is no available mana gem, potion or evocation.

One last thing. About cooldown stacking. I'm not sure how you model this, but since you've fixed the fight length at 5 mins, there is a simple heuristic. Stack IV+AP+Gem all together twice: once after 20 seconds, and once 20 seconds before the end. In between this, AP can be used twice on cooldown (total 4 times), and mana gem once on cooldown (total 3 times).

And before anyone points out that we could squeeze in another IV into 5 minutes: yes we could but it means that you lose most of the mana from the first mana gem - due to having to use it earlier. It likely means you'll have to do a longer evocate towards the end of the fight. Since evocate does 0 dps, any small gain from the extra IV is lost. As Vontre once said: "I removed the cooldown from evocate... dps went down."

Criticism welcome

I'm actually considering AB3AM a "high-dps" mode. Arcane Power is something of an exception. When under AP, it ignores the current dps/dpm rotation choice.

actions+=/mirror_image/arcane_power/icy_veins
actions+=/presence_of_mind,arcane_blast
actions+=/mana_gem
actions+=/evocation
actions+=/choose_rotation
actions+=/arcane_blast,max=2
actions+=/arcane_blast,max=3,arcane_power=1
actions+=/arcane_missiles,barrage=1
actions+=/arcane_blast,max=3,dps=1
actions+=/arcane_missiles
actions+=/mana_potion

We only reach an AB 3-stack if under AP or Rotation=DPS (bold lines).

To get AB spam until Barrage proc while under AP, we add the following bold line:

actions+=/mirror_image/arcane_power/icy_veins
actions+=/presence_of_mind,arcane_blast
actions+=/mana_gem
actions+=/evocation
actions+=/choose_rotation
actions+=/arcane_blast,max=2
actions+=/arcane_blast,max=3,arcane_power=1
actions+=/arcane_missiles,barrage=1
actions+=/arcane_blast,arcane_power=1
actions+=/arcane_blast,max=3,dps=1
actions+=/arcane_missiles
actions+=/mana_potion

However...... I find this to be a slight dps loss because I end up being in a "dpm" rotation (AB2AM) for a higher period of time.

What actually concerns me the most is that you are able to maintain AB3AM indefinitely in Rawr.Mage while I am forced into AB2AM for some portion of the fight. This leads me to believe that I am either over-consuming or under-recovering mana. Unfortunately, my primary platform is UNIX. I'll try to get some time on a windows box and see if Rawr.Mage provides some mana-related breakdowns that I can use.

Calculating your DPS based on an AI that has no clue what the fight length is, is both extremely complicated and pointless. It's pointless because you always at least have some idea on the fight duration. It's extremely complicated because you'd have to maximize the expectancy value of your DPS over the random fight duration which can be anywhere from 0 to infinity with equal chance (since you know nothing better about it) - this will actually probably give you the same rotations as you would get over an infinite duration fight anyway since the biggest contribution to the expectancy value comes from very high fight durations.*

If your AI doesn't know the fight duration and has more than 1 rotation to choose from, of course you'll run into problems, as like said above you need to maximize the expectancy value of your DPS over the random duration rather than your actual "DPS right now". Yet you still let fire use a rotation with living bomb, even though over a very very long fight living bomb is not a good idea, and if you know nothing of the fight length the main contribution to the expectancy value comes from very long fights.*

In reality, you either know exactly the fight duration (thus do the rawr calculation), or have a good estimation for it. While RAWR doesn't handle random durations (yet, at least), you can still input various durations and see how they effect your rotations and optimize accordingly, for example:
- High chance for 5 min fight and low chance for 6 min fight, but can change rotations in a way that will greatly increase your 6 min DPS and hardly affect your 5 min DPS, then you should use the 6 min rotations.
- High chance for 5 min fight and low chance for 6 min fight, and can change rotations in a way that will reduce your 5 min DPS by X and increase your 6 min DPS by X - use 5 min rotations.
- etc...

Basically simulationcraft is assuming simple AI, and the AI needed to play arcane properly just isn't simple. You could code AI based on rawr results but then why not just use rawr. If we didn't have arcane power the AI would be much more simple but still require knowing the fight duration. Note that if you can actually could let rawr play your character for you, it could re-calculate optimal rotations based on changes in fight duration based on increased/reduced raid dps from what was expected and based on the new mana conditions created due to spending more/less mana than expected. I wish we could make a practical implementation of in-game rawr that will actually tell you what to do in real time

*The reason for this is that expectancy value for DPS over a fight of totally unknown duration to be only dependant on infnite-long-fight-rotations:
(1) Average DPS over given T duration = (1/T) * integral(from 0 to T) dt*DPS(t)
(2) Average DPS over varying fight duration = lim A->infinity [ (1/A) integral(from 0 to A) dT AverageDPS(T) ]
Where AverageDPS(T) goes by formula (1) and DPS(t) is the function that will maximize (2). Let's say there's a finite fight duration Thigh above which the optimal rotations are cyclic and just repeat themselves. Since the integral goes up to infinity you will have a finite portion that goes from 0 to Thigh and then an infinite portion that goes from Thigh to infinity, and since the result is of the finite portion is also finite and is divided by infinity, it's contribution is 0, and your expectancy value for your DPS over a random duration fight is then only based on the optimal rotations for an infinitely long fight. This is obviously not realistic and therefore at least some information of the fight duration must be provided to get optimal DPS rotations.

Question !

Hey everyone I have a question , I'm an Arcane mage and I just broke 5k dps on patchwork and I feel that's ok. But I'm wondering if my gems should be moved around +16 haste to 9 Spell power + 8 haste or keep it like I have it? Also, should I keep on the t 7.5 legs or the ones that drop from malygos?

Here is the link to my armory The World of Warcraft Armory

Any help would be appreciated =).

- Joobieo

You're losing damage if you're going from T7.5 to leggings of the wanton spellcaster because you're breaking your 4pT7 bonus.