SimulationCraft for Warlocks (4.0.3a numbers at 80 and 85) - Page 74

Warlocomotif · 08/25/09, 8:05 PM

Originally Posted by dedmonwakeen

Interesting, so it seems their base crit values are designed to give them 5% crit when unbuffed and unmodified by talents.

That was my initial assumption, but I'm not entirely sure if thats accuratel I currently have 900 firebolts logged on the practice dammy (and will keep going for a bit), but the current crit% is at 12%, I'm specced 0/14/57 with 2 pieces of T8, 1 piece of T9, and no talents that enhance pet crit% other than fel vitality. Getting a bigger sample size still though- so give it some time before making any changes.

[edit] As the sample size increases the crit% is steadily dropping. 12% currently seems to have been a fluke, but will post with numbers later.

[edit 2] I've let the imp keep going, and the crit% steadily kept dropping. 12% definitely seems to have been a luke, and the model I posted above I currently think will prove to be accurate. After 2140 firebolts, the crit% had dropped to 7.8%, which means that in the segment that came after the first 900 firebolts, only 59 more crits happened (That's out of 1240 more firebolts, which is about 4.8%- which still isn't what we expect, but it's pretty close.

I'll try extend the sample size to 5000+ firebolts tomorrow, for now the server is down for maintainance.

Cvarto · 08/25/09, 8:25 PM

Is there some sort of Simcraft for dummies guide? I'm not exactly code savvy (ha… ha… far from it), and I rock with my Mac OSX. Any help on how to get Simcraft up and running would be greatly appreciated.

jcmaster · 08/25/09, 8:32 PM

Where does the dps fr a deep affliction warlock with a DG out stands in comparison to the other specs? From what I can see, no DGs were used for these simcrafts...

Cvarto · 08/25/09, 9:37 PM

Originally Posted by jcmaster

Where does the dps fr a deep affliction warlock with a DG out stands in comparison to the other specs? From what I can see, no DGs were used for these simcrafts...

DG is worse than the Imp for deep destro. Improved Imp, Demonic Power, and Empowered Imp make the Imp better for a deep destro lock.

jcmaster · 08/25/09, 9:44 PM

I was referring to a deep afliction build, not a deep destruk one. I was asking because all the theory crafting so far for deep afflic has been done with a succubbus as the pet and not the DG. Does the graphic in page 1 of this thread still holds with t9 gear?

edited; added t9 gear...

Cvarto · 08/25/09, 9:56 PM

Oh, whooops haha. The first iteration listed is for deep afflic with DG.

Warlocomotif · 08/25/09, 10:02 PM

[1] The doomguard is shown in the first post. 53_00_18_dg = doomguard.
[2] If you are intent on commenting on someone's gear, you're likely better off sending a private message
[3] How to use Simcraft has been asked many times before, I recommend you either search or read through this thread.
[4] The exact effect of the 2 piece T9 bonus is currently not really known for sure (Particularly for the melee pets), so until it is, the numbers would be very valueable anyway.

dedmonwakeen · 08/25/09, 10:30 PM

Pet crit rates fixed in r3177. I'll put out a release tonight.

EDIT: SimulationCraft r3178 available for download, includes Warlock pet fixes

Yeda · 08/26/09, 10:58 AM

In my opinion its still relevant to put 00/13/58 without LT glyph in the front post. The debate of the real value of LT glyph is still upthere, and many warlocks still think the "freedom" of not LTing at certain times its a DPS increase. In my last simulation it came behind 100dps more or less, so in a real raid situation things could change. Just my opinion, not asking to do the sim for the glyph/spec combo since i did that myself, just think that would contribute to have it in the front post to keep comparing both "versions" as time / gear sets evolve.

Cheers for the work and "welcome back" Zakalwe

Warlocomotif · 08/26/09, 12:15 PM

Originally Posted by Yeda

In my opinion its still relevant to put 00/13/58 without LT glyph in the front post. The debate of the real value of LT glyph is still upthere, and many warlocks still think the "freedom" of not LTing at certain times its a DPS increase. In my last simulation it came behind 100dps more or less, so in a real raid situation things could change. Just my opinion, not asking to do the sim for the glyph/spec combo since i did that myself, just think that would contribute to have it in the front post to keep comparing both "versions" as time / gear sets evolve.

Cheers for the work and "welcome back" Zakalwe

I generally agree with that you're saying, it would probably be interesting to see how glyph of Immo/Incin/Conflag would compete with Incin/Lifetap/Conflag after you enabled the movement settings of Simulationcraft.

duhwhat · 08/26/09, 1:17 PM

I think another relevant build/action priority list would be a Demonic Pact build that doesn't use Immolation, and thus going with the Immolate rather than Meta glyph. In some cases, Immolation can be dangerous, but maximum single target DPS is still required of the DP player (think Algalon), and it would be nice to be able to gauge performance relative to this kind of sim.

TheRabidSniper · 08/26/09, 2:56 PM

My Deep Destro Spec

Me: 27.58% crit (including Firestone enchant), 916 INT
My Imp: 739 INT (424 + 315, "Increases Spell Critical Hit by 5.34%")

Using a level 80 dummy in Darn, untapped:
1075 firebolts - 1020 non-crit, 55 crit - 5.1% crit

My Deep Demo Spec
- note: includes DT and ImpDT

Same gear; firestone still enabled. 37.78% crit chance, 916 INT
My imp: same as above
- So now, I'd assume the Imp has a crit chance of X+5(MD)+10(DT)+11.33(ImpDT - 0.3*37.78), where X may be just his 5.34% from INT, for a total of 31.67%

Same combat dummy, fewer just for resonable confirmation of influence of talents:
550 firebolts - 374 non-crit, 176 crit - 32.0% crit

Felguard: 90 AGI (5% melee crit), 487 INT (172 + 315, "...6.26%")
- interestingly, the FG gains more spell crit per point of INT

A few notes:
- The dummy I had chosen for the imp was a poor choice as it was right next to another dummy; meaning for each of the FG's Cleaves, he would hit at least two targets. However, I later found out that no dummy in Darn is far enough away from the others to stop multiple dummies being hit by one Cleave. Still, I needed to ensure he wasn't attacking two non-80 dummies (as that can have unwanted effects on crit chance), so I chose a different dummy for him to swing at, Cleaving into a second level 80 dummy.
- Multiple dummies being hit by one cleave doesn't seem to matter, as it turns out: as far as Recount cares, each target that is hit by a single Cleave attack counts as a separate Cleave attack.
- No matter how I am oriented to the dummy when I send the FG in, he always re-orients himself so that he's directly behind the dummy.

So, two level 80 dummies, for soaking hits and the extra Cleave:
Melee attack: 1693 attacks
- 1091 hits, 442 crits, 98 glancing, 42 dodges, 20 misses - 26.1% crit

Cleave attack: 1127 attacks
- 794 hits, 271 crits, 35 dodges, 15 parries, 12 misses - 24.0% crit

For the most part, they were within 1 or 2 percent of each other, whenever I checked them. The parries in the Cleave are only due to the second dummy that was slightly facing the FG, I believe.

Warlocomotif · 08/26/09, 6:58 PM

The sample size is still pretty small, but itis close enough and pretty much consistent between the various tests- to assume it's correct.

O got 2xT9 today, and out of curiousity due to this post:
Demonology Thread

I had a quick look to see how much my imp would crit (after all, auto attacks benefitting differed between the succubus and the felguard, so you really never know). after 1400 firebolts, I have a crit% of 15.3% recorded- which both, confirms that the bonus works for imps- and is close enough to the expected 5.63% to re-affirm that.

I'll also check to see if the succubus indeed doesn't benefit (and that the felguard does), but that will take me some more time.

Orgath · 08/27/09, 11:49 AM

There are various numbers floating around regarding 2pT9 Bonus so here is some math to clear things up a bit:

+10% pet crit =

pet base damage + pet crit rate +10 / pet base damage + pet crit rate => pet damage increase
assuming 21% base crit rate for raidbuffed destro imps (see simcraft for 13/58) = 131/121=108%
48% higher Empowered Imp uptime (31/21=1,48) from 11% (see simcraft for 13/58) -> 16,3%
simcraft already models crit rates around 55% for the affected spells, which brings the increase down to 2,4%crit increase mostly consumed by CB, Immo and IN due to mechanics.

Mystearica · 08/27/09, 6:12 PM

Originally Posted by Tenthplanet

You can probly use the dummy @ the Argent tournament grounds.

I'm still wondering if it will be a marginal boost or loss for going with the 4p t9 sets, has anybody been able to test this out yet? I only have 1 piece of t9 so far.

Well I'm no math expert or anything, I used the Optigear(manually updated the loot list) found on these forums & the current weight factors from simulationcraft to play around a bit.
Without putting a weight for the T9 bonuses(ie T9 2/4pc do nothing) We'll still be using at least 3pc of T9 (helm, shoulders, & legs).
So the question then becomes is it really worth not wearing a 4th piece of T9 over off set gloves & robes? I'm assuming the gain of wearing 4pc will out weigh wearing both off set robes & gloves.
The real problem with this debate tho, is that there's no ilv258 Helm or shoulders from ToC. So at ilv245 it may be better to only wear 2pc until you have access to hard mode loot. (which almost everyone should judging by the current difficulty of ToC)

XviiLock · 08/27/09, 8:05 PM

At what point is the 2 pc T7 bonus out done by raw stats?

mjball · 08/28/09, 12:04 AM

Originally Posted by XviiLock

At what point is the 2 pc T7 bonus out done by raw stats?

The short answer: very quickly. Also see Corruption Triggers Crit - Spell - World of Warcraft for comments about the proc.

The long answer:
Are you actually trying to figure out when you break your T7 2pc? If so, then you're wasting your time because T7 is such low-end gear these days. If not, why the academic exercise? At any rate, for those who can't remember it (like myself), [Valorous Plagueheart Circlet] is a reference to the bonus he's talking about. Particularly once you compare T7 gear and set bonuses to the rest of the warlock tier set bonuses and tier gear (T8/T9) in general, I really don't know why you're asking this question.

Edit: Okay, so I just actually bothered to look at your armory page. Not to be a dick or anything - but go get some real gear, proper gems, and come back with questions that aren't a waste of time.

Relent · 08/28/09, 1:11 AM

An idea popped into my head the other day: Why not create a program to write a simcraft profile text file for every possible set of spec, rotation, and glyphs and run them all through? I could easily arrange for such a thing to be written.

I think that this comprehensive approach to theorycrafting has the potential to greatly benefit the community. The kind of exhaustive, scientific analysis would include all possible creative efforts and analyze them all in one batch.

Does Simcraft have memory leaks or other performance issues which would prove to be obstacles in the achievement of such a computation?

Yeda · 08/28/09, 8:26 AM

I was running SimCraft today (version on SVN) for scale factors, and I noticed a strange thing. Spirit in 00/13/58 with LT glyph is at 1.09 DPS and at 00/13/58 without LT glyph is at 1.24. Dont know if its something with smooth_scale_factors, but im running a sim atm without it activated. Still weird how much spirit is valuated for a spec without the glyph.

dedmonwakeen · 08/28/09, 8:32 AM

Originally Posted by Relent

An idea popped into my head the other day: Why not create a program to write a simcraft profile text file for every possible set of spec, rotation, and glyphs and run them all through? I could easily arrange for such a thing to be written.

I think that this comprehensive approach to theorycrafting has the potential to greatly benefit the community. The kind of exhaustive, scientific analysis would include all possible creative efforts and analyze them all in one batch.

Does Simcraft have memory leaks or other performance issues which would prove to be obstacles in the achievement of such a computation?

The performance cost would be prohibitive. Consider the number of variables you are talking about. Just doing gear alone would be spectacularly expensive given the number of slots.

Even for just gear, you would be better off pre-calculating a 4-D surface: (SP,Crit,Haste)=>DPS

Find reasonable ranges for each of SP, Crit, Haste. This collapses the problem down to three variables with fine-tuned control on the range of the variables. Let's say you allow for 10 buckets, you end up with 10^3 combinations.

Now consider 17 slots of gear. Even if you limited the gear choice to 3 pieces (including gem configs!!!) you end up with 3^17 sim runs.

You are better off building a DPS surface lookup-table and then quickly iterating through gear choices to see where you end up on that table. Even still, you'll need something smarter than a simple recursive search....

This is just at the edge of possibility: You could run with smooth_rng=1 and iterations=100 to get a reasonable DPS value. Covering a 10x10x10 surface of such runs is the equivalent of a single 100k iteration run which is not really that painful on modern multi-threaded processors.

EDIT: The interesting thing is that the interpolation table only needs to be generated once per talent/glyph/patch combo and then can be re-used. Time for some late night shenanigans on those 128X big-iron servers......

Originally Posted by Yeda

I was running SimCraft today (version on SVN) for scale factors, and I noticed a strange thing. Spirit in 00/13/58 with LT glyph is at 1.09 DPS and at 00/13/58 without LT glyph is at 1.24. Dont know if its something with smooth_scale_factors, but im running a sim atm without it activated. Still weird how much spirit is valuated for a spec without the glyph.

I'm afraid I have currently lost my faith in smooth_rng when used to reduce both iteration count and the scale deltas (which is what happens with smooth_scale_factors=1). I would recommend against smooth_scale_factors=1 and instead iterate in the 10k to 20k range when generating scale factors. I'll eventually work on solving the problem because it still has the greatest performance potential of any reasonable alternative..... but my plate is full at the moment.

Warlocomotif · 08/28/09, 10:18 AM

This is why you can download Simcraft and run the scale factors based on your current armory. The sale factors in the OP should be taken as an indication, not as a "these are the numbers you should go by for your own gear".

angaroth · 08/28/09, 1:08 PM

So is there a self-crit to 2pcT9 conversion yet? I assume that the higher your own crit level is, the less 2pcT9 is worth and obviously it varies from spec to spec. My thought would be that the machine-gun imp would provide the smoothest uptime and thus gain the most benefit but perhaps the FG auto-attack occurs more often that firebolt.

Yeda · 08/28/09, 1:34 PM

Originally Posted by dedmonwakeen

I'm afraid I have currently lost my faith in smooth_rng when used to reduce both iteration count and the scale deltas (which is what happens with smooth_scale_factors=1). I would recommend against smooth_scale_factors=1 and instead iterate in the 10k to 20k range when generating scale factors. I'll eventually work on solving the problem because it still has the greatest performance potential of any reasonable alternative..... but my plate is full at the moment.

Yes, smooth_scale_factors was causing this problem. Even tou i still think some stats look a bit odd:
- spirit its looking a bit to much atm, but thats just an opinion;
- Intelect also used to be higher on non-LT glyphs specs, but atm its close to the same value;
- And just from top of my head, shouldnt haste be me in pair in both specs? Or even a bit higher on non-LT glyph spec (since it Life Taps less, therefor spending more time casting)?

                                Int        Spirit    Spower        Hit        Crit      Haste
Warlock_T8_00_13_58	        0.57	    1.28      1.56	  3.57        0.85	1.21
Warlock_T8_00_13_58_nolt	0.55	    0.96      1.59	  3.57	      0.85	1.12

Did 20k iteration like dedmonwakeen recomended

Naoe · 08/28/09, 2:04 PM

Originally Posted by madlaxer

I have always wondered how the coefficients for the various stats were determined. Clearly they are VERY important in the outcome of any tests or calculations when comparing different stats. They are different for each spec, so just using arbitrary numbers.... why is haste worth 1.275, and crit worth .8613, etc etc?

And i know it is much more complicated than just using constant numbers, but shouldn't those values be changing as your stats change? Not even talking about specific "caps" but just in general. Obviously once you are hit capped the value of hit is zero, and the value of haste goes down once your 1.5 casts hit the 1.0 sec ceiling. But more on a general level, all the stats that increase damage are related and become more and less valuable as each of them move up and down. I know that we are limited to the available loot, so we cant always pick the exact ratio of all stats. But if you take the coefficients as they are, you should cap your hit, and then ONLY take haste. Each point of haste is always better than a point of crit, and given the choice, you should never choose crit.

Your warlock with its current stats, among other things, has an average damage done per cast, an average cast time per cast, and total damage done. Referring to these as AVG Damage, AVG cast time, and total damage. You have the choice, to increase either haste or crit, according to the constant coefficients you should always choose haste. But is that always the right choice? Choosing haste will lower your AVG cast time, allowing for more casts in the same amount of time, increasing Total damage because you were able to cast 1 extra spell (more realistically we are talking about fractions of a spell). So lets just say your AVG Damage per cast is 1000. You increasing haste allowed you to cast 1 extra spell, and increased your total damage by 1000. The other option would be to increase crit, and effectively increase the AVG Damage per spell. I would not gain the 1000 from the extra spell, but every other spell would be doing more damage. Lets say that increasing crit would increase AVG Damage by 100, so now each spell does 1100. If i only cast 5 times, my increase is only 500, so yes, haste is still better. But if i have so much haste already, that i cast 15 times, the 100 increase per cast is worth 1500 total damage, greater than simply casting 1 more spell.

My point is, that as any of your stats get larger, it slowly starts to diminish the gains from that stat, until the others catch up some. This effect is completely lost when you lock in the values for all the stats.

This effect happens with hit also. It is universally proclaimed that you are a complete idiot if your not hit capped. The opinion of so many is that you know nothing, and your dps is completely irrelevant if your not hit capped. The same argument from above applies. Lets consider the hybrid gem, +10 to hit, and lets say you are 10 hit rating away from cap, do you use the hybrid gem, or forget about the last 10 hit and just go pure spell power? 10 hit is less than 1/2 a percent, so we are talking about over 200 spell casts per miss. If you add the hit, you will guarantee that you don't miss that 214th spell, and increase your damage by that spell. Or you could add a spell power gem, and make the first 213 spells do more damage. What if you are only 2 hit rating away from cap? Less than a .1% chance to miss. Meaning on avg you will cast over 1000 times per miss. Im sorry, but i would bet anything that increasing the spell power of those 1000+ casts is better than removing the .1% miss chance. And i understand that not all casts are created equal, take Haunt for example, it hurts your total damage more if a haunt is resisted rather than a CoA. This does not really change anything though, it only skews the value of hit slightly higher to compensate for the random number generation chance.

Sorry for the really long post.

You are very correct in the fact that the values of stats change depending on gear. Haste and crit increases the value of spirit and spellpower, spirit and spellpower increases the value of haste and crit. If you have extreme amounts of haste, but a small amount of crit, the relative value of crit goes up comparing to haste and vice versa.

The statvalue numbers are generated by simply simulating a fight several times and increasing the interesting stat by a small amount and see how big the difference is.

The reason for haste being theoretically higher dps than crit isnt the fact that it increases the number of casts rather than the damage of spells, but that it is cheaper. All combat ratings have the same cost for item budget, but different amounts are required to increase your haste/crit/hit by 1%. Hit requires a small amount and scales with everything which makes it THAT good. Haste is still pretty cheap and thus good, but crit is extremly expensive at almost 40 rating per percent. sure, missing 0.1% hit from cap only makes you miss once in every 2-3 fights or so, but when that miss comes it can be devastating, especially for affliction if haunt misses.

I hope this helped abit

prorsus · 08/29/09, 12:26 PM

Originally Posted by Yeda

I was running SimCraft today (version on SVN) for scale factors, and I noticed a strange thing. Spirit in 00/13/58 with LT glyph is at 1.09 DPS and at 00/13/58 without LT glyph is at 1.24. Dont know if its something with smooth_scale_factors, but im running a sim atm without it activated. Still weird how much spirit is valuated for a spec without the glyph.

It's possible the scale calculation step is too big. Because both spirit and int will change lifetap number in a short duration. For 5 minutes, it will be probably 7 or 8 taps without glyph. Due to the integer nature of life tap cast, the scaling for any stats which affect life taps will result an uneven effect on dps.

I would suggest use a much small step such scale_spirit=+30. Also try different base spirit and different durations.

Haste value is more complicated stasts. It not only changes lifetap frequency due to mana consumption, but also change possible casts between 2 conflagrate which will result a huge step in dps change. I think you probably should try both positive and negative on haste value.