What causes the differences in frostbolt crit chance between the simulations? It would seem to me that the frostbolt spam spec should have the highest frostbolt crit chance of all (thanks to SOTM), but it comes second.
What percentage of ice lances need to fail to hit a frozen target before they actually become a DPS loss? (This is why I don't rely on the ice lance glyph.)
All I can think is "RNG," but I ran 20,000 iterations, so it seems like that should smooth out the bumps pretty well.
As for the Ice Lances, really really rough guess is that if you miss more than about 1 in 9, it's a loss. Somewhere between 1 in 8 and 1 in 9, the damage per execute time drops below Frostbolt.
At Veridian Dynamics, we can even make radishes so spicy, people can't eat them. But we're not, because people can't eat them.
I think that unless we can logically explain the differences in crit, the results should be considered suspect to some degree?
In the actual game with a real player (let's say me), there's some loss of FoF because of human reaction time and travel time effects (FoF procs on impact if you don't have frostbite, so it shows up rather late). That, however doesn't explain how the simulation could have a higher frostbolt crit chance in the BF+IL case vs. the pure frostbolt+SOTM case.
If I remember correctly, using /petattack on your frostbolt spam macro can cause your pet to reset his cast everytime you hit it. I usually keep that command to the summon, freeze macro only.
As a side note: I usually use my cold snap 20 seconds after proc'ing both my water ele and icy veins, in order to use icy veins back to back, giving me 40 seconds of haste. In a 3.5 minute fight, icy veins would have been activated 3 times.
The first one summons the water elemental, or if it exists, puts it on defensive and does a /petattack. I hit this key twice when I summon the elemental.
The second macro does a /petpassive and /petfollow, allowing me to call off the elemental and making it ignore all incoming damage sources. This is extremely useful in PvP and somewhat useful in PvE.
I have been considering an addon that would show a warning message each time the elemental changes targets. This would be useful in PvP, where a rogue vanishing will cause the elemental to switch to a healing priest, which might be my CC target.
In a simple fight, the right time to use cold snap is right after you have activated IV and the elemental, allowing the cold snap cooldown to start counting down as early as possible.
I spent some time reading Simulationcraft source code late last night, but I haven't figured out what would cause the frostbolt crit chance discrepancy in Lhivera's simulation runs. Using a higher crit chance on my own simulator shows that increasing crit chance significantly will also somewhat reduce the relative benefit of brain freeze. The crit chance for frostbolts didn't change depending on wether brain freeze was used or not.
I have been considering an addon that would show a warning message each time the elemental changes targets. This would be useful in PvP, where a rogue vanishing will cause the elemental to switch to a healing priest, which might be my CC target.
I have actually thought about this same thing after having my ele break my sheeps in 2s against rogue teams. I tend to be bad at watching my pet target box when I'm watching everything else. Anyone feel like writing a simple addon? If not, I'll start on it but it'll take some time with my schedule lately.
I think that unless we can logically explain the differences in crit, the results should be considered suspect to some degree?
In the actual game with a real player (let's say me), there's some loss of FoF because of human reaction time and travel time effects (FoF procs on impact if you don't have frostbite, so it shows up rather late). That, however doesn't explain how the simulation could have a higher frostbolt crit chance in the BF+IL case vs. the pure frostbolt+SOTM case.
Lhiv, if you can pile your Mage configs into one file and post them on that site, I'll dig into the results and provide a detailed explanation.
I do not believe that reissuing /petattack will reset your elemental's water bolt cast time, but reissuing /petpassive will. I want my elemental to never be on defensive (that's where most pvp misbehavior comes from), but using /petpassive followed by /petattack will interrupt the water bolts if you spam it like a bad player. Since I'm a bad player, I've been instead using
/petpassive [target=pet,nocasting]
/petattack
which can very occasionally interrupt a legit water bolt thanks to latency, but only early in the cast. As long as it's not baked into your frost bolt spam macro it should work fine.
Originally Posted by Kyth
The only true error is in not learning how to make your second kill better.
If I remember correctly, using /petattack on your frostbolt spam macro can cause your pet to reset his cast everytime you hit it. I usually keep that command to the summon, freeze macro only.
I don't know if it still does, but it definitely used to. If necessary, you can prevent that problem like so:
If the Water Elemental is already attacking something, the /petattack command won't be issued. It basically means, "start attacking my target if you're not already doing something else."
It occurs to me now that /petattack [target=pettarget, noexists] might accomplish the same thing, but I've never tried it.
Originally Posted by dedmonwakeen
Lhiv, if you can pile your Mage configs into one file and post them on that site, I'll dig into the results and provide a detailed explanation.
I found the reason for the extra crit with ice lance: Simulationcraft is allowing all frost spells to proc fingers of frost, when it should only be proccing on spells with a chill effect:
static void trigger_fingers_of_frost( spell_t* s )
{
if ( s -> school != SCHOOL_FROST &&
s -> school != SCHOOL_FROSTFIRE ) return;
This means that the ghost shatters with ice lance are adding more fingers of frost procs. We'll see frostbolt crit rates and FoF proc counts going down after this is fixed.
I'm testing a change to fix that (it also applies to Brain Freeze), and dedmonwakeen knows about it (he'll probably have it fixed before I've tested my version and sent him a patch).
He's also looking to add spell flight time, which can help with some human reaction time simulation. I believe Fireball, Frostbolt, and Frostfire Bolt all have the same flight speed as Shadow Bolt. In fact, I'm pretty sure every bolt has one of two flight speeds, with most using the same speed as those four spells, and Ice Lance and perhaps a couple other exceptions using a faster speed. Does anyone know for sure?
7338 Fire (for point of reference)
6226/64.4% crit: 18/0/53 w/spell impact, using brain freeze and using ice lance on ghost charges, w/Ice Lance glyph
6192/64.9% crit: 19/0/52 no spell impact, no brain freeze, no ice lance, w/Water Elemental glyph
6140/61.7% crit: 18/0/53 w/spell impact, no brain freeze, using ice lance on ghost charges, w/Ice Lance glyph
6104/60.1% crit: 18/0/53 w/spell impact, using brain freeze, no ice lance, w/Water Elemental glyph
It seems to me that the crit rates on the three 18/0/53 specs should be a lot more similar. It's possible I did something wrong with my change to the Brain Freeze and FoF proc code.
Last edited by Lhivera : 07/15/09 at 12:36 AM.
At Veridian Dynamics, we can even make radishes so spicy, people can't eat them. But we're not, because people can't eat them.
Assuming I read the source code correctly, Simulationcraft doesn't stop using fireballs during extreme haste periods. The T8 test mage has a +432 haste trinket, IV and heroism. I guess even base gear haste+IV+heroism will cap the GCD and going further than that will make brain freezes and ice lances a DPS loss during that period.
Thanks Lhivera for running these, it's a very interesting set of runs. I am trying to get a handle on these results, with a view to reproducing them in my own model.
A couple of questions, looking at the summary graphs. Why is Fingers of Frost active for such a high proportion of the fight? In the "19/00/52 FrB [BF]" example it is up for 38%, is this not a little high? With a 15% proc chance lasting 2 spells it should be ~28% on average - especially in the absence of any ghost charge utilization.
The "18/00/53 FrB [BF] [FoF3 IL]" and the "18/00/53 FrB [FoF3 IL]" seem to be missing Focus Magic feedback.
Is Heroism up for these runs?
Not trying to be picky, I'm just very interested in seeing how this pans out. I realize it's work in progress, very much appreciate the effort.
In modelling this myself, I can see how deceptively complex the rotation can actually be. There are a surprising number of permutations, with unintuitive, but mostly marginal, DPS adjustments...
FrB = Frostbolt, [BF] = On Brain Freeze proc, Fireball, [FoF2 IL] = On FoF proc, second charge, Icelance, [FoF3 ..] = Ghost charge etc
I found it disturbingly easy to bugger up the rotation when trying to automate it within the model - consecutive or near consecutive FoF procs can have bizarre results, double ghost charges etc, which are obviously not possible. Add to this the decision about any GCD limited Instant cast in hyper-hasted situations.
Edit: some more observations/questions as I find/think of them. I am having trouble tuning my model to reach these DPS levels. Possible discrepencies in the Simcraft model:
1. Scale of Fates is active ~100 seconds, from about 3 activations. Shouldn't it be ~60secs? (3x duration 20secs, CD 2 minutes) Or is this the inactive time between activations? I see that the numbers for procs make sense as far as Internal Cooldowns go.
2. Is it stacking Cooldowns? SoF on IV, Elemental up.
It's possible that other frost school spells are still proccing FoF in the simulator (IV, water elemental, cold snap) and that the uptime calculation isn't what we expect.
For pure frostbolt spam in an infinitely long fight, the chance of FoF being up for any given frostbolt is 100% - 85%^2 = 27.75%. However, if you use evocation while FoF is up, it doesn't affect your frostbolt crit chance in any way, but it increases the relative uptime of FoF. The same goes for summoning the water elemental and using mirror images while FoF is up (one GCD).
The uptime calculation for FoF uses:
p -> uptimes_fingers_of_frost -> update( p -> _buffs.fingers_of_frost != 0 );
On ghost charges, the value is -1, so while casting a frostbolt on a ghost charge will not actually benefit from FoF, the cast time is included in the uptime of the buff.
This would mean that on pure frostbolt spam in an infinite fight, the uptime would approach 100% - 85%^3 = 38.5875%. The chance of FoF being up for the first frostbolt is zero and then goes up from there. In a limited duration fight 38.1% sounds reasonable using the faulty uptime calculation method.
Right, see if I've got you. The FoF uptime is the FoF state for cast start, not release, in effect, Uptime = casts started with FoF active/ total casts . It is not counting those ghost charge Frostbolts as Shattering though. (?)
The way it is set up (at least on the source I grabbed from the SVN two days ago), is that FoF can be -1, 0, 1 or 2. If it's 0, it's not active. If it's positive, it's available for all spells. If it's -1, it activates shatter only on instant cast spells.
The damage done should be correct, but it looks like the logged FoF uptime includes frostbolts that can not benefit from ghost charges.
The aura loss is performed in the same function that then later on checks for a new proc (mage_spell_t::execute) that causes the aura gain, so wether this happens at cast time or at impact doesn't affect the uptime calculation.
If travel times were used, the uptime would drop much below the value I predicted above. This is because FoF (without frostbite) procs on impact, so you already have a new frostbolt casting when the proc happens and in general (without counting ghost charge time), you end up with maybe 1.5 frostbolts worth of uptime per proc even though two full frostbolts benefit from each proc.
To be meaningful as a statistic, I think the uptime should include the cast time budgets for each spell that benefited. That's the full cast time for frostbolt and the GCD+instant cast lag for fireballs/ice lances. The uptime accumulation could be moved to the shatter code, where it would add the cast time or GCD to a running total.
I bet the non-programmers reading this are about to dump cores...
The simcraft mage model has gotten more peer review over the last 7 days then it has in the last 7 months.....
I want to make sure I get the FoF model right. SimulationCraft supports travel time and human-reaction time for buffs. I just need a little more info.
(1) Travel speed for Frost Bolt and Frostfire Bolt. For now I'll just reuse our values for Incinerate: 21mps (Actual value used is a gaussian distribution centered on 21.)
(2) I understand that FoF procs on-hit, but what about consuming the buff? Is the "ghost charge" due to the buff being consumed on-hit, allowing one to combo a cast-time-plus-instant?
The damage done should be correct, but it looks like the logged FoF uptime includes frostbolts that can not benefit from ghost charges.
Based on your description of the problem above, I'm assuming uptime is related not to the number of spells it affects as much as the simple question of how much time the buff appears in your buff list. And this strikes me as the "correct" way to do it, given the "time" in "uptime." It's perhaps a more appropriate measure for time-based things like the Water Elemental than it is for charge-based buffs, but this is probably a case where it's close enough for government work.
Originally Posted by dedmonwakeen
I want to make sure I get the FoF model right. SimulationCraft supports travel time and human-reaction time for buffs. I just need a little more info.
(...)
(2) I understand that FoF procs on-hit, but what about consuming the buff? Is the "ghost charge" due to the buff being consumed on-hit, allowing one to combo a cast-time-plus-instant?
Ah, would that it were. That would be working like a "real" Shatter Combo. But no, that is not the case. FoF charges are consumed on cast. The "Ghost Charge" is a very small window, based at least somewhat on your latency, during which it is possible to cast an instant spell before the server realizes that the second charge has been consumed. It is a very ugly mechanic, and my personal preference would be for it to go away. But people do it, and the question of whether it's actually worth doing is dependent on how reliably it can be executed (one failure in ~8 attempts makes it a DPS loss).
Last edited by Lhivera : 07/15/09 at 12:12 PM.
At Veridian Dynamics, we can even make radishes so spicy, people can't eat them. But we're not, because people can't eat them.
- You can "buffer" one spell about 200 ms before a cast time spell ends (as we all know), so you cast the fireball at that point.
- This will cause the server to schedule that spell for casting as soon as the previous spell finishes casting.
- However, the server doesn't execute in a single thread and the casts are split among multiple threads.
- Variables like "FoF charges" are not really variables that you can atomically read/write any time, so what you end up is with two parallel threads with a "1" start value for FoF, which will then both issue a "decrement" or "set to 0" request. Both spells end up seeing FoF as being up and the FoF counter is likely to end up at -1.
I could be wrong on the details above - I'm just guessing what the server processing is like based on observations.
In order to pull of a ghost shatter without failure:
- You have to queue the fireball or ice lance early enough so that it is buffered on the server before the frostbolt finishes casting. This requires low latency variation. Latency could be caused by your Internet connection or server load.
- The server load must also be low enough so that both events are processed before the decrement operation from the first one to execute (probably the frostbolt?) consumes the FoF buff.
It works pretty much 100% for me under normal load, but I'm not willing to trust the mechanism at prime raid time in a 25 man raid. I will use ice lances on ghost charges when I absolutely must move. I will also use brain freeze fireballs on ghost charges, because the penalty for not shattering those is low and I would want to cast them at that point in any case.
Got a note from dedmonwakeen today, saying he's implemented:
- FoF proc on impact (instead of cast)
- FoF uptime calculated as a percentage of times a spell capable of a critical strike benefits from FoF
- haste<=0.xx option on all actions, so I can use ",haste<=0.50" on Brain Freeze Fireballs.
So, I made a few changes to the spell priorities (you'll be able to see them in the source links in the output). The result tells me there's probably something wrong, but I don't have the time to investigate deeply today; perhaps somebody else can. Basically, the Frostbolt crit rates skyrocketed. The new results are here: Simulationcraft Results
The summary is:
7338 Fire (for point of reference)
6536/73.2% crit: 18/0/53 w/spell impact, using brain freeze and using ice lance on ghost charges, w/Ice Lance glyph
6524/74.5% crit: 19/0/52 no spell impact, no brain freeze, no ice lance, w/Water Elemental glyph
6472/71.6% crit: 18/0/53 w/spell impact, no brain freeze, using ice lance on ghost charges, w/Ice Lance glyph
6406/70.6% crit: 18/0/53 w/spell impact, using brain freeze, no ice lance, w/Water Elemental glyph
You'll note that this puts the top two Frost results within 12.5% of Fire. But the vast increase in frostbolt crit rate over the previous results seems highly suspicious.
ETA: One thing that's clear is that the top two contenders are either using the Ice Lance glyph and using Ghost Charges on both Lances and BFF's, or maxing SotM and using pure Frostbolt spam. Attempts to mix the two approaches fall behind. So I'm eliminating those other two 18/0/53 options form my list for the sake of faster runs.
I have modified the spell priority from the default files as follows:
1) Frostbolt if FoF is up
2) Fireball if Brain Freeze is up and haste < 50%. I believe this should catch the FoF ghost charge if available.
3) Ice Lance if FoF is up and haste < 50%.
4) Water Elemental (this used to be behind Mirror Image)
5) Icy Veins
6) Mirror Image (this used to be above both Water Elemental and Icy Veins)
7) Cold Snap
8) Frostbolt
The results I posted earlier this morning had the Fireball priority change (it used to be lower on the list, I forget exactly where), but not the cooldown priority changes. I'm doing another run now.
Last edited by Lhivera : 07/16/09 at 1:30 PM.
At Veridian Dynamics, we can even make radishes so spicy, people can't eat them. But we're not, because people can't eat them.
- FoF uptime calculated as a percentage of times a spell capable of a critical strike benefits from FoF
I'm guessing the sudden addition of 10% extra crit is coming from this new FoF uptime calculation. What's most odd is that the rotation which doesn't use ghost charges has the highest crit rate. 74.5% seems absurd, but even your earlier analysis showing 64.9% is very high. Try removing FoF talent and see what it says.
For comparison, in my gear selection which has like 40% crit, latest Rawr says my average frostbolt crit rate should be 57%, and if I remove FoF talent then it's 43%. These numbers seem more reasonable.
I'm guessing the sudden addition of 10% extra crit is coming from this new FoF uptime calculation. What's most odd is that the rotation which doesn't use ghost charges has the highest crit rate. 74.5% seems absurd, but even your earlier analysis showing 64.9% is very high. Try removing FoF talent and see what it says.
For comparison, in my gear selection which has like 40% crit, latest Rawr says my average frostbolt crit rate should be 57%, and if I remove FoF talent then it's 43%. These numbers seem more reasonable.
The rotation that's not using ghost charges has a higher Frostbolt crit rate because it replaces spell impact with Student of the Mind. That's not a surprise, really. What is odd is that the three 18/0/53 builds should have the same crit rate.
I'll try a run without FoF after this and see if it's adding more than around 14% crit.
ETA: Shuffling the cooldown priorities had no effect on DPS.
ETA again: Run without FoF complete. This resulted in crit rates of 50.9% (18/0/53 w/ghost charge activity) and 51.7% (18/0/50+3 w/frostbolt spam). So FoF is adding 22.3% and 22.8% crit, respectively, around 8.5% more than it should. Best guess is that it's getting an extra Frostbolt in on each FoF proc for some reason.
Last edited by Lhivera : 07/16/09 at 3:17 PM.
At Veridian Dynamics, we can even make radishes so spicy, people can't eat them. But we're not, because people can't eat them.
For pure frostbolts spam, the expected increase of crit from the FoF talent is (100%-85%^2)*50% = 13.875%, which agrees very closely with what RAWR was telling Wizeowel. The crit chances on these Simulationcraft runs are much higher than what I typically get in a raid, but that's partly due to circumstancial differences.
It should be easy to calculate a predicted crit rate for frostbolts spam:
Adding up the above gives me 55.3% for PTR talents and 58.3% for live talents. Maybe I'm missing something? 74.5% seems way too high.
BTW, 74.5% crit chance implies an over 60% chance to crit even without FoF, which would actually put the gear/spec over the soft cap for frost crit. If you are critting over 64% of the time as frost, you are likely to be over the soft cap (reducing the value of crit rating even further). (Being over the soft cap means that your crit chance is higher than 100% when FoF is active, so it's capped at 100%.)
If you want to approach testing the simulator systematically, it makes sense to test the really simple case first and check that it matches your expectations and real game performance. If we determine that the simulator is wrong (and we know by how much), we can start looking for a reason.
====
More on brain freeze, but unrelated to simulations (just simple calculations on a spreadsheet):
I did some spreadsheets today, calculating the spell damage scaling of non-crit frostbolt vs. non-crit fireballs. It's interesting how close these spells scale in a typical frost raid build. Fireballs start out hitting slightly harder than frostbolts (about 50) and end up hitting for slightly less. The conclusion from that is that increased spell power does not alter the DPS value of the brain freeze talent. The scaling is too similar for there to be any noticeable difference when you increase/decrease spell power within a reasonable range.
Now of course the difference really comes from critical strike scaling, since fireballs do not benefit from ice shards. The value of the talent is higher when your critical strike chance is lower. So, for a non-ideal raid (say, a 10-man) and with non-ideal gear, brain freeze is a better talent than for a top of the line raid in top notch gear. If we find that the simulator is calculating crit chance incorrectly, the value of BF will change once again...
In my own simulations and in actual raiding, I'm finding that for a BF-specced frost mage, the crit chance of frostbolt and fireball are very close when using ghost charge fireballs + non-FoF fireballs.
This one presumably goes away in patch=3.2.0 mode (which is what I've been running with), being replaced by the -0.2 sec cast time.
We probably need to examine the gear set. A PvE Frost Mage doesn't want to exceed 50% crit before Fingers of Frost is active. Isn't there a tool that calculates BIS gear for a specific spec? We may well find that the gear list in the Frost config files isn't ideal.
Also, is crit suppression simulated? I forget what the value is, but hasn't it been determined that a +3 level boss reduces crit chance by some amount?
Lhiv, I'm not sure why you are seeing such a large drop when FoF is removed....... unless the "without" ran a simple profile without the FM Circle Jerk.
I used exactly the same files, with only a change to the talent links (I shifted points from Fingers of Frost into Frozen Core).
Last edited by Lhivera : 07/16/09 at 5:26 PM.
At Veridian Dynamics, we can even make radishes so spicy, people can't eat them. But we're not, because people can't eat them.
I got the selfish benefit from winter's chill (3%) and empowered frostbolt (4%) mixed up a bit in my post. I took the amount of spirit from the Simulationcraft log, but it wasn't 100% clear if it was with kings & everything. 10% from molten armor sounds like a lot to me though. And I forgot focus magic...
Let's try that again (I'll keep 10% for molten armor for now):
Gear crit (18.47%) + molten armor (10%) + crit debuff (5%) + moonkin aura (5%) + totem of wrath (3%) + winter's selfish chill (3%), empowered frostbolt (4% or 0 for PTR) + FoF (13.875%), focus magic (3% or 6%)
Adds up to about 68% (or 64% on the PTR), which is a lot closer to 74.5%, but not there yet. It's interesting that the 3.2 number sits exactly at the soft cap for crit. The T9 set bonuses would wreck that though (amazingly bad bonuses for frost mages).
I generally don't recommend that other mages buff me with focus magic, because there's usually someone else in the raid that benefits more from it (a FFB mage, destro warlock, shadow priest or moonkin). For maximum DPS though, I guess you can assume it's there.
07/14/09, 12:30 AM
If not, I'll start on it but it'll take some time with my schedule lately.
Similar Threads
