It's related to Winter's Chill, and it appears to be something on the order of +1% damage per stack of Winter's Chill on the target. None of the Frostbolts I cast while WC wasn't stacked exceeded the expected value while naked. None of the Frostbolts I cast while unspecced and geared exceeded the expected value using the current coefficient.

So it could just be a bug, or it could be an ill-conceived buff (since Winter's Chill will be overwritten by any other spell crit debuff).

I believe someone else noticed this a few weeks ago related to Improved Scorch, affecting all casters.

All three types of the 5% crit debuff increase the spell damage taken by the target by 5% as well in the current ptr build. We have not received any official response about it, nor have they changed the tooltips.

One thing worth noting: SimulationCraft uses three different lag values: queued, instant-cast, channeled (listed in increasing value)

This tends to penalize off-rotation instants that are weaved into your action sequence.

Yes, one of the things I was going to look into was if in fact Lhivera was using the default values or just zeros. I have a combat log analyzer that looks for frostbolt-frostbolt and frostbolt-fireball-frostbolt timings and collects statistics. If you keep your haste constant, you can gather information on the time cost of weaving a fireball (or ice lance) in between two cast time spells.

The results were that on a live realm with about 80 ms latency, the instants were actually saving a little bit of time and on the PTR (with over 300 ms latency), the extra time cost for instants was equal to that for timed spells. Both tests were on target dummies, so a real raid situation might be different. Zaldinar had some plans to write an in-game analyzer for these timings (you can compare the expected spell timings with the actual time cost even when your haste changes - unlike with the log analyzer).

Here's a snippet from a spreasheet comparing calculated timing with measured timing on a live realm with 80 ms latency:

Chained frostbolts were taking an extra 23 milliseconds. Frostbolt+fireball combos were taking 47 mlliseconds less than predicted. Since this was a human player pressing the keys (me): YMMV.

The thread is on the Subcreation mage forum: Instant Cast: Benefit or Penalty?

The last two posts show my current information on the topic...the first ones probably had data distorted by inaccurate analysis of the combat logs (I wasn't actualy detecting the fireballs - I was just binning frostbolt to frostbolt times into two bins and probably had that slightly wrong as well).

Even with an average of 150 ms added per each brain freeze fireball, I think it should still show as a very minor DPS increase. It gains about 1% when you set the lag penalty for all spells to zero and goes up another notch if you use the timing data I posted. Zaldinar has a pretty decent explanation for the effect, but I think it's in another thread.

I am using the default values. I haven't changed anything from default except the talent points and casting priorities necessary to produce the different specs and casting strategies.

Changing the lag values will affect pretty much all casters. Hot streak pyroblasts and refreshing living bomb will also gain in DPS, if the lag values are lowered. The program I used could be easily modified to measure hot streak pyroblast timing instead of brain freeze fireballs. It's unfortunately worthless on actual raid combat logs, because haste is so variable in raids and combat logs do not record haste.

Has anybody experimented with an Incanter's Absorption+Ice Barrier build on fights with a lot of Raid damage?

For example:

The World of Warcraft Armory

I am just wondering if the buff from Incanter's Absorption plus the casting of Ice Barrier + Frost/Fire Ward is enough to buff dps to the point where this spec would be comparable with other "better" frost specs? I haven't played frost for years, but with the massive spellpower boost I have seen through Incanter's Absorption, I was wondering if the buff could offset some of the other loss of dps.

Of course, this build would have one purpose: surviving on fights with a lot of raid damage and utilizing Incanter's Absorption fairly often.

Assuming Glyph of Ice Barrier and a +spell power of 2000, the amount absorbed per casting is = 1.3*(3300+2000*.8067) = 6387. 15% * 6387 = 958 spell damage per cast for 10 seconds. Since you can cast it every 26 seconds, this amounts to an average 958*10/26 = + 368 spell power. Adding in the fights when you can cast fireward or frost ward, it would seem to be a decent bit of +spell power, not to mention the survivability.

Any thoughts?

This has been fixed in the latest PTR build.

Exactly what school our you favouring with that build - Arcane or Frost because you never made it entirely clear? (I'm assuming Frost being the frost thread but a lot of dps Frost talents are skipped)

It's it's Frost to many Frost buffs are lost and increased spellpower isn't enough to pull it up to acceptable dps, especially at the cost of a global cooldown every 26 seconds. If it's Arcane then you've skipped too many key talents including Missile Barrage and increased spell critical strike damage.

In short hybrid builds just don't work due to the nature of Mages firing nukes from a particular school.

I'd agree with Pasture - Loosing things like the WE, FoF and Chilled to the Bone completely will really gimp your DPS. I'd guess about 25% lower than the standard 18/0/53 builds. There just isn't much from going deeper into the Arcane side that makes up for that.

I've got a question about Frost utility on the Vezax fight.

Does Life Leach only heal Vezax if it does damage to a player? If so, wouldn't having Ice Barrier up and absorbing a tick significantly reduce the healing Vezax gets?

Fantastic, thank you Tiga.

Incanter's doesn't even add that much to arcane with a disc priest keeping it up. There just isn't enough raid damage output to maximize incanter's gain (~1158 sp on a fully raid buffed arcane mage in perfect incanter's gear.) It is roughly a 200 dps gain with any realistic amount of raid damage taken, possibly 400-600 dps on the most extreme raid damage fight. If there were a way to hit the max SP gain with reasonable uptime, Arcane would be in a very nice position right now.

It's definitely not worth it for frost due to the afforementioned issue, but not only that frost would have to sacrifice far too much dps, and ice barrier would not be an optimal way to trigger incanter's procs.

I played a bit with SimulationCraft last night. I found a bug in the haste< option handling. I wasn't able to fix it myself, but Dedmonwakeen was very quick to react, so the latest version on the SVN is working correctly.

After some experimentation, the part that determines brain freeze use is like this:

actions+=/cold_snap
actions+=/fire_ball,brain_freeze=1,ghost_charge=1
actions+=/fire_ball,brain_freeze=1,haste<=0.9
actions+=/frost_bolt

What it does is to use brain freezes on ghost charges no matter what and then use them normally if there's less than 90% combined haste (yes, even when your GCD is over the haste cap for it). The differences are quite minor, but they are there. The exact haste value actually depends on the raid setup, the gear and especially the gcd_lag option value. I mostly simulated with an ideal raid setup and the default gear & talents.

The gcd_lag value is probably more important than most people understand. I posted an article on brain freeze scaling yesterday on the MMO Champion forum (before I did all the simcraft stuff) and I'll paste it at the end of this article. It repeats some of the things I have said before on this forum, but I think it's a pretty decent summary on how brain freeze DPS scales as a talent.

Going back to SimulationCraft though, gcd_lag obviously doesn't affect the frostbolt spam mage very much, because that mage is only summoning the water elemental and mirror images on the GCD. The effect on the brain freeze-using mage is much greater. Here's some data running 4 simulated mages in an ideal raid and then all on their own without decent raid buffs:

NoBrainer = SOTM maxed
BrainFreezer = 3/3 spell impact, 1/3 SOTM
NoImpact = 3/3 SOTM, 0 impact
NoStudent = 3/3 spell impact, 0 SOTM

optimal raid:

gcd_lag = 0
6313 25.1% BrainFreezer
6308 25.1% NoImpact
6297 25.0% NoStudent
6238 24.8% NoBrainer

gcd_lag = 0.15
6254 25.0% BrainFreezer
6250 25.0% NoImpact
6244 25.0% NoStudent
6229 24.9% NoBrainer

gcd_lag = 0.3
6216 25.1% NoBrainer
6200 25.0% BrainFreezer
6194 25.0% NoImpact
6190 25.0% NoStudent

non-optimal raid:

gcd_lag = 0
4404 25.2% BrainFreezer
4402 25.1% NoStudent
4397 25.1% NoImpact
4302 24.6% NoBrainer

gcd_lag = 0.15
4363 25.1% BrainFreezer
4361 25.1% NoImpact
4357 25.1% NoStudent
4292 24.7% NoBrainer

gcd_lag = 0.3
4321 25.1% BrainFreezer
4311 25.0% NoImpact
4309 25.0% NoStudent
4278 24.8% NoBrainer

BTW, the difference between 0 and 150 ms gcd_lag is about 200 DPS for a fire mage. They are much more affected by GCD time cost than frost mages.


Here's the MMO Champion mage forum article from yesterday:


- Fireballs and frostbolts scale almost identically with spell power. Adding more spell power doesn't affect the value of the brain freeze talent by any significant amount. If you look at non-crits, the frostbolt:fireball ratio goes from 97% to about 102% over the whole raiding spell power spectrum.

- Frostbolt scales much better with crit than fireball, so if you gear up crit-heavy and have all the possible raid buffs increasing crit, you will reduce the relative DPS value of instant cast fireballs. Looking at the frostbolt:fireball scaling (at about 3200 spell power), we go from 120% to 130% when going form 40% crit chance to 70% crit chance.

- In a purely theoretical environment, as long as your GCD isn't capped to 1 second, the cast time ration between frostbolt and fireball remains absolutely constant. For patch 3.2, that's 2.3:1.5, which is a bit over 1.5. Combining spell power, crit and haste scaling, you still end up finding that the fireballs will do more DPS than your frostbolts - no matter what your gear is or the raid setup is. For typical frost gear in a typical raid, the DPS from an instant cast fireball is maybe 20-30% higher than for frostbolts.


Of course there's a catch...WoW isn't a purely theoretical environment. We're talking about a client-server combination connected with a network connection with non-trivial amount of latency. How do you model that? We have surprpisingly little data on how latency affects instant casts. We do know that the roughly 200 ms "cast window" allows timed casts to be chained almost perfectly, so spamming frostbolts, there's only a few milliseconds of extra lag per cast.

Instant casts are tricky, because there's no time window to pre-buffer a timed cast after an instant cast and the GCD actually ticks on the client side. But... it apparently starts ticking when you queue up the instant cast, which you can do about 200 ms before the timed cast before it ends.

SimulationCraft uses a random penalty on instant casts. The default value averages 150 milliseconds, which is pretty significant. Assuming the player has 25% haste from gear and raid buffs, you end up with (2.3/1.25)1.5/1.25+0.15) as the cast time ratio for frostbolt:fireball. Our nice 1.5 ratio now drops down to 1.36 instead. Add more haste (such as during heroism or from trinkets) and the fireballs become DPS-neutral at best.


So the big question is: what's the timing behavior of frostbolt+instant cast? I wrote a combat log analyzer program to look into that and at least playing normally on a target dummy, there's no evidence of a time penalty for frostbolt+fireball combos. In fact, the combos take up less slightly time than calculated. It probably has to do with the GCD actually starting early on the client side and allowing the next frostbolt to start casting slightly early. Even on a target dummy (no lag and excellent frame rates), I'm only gaining about 50-100 ms, but it's a gain rather than a loss. Plugging a 50 ms gain into the cast time ratio: (2.3/1.25)1.5/1.25-0.05) = 1.6.

If there's a time penalty on instant casts, the ratio gets worse as you increase haste. If there's a time benefit for instant casts, the ratio becomes better for fireballs as you increase haste.

So, depending on how you simulate instant cast spells can have a roughly 17% effect on brain freeze theorycrafted value.

Tiga, the method we used to pick our default GCD lag as to simply spam an instant-cast spell until OOM and then look at start and end times to determine expected-casts vs actual-casts.

Could you try that and see what numbers you get?

I have no doubt that 150ms is a valid result for spamming instant casts (let's say ice lance), but the situation changes when you do combos instead.

The scenario for a combo cast + another cast time might go like this:

(Let's assume a clean 2 second cast time and 1.2 s GCD with 100 ms latency.)
0000: Player schedules a cast
0100: Server receives cast
1800: Player schedules the instant cast, GCD starts counting
1900: Server accepts the instant cast (200 ms until cast completes, but it's OK)
2100: First spell + instant execute
3000: Client side GCD has cleared
3000: Player schedules another spell
3100: Server receives cast

If you look at the time from the first schedule to the last one, it's exactly 3 seconds. However, the combined time of the cast + instant is 3.2 seconds. Because the GCD ticks on the client side, we saved 200 milliseconds. Of course that's with perfect timing and assuming your latency is invariable etc...

Anyway, I think that's the theory. It relies on GCD counting purely on client side, but if that's not the case, maybe there's some other effect in play?

Here's a snippet from a real combat log (22.75% haste on a live realm):

30.886 SPELL_CAST_START,...,"Tiga",0x511,...,nil,...,42842,"Frostbolt",0x10
32.930 SPELL_CAST_START,...,"Tiga",0x511,...,nil,...,42833,"Fireball",0x4
33.846 SPELL_CAST_START,...,"Tiga",0x511,...,nil,...,42842,"Frostbolt",0x10

That's 2.96 seconds for the combo for cast starts. Here's the damage for the same three spells:

33.967 SPELL_DAMAGE,...,"Tiga",...,"...Dummy",...,"Frostbolt"
34.129 SPELL_DAMAGE,...,"Tiga",...,"...Dummy",...,"Fireball"
37.006 SPELL_DAMAGE,...,"Tiga",...,"...Dummy",...,"Frostbolt"

And that's 3.039 seconds for the combo. Theorycrafted time would be 3.259 seconds (4s/1.2275).

While I did very well on the example above (I picked it at random), running maybe 20 minutes worth of combat log through an analyzer said that on average, I was taking 23 ms extra for chaining cast time spells and saving 47 ms for each combo.

Here's the analyzer: http://wow.poista.net/theorycraft/bolttime.lua

There are short combat logs from a 3.2 PTR and from a live realm in the same directory, if you want to have a peek.

http://wow.poista.net/theorycraft/LiveBolts-80ms-T.txt
http://wow.poista.net/theorycraft/PTRBolts-300ms-T.txt

On the PTR, I wasn't getting time savings on combos, but the combos were only taking a little bit of extra time. Considering the PTR is across the Atlantic and i'm not used to playing with high latency like that, it wasn't half bad. There's probably a huge "human factor" in these timings and of course your connection quality might also be a factor.

P.S. After a quick ice lance test (57 ice lances) with 22.75% haste, I get:

56 1.3371 Tiga

The expected value is 1.222s, so I'm about 115ms slow. Matches quite well what you use in SimulationCraft.

I think it would be good to check if the data is valid for a typical raid as well. Real time changes in haste rating make it a bit more difficult though. I could maybe measure the chain cast time of a spell and assume that it behaves close to the theoretical model and then apply the haste value from that to combos close to it. Even then, the data might be tainted by movement periods. Zaldinar has suggested that an addon could be used to do this properly.

A raid environment may also impact this behavior in other ways -- greatly increased client/server communication could have an effect on lag, changing haste rates could increase error rates, etc. I don't really know what a good way to extract useful data from an entire raid encounter would be, but I suspect that the results may be significantly different from the target-dummy tests.

ETA: I'm running simcraft with the latest updates; I'll update both the tables, and the "compilation" on my site later today. dedmonwakeen has added a comparison feature that will make the job much easier.

ETA again: Real life issues have delayed my update. Hopefully tomorrow; this weekend at the latest.

Another edit: Sorry for the delay. Double-checking some of the config info on the files before rerunning again -- I believe earlier numbers had the Frost Mages using an extra profession, producing higher than expected results. The newest files seem to have corrected the problem.

Looking at the options available for buffing frost, most of them revolve around either buffing Ice Lance or somehow getting Shatter Combos to work. Deep Freeze has been pretty much ignored - because it was deliberately changed into a PvP-only spell.

As it stands, Deep Freeze does not affect bosses at all, but what would be the effect if it just caused them to be considered "frozen" for the duration of the debuff, instead of "stunned+frozen" as currently works? Would 5 seconds (less a GCD for casting it) of controllable frozen status every 30 seconds help Frost DPS any, or would the damage-less GCD from casting Deep Freeze outweigh the bonus damage from the frozen target?

That's about 2 FrB more with 50% increased critchance less one which has normal crit chance. I'd say it's marginal. Furthermore, it's completely useless if FoF procs anyway. Letting the debuff increase one's frostbolt damage by 50% for 10s on targets of higher level (some kind of AP) could give Frost a nice burst and maybe close the gap.

OK, here's the latest output, at long last. I've made a couple changes to simplify the output a bit:

First, I'm now using almost entirely default .simcraft config files. The only exception is the 18/0/52+1 build. However, that too is autogenerated from a wowhead profile now. All of the output files, config files, and the script I use to run them, can be viewed at Index of /misc/simcraft

Second, I dropped the 7.5 and 15 minute fights, and added 6.5 and 8 minute fights (so we have 5, 6.5, 8, and 10). This is largely because after 8 minutes is when Brain Freeze's efficiency starts to matter.

Second, because the performance difference between perfect ghost charge usage and 25% ghost charge usage was so small (generally on the order of 10 DPS or so), and because the difference between a spec with 3/3 Spell Impact and 1/3 Student of the Mind vs a spec with 1/3 Spell Impact and 3/3 Student of the Mind was similarly miniscule, I've consolidated everything down to two specs:

- 18/0/53 has Brain Freeze, 3/3 Spell Impact, and 1/3 Student of the Mind, and uses Ghost Charges with a 50% success rate. This spec has Ice Barrier, but only 30 yards range.

- 18/0/52 does not have Brain Freeze, has 0/3 Spell Impact and 3/3 Student of the Mind (the filler point is spent in Magic Absorption), and spams Frostbolt. This spec has Ice Barrier, 1/2 Magic Absorption, 36 yards range, and one extra point to spend.

The advantages of one spec over the other are trivial up until fights exceed 8 minutes in length. So, some conclusions:

1) The two specs will put up essentially identical performance on the majority of encounters. If you're really, really good at using Brain Freeze procs on Ghost Charges, then 18/0/53 is a bit better for you; if you're really bad at it, 18/0/52 is a bit better for you. If you're mediocre at it or don't know what Ghost Charges are, then use either spec, it doesn't really matter.

2) The tipping point for Brain Freeze is 8 minutes. If you're going to be fighting longer than 8 minutes without downtime, you need Brain Freeze.

3) For 18/0/53, Spell Impact and Student of the Mind are interchangeable. If you want to put a point or two into Magic Absorption, you can, with < 1% reduction in DPS.

4) DPS gap is larger than previous runs indicated, as some problems with Fire modeling have been addressed. The difference is now about 21% between Frost and Fire on short, stationary fights, and larger as fights get more mobile. If Frost DPS were increased by about 15%, it would still be 5% behind Fire on short, stationary encounters, and around 10% behind on mobile encounters.


5-minute stationary fight

Spec	DPS	% below Fire	% below top Frost
20/51/0	7828	n/a	n/a
18/0/52	6457	21.23%	n/a
18/0/53	6430	21.74%	0.42%

5-minute fight, 3 sec movement per 30 secs

Spec	DPS	% below Fire	% below top Frost
20/51/0	7338	n/a	n/a
18/0/53	5730	28.06%	n/a
18/0/52	5671	29.40%	1.04%

6.5-minute stationary fight

Spec	DPS	% below Fire	% below top Frost
20/51/0	7787	n/a	n/a
18/0/53	6389	21.88%	n/a
18/0/52	6360	22.44%	0.46%

6.5-minute fight, 3 sec movement per 30 secs

Spec	DPS	% below Fire	% below top Frost
20/51/0	7297	n/a	n/a
18/0/53	5648	29.20%	n/a
18/0/52	5576	30.86%	1.29%

8-minute stationary fight

Spec	DPS	% below Fire	% below top Frost
20/51/0	7461	n/a	n/a
18/0/52	6267	19.05%	n/a
18/0/53	6262	19.15%	0.08%

8-minute fight, 3 sec movement per 30 secs

Spec	DPS	% below Fire	% below top Frost
20/51/0	7017	n/a	n/a
18/0/53	5559	26.23%	n/a
18/0/52	5477	28.12%	1.50%

10-minute stationary fight

Spec	DPS	% below Fire	% below top Frost
20/51/0	7298	n/a	n/a
18/0/53	6199	17.73%	n/a
18/0/52	5772	26.44%	7.40%

10-minute fight, 3 sec movement per 30 secs

Spec	DPS	% below Fire	% below top Frost
20/51/0	6839	n/a	n/a
18/0/53	5454	25.39%	n/a
18/0/52	5073	34.81%	7.51%

Some points in favor of brain freeze:

- If frequent there's casting pushback, instants are immune to pushback.
- If there are AOE phases, being 9% more mana efficient on single targets will allow you to spend more mana on AOE.
- The Simulationcraft numbers are based on a perfect raid. If the raid composition isn't perfect, a spec with brain freeze will suffer less. This is especially true in terms of critical strike chance -> brain freeze scales relatively poorly with crit.
- The exact lag behavior of cast combos is not perfectly known for laggy raid situations, but for non-laggy situations, there seems to be a timing advantage (I assume the old 150 ms default instant cast lag was used) for Lhivera's numbers above? (However, fire will gain far more DPS if/when a lower gcd_lag is used.)