Sorry that this is a bit off topic, but is Hydross the only boss in SSC that is immune to bleeds? I'm having an argument with someone in my guild, and he's saying that rupture isn't very good because all the bosses in SSC are immune, which I don't believe.

Yes, only hydross in ssc

Hydross is the only boss that is immune, but also the Elementals at Vashj are immune to both bleed and poison.

As far as I can remember everything else is susceptible to bleed and poison

I could have sworn I'd heard that the crit scope affects melee as well. Not that I've tested it myself since I have Serpentshrine Shuriken...

--edit--

Sorry, double post

At the risk of shameless self promotion, here is a cross link to a post I have on the Rogue DPS Spreadsheet thread discussing the modelling of Deadly Poison between the two sheets and how it might be improved. I basically concluded that the DPS sheet underestimates poison damage and the Gear sheet overestimates it (due to not accounting for stack-up time after a fall off). I came close to putting together a better formula to use, but I got stuck converting the probability of DP falling off to an average time before it does.

Take a look if you want some additional ideas on DP modeling. Unfortunately, not being a statistician I can't really vouch for the mathematical correctness of my approach, but I think the concept of what I'm getting at is valid.

I know that ^^ But as I'm Engineer, I also show some interest on which scope I build is most efficient für our hunters.


No, it's stated on the scope itself that it only improves "the weapon it's used on".

So, the thing is, I did actually try to account for stack-up time when I wrote the model in the Rogue Gear sheet. Looking back, though, it does appear that I may have bungled it. So, in the interest of professional curiosity, lets work over deadly poison modeling for a moment.

Let n = the number of attacks that hit in the average second made with weapons that have deadly poison on them. That is, if you're using Deadly Poison on OH, this is the number of OH attacks per second times the chance that an OH attack hits.

Let m be the probability that Deadly Poison procs on a given hit.

Then p = .83 * m * n is the probability that Deadly Poison is applied in a given second. Let q = 1-p.

It is easy to see that the uptime of Deadly Poison - that is, the fraction of the time when you have at least one proc up - is 1-q^12. But how do we account for ramp-up time?

Well, the immediate idea which occurs to me is that we can estimate the duration of the average Deadly Poison gap, and use that to estimate how frequently such gaps occur.

Let us assume that 12 seconds have elapsed since our last proc, so Deadly Poison has just ticked off. At this point, Deadly Poison starts up again with probability p each second thereafter; that is, with probability p it occurs in the first second, with probability p*q in the second, and so on. Thus, the expected time until the stack starts again is the sum of the infinite series p + 2pq + 3pq^2 + .... which is easily shown to be equal to 1/p. Hence, the average downtime between the end of the last proc and the beginning of the next is 1/p, and the uptime is 1-q^12; hence, to determine X, the average duration of one "cycle" - that is, one uptime period followed by one downtime period - we simply solve q^12 = (1/p)/X, or X = 1/(pq^12)

Next question: how long does it take poison to ramp up? Well, by the above, it's an average of 1/p seconds between procs, so it takes an average of 5/p seconds to stack back up. Thus, out of our cycle of length 1/(pq^12), we spend 1/p time with no poison up, 1/p with one stack on, 1/p with 2 stacks on, and so on. Thus, the total damage *lost* relative to continuous uptime is (5+4+3+2+1) * 1/p in each period of length 1/pq^12. Thus, the average number of stacks up is given by 5 - (15/p)/(1/pq^12) = 5 - 15q^12 = 5 * (1 - 3q^12). Now, in the Rogue Gear Spreadsheet, I have 5 - 3q^12, so it appears that the bug in the gear sheet is that I neglected to distribute the 5 correctly when I entered it into the sheet. I'll make a point to fix that in the next release.

Note that this still isn't a 100% accurate model, since it implicitly assumes that poison always stacks up to 5 before dropping off - which isn't true. This is why, for instance, if you plug p = 0 into the above formula, you find that you average -10 stacks of poison up, which clearly isn't true. But for reasonably high poison application rates - that is, where 5/p is small compared to 1/pq^12, it should be reasonably accurate. But when implementing this in, for instance, a spreadsheet, you probably want to perform a check to verify that the poison application rate is high enough to get good results.

As an additional note: this provides a lower bound for poison uptime; we obtain an easy upper bound via the simple uptime calculation (ignoring ramp-up time). So it is valid to say that N, the average number of stacks up, satisfies 5*(1-3q^12) <= N <= 5*(1-q^12)

Edit: Oh, it also doesn't factor in that uptime periods are at least 12 seconds in length, which skews the probability distributions some. Still, it's not a bad estimate when p is on the order of .2 - which it usually is.

First and foremost I wanna thanks those before me for the massive work done on our theorycrafting. The post above that explores DP on the spreadsheets while a great read should not be attempted before at least two cups of coffee, as now my head hurts. I've been lurking for months here and, so thanks for all the late night reading and topics to ponder in afternoon traffic.

I don't know everything about everything nor recall (or have likely not read) every single post here. In instances where the killing is fast, be it simple mobs or overload DPS, I use IP both MH and OH. Like running Kara with the group I was recently running with, it made no sense to try to build my 5 stack and then sustain that for a certain time to make DP an effective/superior substitute for IP.

I like numbers, am not NEAR the math head that some of the big guns are here. Is there a simple formula or model to determine approximate length where DP comes into it's own? I am thinking that I can probably extrapolate it from the aforementioned poison post, but I've yet finished my first cup of coffee.

If it is simple and I just can't spy it, forgive the silly questions and thanks again for making my rogue an effin' blast to tinker with!

I'm pretty sure I followed all that, with the exception of the simplifications of the series (which I am totally willing to trust you on). Where I got hung up was the 3*q^12, which you are saying should be 15*q^12. When it was just the 3*q^12, it basically looked like the assumption was that the poison was dealing 3/5 less DPS for the times it was down, since downtime is given by 1*q^12. Following your math, it makes more sense now.

It certainly seems a good way to model it without being too overly complicated. The factor of 5 difference in the "average stacks" formula will reduce poison DPS by a good bit, which makes sense intuitively to me.

I was getting hung up on the 12 second thing just as you are... since the periods have to be at least 12 seconds long I didn't know how to approach it. However, as long as the average cycle length 1/(pq^12) is fairly long, then few enough "dropoffs" will fall way low on the probability curve and it looks like a good model.

I like it.

Now, models are great, but in engineering we always put real numbers back in to the model to verify that we get results that make sense. So, let's do that for a typical rogue. For example, me (as I'm mid-geared). I have, unbuffed, +18.33% hit from gear, +5% hit from talents, and +10 expertise (-2.5% dodge) from talents. Thus my hit table for my offhand looks like:

Miss + Dodge = (6.3 + 28) - (18.33 + 5 + 2.5) = 8.47%
Hit + Crit = 91.53%

Offhand weapon speed is 1.5, increased by Haste to 1.47, increased by SnD to 1.13.

DP Proc Chance = 30% (no talents in improved)

p = (1 - 0.17) * (0.3) * (0.9153/1.13) ~= 0.202

This is in the range where we thought the prediction should be valid. Good.

q = 1-p = 798

Average downtime = 1/p = 4.95 sec
Average uptime = 1/(pq^12) = 74.23 sec
Cycle time = 4.95 + 74.23 = 79.18 sec ~= 1 min 19 sec

Hmm, average uptime seems a bit low. Does DP really fall off so much? That surprises me.

Average stacks = 5 - 15*q^12 = 4.00

This seems to be a reasonable result. I can try to check my poison DPS on some fights to see if I'm hitting more or less 4 stacks on average. Should help verify the model. Others could try their own numbers and compare to WWS or w/e as well.

Interesting; I'm only getting an average of 4 stacks on the target due to DP falling off. This begs the question, how does it compare to IP?

DP -> 4.00 stacks * 180/12 = 60 DPS
IP -> ((146 + 194)/2) * (1 - 0.17) * (0.2) * (0.9153/1.13) = 22.85 DPS

DP is still clearly better.

What about Improved Poisons vs. Vile Poisons? Well, with 5/5 Vile you see a 20% increase. With 5/5 Improved?

p = (1 - 0.17) * (0.4) * (0.9153/1.13) ~= 0.269
q = 0.731
Avg. stacks = 5 - 15*q^12 = 4.65

4.65 / 4.00 ~= 1.163, or a 16.3% DPS increase. (Also, it gives a smoother uptime and faster ramp-ups after stopped DPS.)

So, comparable, with Vile being a little better for static fights and improved possibly being better for dynamic ones.

Yes, calculating DP damage is extremely hard. There is actually very little in terms of simple models that you can use beyond just running a simulation. The problems are due to the damage being dealt every three seconds and not caring about what happens other times, a varying number of attacks each three seconds, and another very interesting feature for which I have to make a few assumptions clear before stating:

1) Three seconds after the first application, and every three seconds there after, damage will be done based on the number of doses on at that time.
2) The stack will fall off if it does damage 4 times without being refreshed in the meantime.

These are based on observations I made as a rogue where a stack will fall off before the Blizzard UI says it should, and always right as it ticks with less than 3 seconds left. This means that the probability of the stack falling off is greater when it is refreshed closer to the time it ticks. Given that we're looking to get a good grasp at this probability, this feature puts a whole new twist into it. The idea though is to hope that things average out in the long run and probabilities calculated will tend towards the actual events in the limiting case. The question just becomes how to model it effectively, as you can't just look a the probability of it not being refreshed in 12 seconds.

I'm just starting a class in stochastic processes, where stochastic is a fancy word meaning "random". I anticipate this class will introduce tools for modeling behaviors very similar to DP. However, I would have thought that given its a low-level grad course, that *someone* on these boards would have already taken courses in this sort of thing.

The first topic discussed in the class is something called Markov chains, which I first encountered in middle school when looking at probabilities of landing on spaces on a Monopoly board. They model stochastic processes where the probability of transitions from one state to another are solely determined by the current state. Under this assumption, it is very easy to determine long term behavior of the system and relative time spent in each state. Consider the following states:

1: No DP
2: 1 DP dose, has ticked 0 times since last refresh.
3: 1 DP dose, has ticked 1 time.
4: 1 DP dose, has ticked 2 times.
5: 1 DP dose, has ticked 3 times.
6-9: 2 DP dose, has ticked 0-3 times.
10-13: 3, etc
14-17: 4, etc
18-21: 5, etc

Recall that if it ticks 4 times without being refreshed it will fall off regardless if that's 9.1s or 11.9s since last application.

Every 3 seconds the state changes (or in the case of states 1 and 18, maybe not). State 1 goes to state 2 with probability p, where p is the chance for exactly one poison proc, it goes to state 6 with chance q, where q is the chance for exactly 2 procs, and so on. States 1, 5, 9, 13, 17, and 21 go to State 1 if no poison procs occur, while in any other state it jumps to the next higher numbered state in that case.

If we're truly modeling long-term behavior we can average out SnD downtime/BF/Drums/etc into weapon speed and thus poison procs so that the probabilities are constant. One problem with this model is that in reality when DP falls off the 3-second timer resets when DP is reapplied (by assumption - correct me if I'm wrong), while this model assumes the state changes every 3 seconds. There might be a way around this, but I haven't thought hard about it yet.

One can form a transition matrix where the entries correspond to the probabilities of going from one state to the next, and the long term behavior can be found by taking the matrix and multiplying it by itself until the entries either become stable or change periodically. I'm assuming that the former will happen in this case. This is a rather naive approach to using the transition matrix to determine long term behavior, and there may be better ways of solving them. However, it is something that can be programmed for a computer to do without thinking and thus possibly calculate the behavior based on changing variables as spreadsheets do currently.

It looks to be quite a daunting task to figure out each of the necessary probabilities in general form and I'm not all that swift with spreadsheets. I'll try to think about the situation more and chime in when I figure stuff out.

In the interest of checking the accuracy of all this, I wrote a little program to simulate what we're modeling - that is, that every 1 second there is a chance p that poison is applied, which stacks to 5 and refreshes, and then totals up the average poison level across the duration of the test - specifically, the following Python program:

I ran 2 trials of 100 million seconds of simulated poison application each; the average number of stacks applied was 4.0847 in the first test and 4.0858 for the second. So, suffice it to say, the average number of stacks for your number is about 4.085 in our model, versus 4.00 predicted by my estimate. So, we're about 2% low, which is probably adequate for the moment.

Then, of course, there's the issue that our model doesn't exactly reflect reality; in reality, the chance to apply does not occur once a second, but once per attack (which can be either over or under once per second, can change over time, etc.), and the fact that Deadly Poison doesn't tick 12 times spaced one second apart, but 4 times spaced 3 seconds apart. On the whole, though, these strike me as unlikely to significantly change the overall answer, so I'm content with this estimate for spreadsheet purposes.

Oh, and re: Glowacks. You're right, we are making many simplifying assumptions in these models which may or may not affect the accuracy of the estimates; however, solving the problem analytically is prohibitively hard. One could solve this average stacks problem reasonably easy if one could diagonalize the 15x15 transition matrix; but computing Eigenvalues of a 15x15 matrix is pretty hard, and, in particular, does not have any closed-form solution. So the trick is not so much knowing the math to write down the exact formula for the answer, but finding a reasonably good estimate that can be included in spreadsheets for purposes of damage estimation. I make no claims that our simplifying assumptions will be exact - but as long as they're close, that's enough.

I know the Hemo vs. Combat inflection point thread was closed, but I read the 60+ pages and never had the chance to ask this question before the thread was closed, I hope asking it in this thread is appropriate.

I'd like to spec something like Talent Calculator - World of Warcraft just for a change of pace, I've been combat for as long as I've raided and the idea of playing something different seems refreshing.

From the old thread it seemed as if the consensus was that the core DPS talents in sub are Serrated Blade, Hemorrhage, Dirty Deeds and deadliness with the lower tiers being mostly filler. I'll be taking some points in deadliness since the best MH is have is the Big Bad Wolf's Paw. The latest spreadsheet from http://rogue.raidcal.com with the 2.3.3 changes has 11/20/30 slightly higher than 11/21/29 in my gear, but I was wondering if anyone had practical experience from using blade flurry over that last point in deadliness? My intuition tells me that the point in blade flurry will serve me better but it's hard to argue with the spreadsheet.

It's a shame that the old thread devolved into what it did, it was a good read while it was civil.

The spreadsheet assumes you are attacking a single target constantly. Blade Flurry certainly has more uses that just boss fights, and any time you hit more than your target, its value goes up a LOT. I would definitely pick up Blade flurry over 2% more AP.

From the first page of this thread:

One thing you gotta keep in mind is that the spreadsheet models indefinite DPS and puts a static value on each talent. It can't model you Blade Flurrying 2 mobs at once nearly as well as it can model what an additional 2% AP will do for you.

EDIT: Damn you Latito...

Regarding your spec, why 5/5 Opportunity? The additional Garrote damage it provides is pennies, really. Elusiveness, Inititative or Setup would be much better filler talents I would think. But if you're going for 100% raid DPS, then just ignore me.

Thanks to both Ozzmar and Latito for the quick response! Sorry about missing the bit at the front of this thread, that'll teach me to read all of the other one and not reread the first page of this one post patch

Since I took prep to get down to deadliness I figured a 3.5 cooldown on a Vanish which is reset by Prep is kind of overkill and it seems like the old thread indicated that you couldn't really count on Setup for reliable combo points even with Evasion/CloS since it's only a 45% chance. My reasoning was that some degree of predictably increased damage was better from filler than maybe getting a combo point.

First-time poster, long-time reader; love your site.


My current spec is 20/41/0, swords, and has been for some time now. I am completely satisfied with my current damage output.

However, I am inquisitive, and after looking at the patch notes for 2.3.2 I thought I would play around with some of the numbers. I posted my results and asked for constructive feedback on the WoW class forums, but received no good discussion from it.

Here is a link to the post:
WoW Forums -> 20/0/41

To sum it up, I wanted to ask you guys to give me some figures as to why applying Shadowstep to a 5cp Rupture 19-20 times over the course of a 10+ minute boss fight (combined with the other raiding benefits of the Subtlety tree with a 20/0/41 build) does not make up for the loss of energy regen and attack speed bonuses found in the combat tree.

I appreciate your input, and look forward to reading what you come up with.

My apologies if this was posted earlier, I looked but was unable to locate any related topic.

So, short answer:

A fully talented, mangled rupture does on the order of 2500 damage. Hence, each Shadowstep adds about 500 damage. However, each Shadowstep also costs 10 energy, which is 10 energy no longer being spent on Hemo; assuming a typical hemo hits for around 600 (which is probably actually low), the energy loss also costs you about 150 damage. So the next damage increase of Shadowstep is on the order of 350 damage every 30 seconds - or about 12 DPS.

Now, compare to, say, a single point in dual-wield spec. If you have a 90 DPS OH and 2100 AP (pretty easy for any rogue in a 25 man raid), each point in Dual Wield spec is worth about 18 DPS. So the total contribution of Shadowstep is less than a single point in DW Spec.

Fundamentally: sure, there are some good DPS talents in subtlety. Hemo is a great talent, and Deadliness, Sinister Calling, and Serrated Blades aren't bad. The problem is: there are even more good talents in combat, and the combined damage contribution of Improved Sinister Strike, Precision, Imp SnD, DW Spec, Sword Spec, Weapon Expertise, Blade Flurry, Aggression, Combat Potency, Surprise Attacks, Vitality, and Adrenaline Rush turns out to just be a lot bigger than the damage you can get from the Subtlety talents.

Well put, and exactly what I was looking for. Thank you both.

So on the poisons, would it be safe to state the following?

1. Assuming infinite time assuming a completely stationary target with a 1 second weapon, Vile is slightly better.
2. For a varied fight, Improved Poison is better.

And if that's not always true, is there a percentage of needed DPS time on the boss to keep Vile better than Instant?

Hello, my char. name is Gull.
I have been having problems with my DPS. In Kara my guild just faught Aran and my DPS was only 491. That was with popping my trinkets using blade fury and using the SnD/Rupture combo. I have great gear and great enchants, so my question is does anyone have any advice to help me get my DPS where it needs to be? I'm on WoW armory too if you wanted to look at my gear and talent build......Thank you for any advice

Gull

Your gear looks generally okay, although you could use a new chest at some point. There's certainly better stuff than what you're using, but as I don't know what else you have, it's hard to say whether what you're wearing is right or wrong.

Talentwise, the most immediate issue I see is that you don't have Improved Sinister Strike. You should drop 2 points from Lightning Reflexes and get it, as it's one of the best DPS talents there is for a sword rogue. You're also missing Aggression in combat, which is a definite boost to DPS. Also, it is typically better to take Lethality than the filler in the lower parts of Combat. So, if it were me, in addition to getting Imp SS, I'd probably drop Deflection, Riposte, and Fist Spec (as you're not using fists) and spend those 11 points to get Aggression and Lethality.

For additional details, you can refer to my spec to get an idea of what's more typical for a combat sword rogue.

However, fundamentally, if you're under 500 on Aran, it's usually not entirely a gear + talent problem. Typically, it means you're breaking one of the 7 commandments of rogue DPS. In roughly descending order of importance:

1) Don't die.
2) Don't do anything that risks wiping the raid.
3) Maximize your time on target
4) Don't let your energy cap out.
5) Don't let SnD drop
6) Use one of the spreadsheets to figure out your best cycle; this will usually be the highest rupture uptime cycle that doesn't violate rule 4 or 5.
7) Use your cooldowns.

Without having seen you play, I can't comment on whether or not you're doing these. But if you're not, you should be.

Currently revising...