@Elimbras:
You made several excellent points about specific parts of the sim being models. The difference that I was really going into was that Rawr utilizes a 100% model, rather than modeling out things like the attack table then running a 10,000 iteration check of what a fight ends up happening with randomized numbers (the Range for attacks, like weapon damage having 500-550) occurring. Which you gave a really nice explanation of, I may steal that for the future :P

And there's always a chance that things are occurring behind the scenes we don't know about. Those are never fun to try and figure out

Let me reiterate that there are advantages and disadvantages to each method and having both at your disposal is always a nice check back and forth.

@Rysidion
I think I need to do some more research on the relationship of Thunderclap and Rend and what actually occurs when Thunderclap is refreshing the Rend. I can certainly (and without more than about an hours work) try it each way and see what happens. Thank you for posting all that raw data, would you mind sending me the character file that you are working with in pm? I could then load the same toon into Rawr and give a direct comparison.

Quick note without posting a ton of data. When I change it to not apply Rend's Initial Damage when you have Blood and Thunder, you lose Multi:103|Single:77 DPS from Rend. I also added that when you have BnT, you lose a Rend Tick since you have to refresh it before the last tick (50% of the time at 1/2 points).

All the other numbers stay the same. We still have a 1.8k difference between Incite and BnT specs without multi-targets and 2.1k with.

@Rysidion:
I just noticed you had asked how I could have that much mastery. Its all from gear, enchants and gems. I added it up manually and came to the same result on screen.
slot | Base+Gems+Ench+Refg
Head: 0+20+35+102=157
Neck: 143+0+0+0=143
Shld: 172+0+0+0=172
Back: 127+0+0+0=127
Chest: 0+20+0+86=106
Wrist: 0+60+0+45=105
Hand: 171+60+65+0=295
Belt: 0+40+0+76=116
Legs: 257+80+0+0=337
Feet: 191+20+35+0=246
Ring1: 143+0+0+0=143
Ring2: 153+40+0+0=193
Tnkt1: 0+0+0+114=114
Tnkt2: 0+0+0+145=145
MHnd: 0+0+0+102=102
Rngd: 72+0+0+0=75
Total 2574 (there could be a rounding difference on reforges, need to verify that)

To borrow some useful terminology from the mage forums: the two broad kinds of modeling tools for WoW theorycrafting are Simulators (simcraft, kahorie's DK sim, etc) and Formulators (most spreadsheets, magegraf, apparently Rawr). The flaw of simulators is the computational requirements. It takes a long time to get results, but other than that, it ought to be superior in just about every way. The major flaw with formulators is that they are more prone to faulty assumptions. To borrow an example which turned up in the mage forums: magegraf and simcraft were generating vastly different results for Frost mage dps. Ultimately, the reason is that Frost has unexpectedly high mana problems (to the point where Mage Armor may be preferable to Molten), but the particular version of magegraf had been based on prior assumptions of Frost having effectively unlimited mana.

To that end, I question whether Rawr, which is returning significantly different results than simcraft, might also be based in part on faulty assumptions. Perhaps the BnT presumptions don't properly account for the lost Slams due to the increased number of GCDs required to keep Rend uptime high, or the lost Heroic Strikes from the increased rage cost of using TC to refresh Rend, or so forth. These are nothing more than "just so" stories which might explain the differences, since I lack the programming ability to check how Rawr compared to simcraft is calculating things.

The way the Arms rotation code works, it will have taken those points into account. More GCDs eaten by TC would reduce slams and eat more rage that could have gone to HS/CL.

That said, I'm investigating a couple of specific points that have to do with CS buff uptime and OP chance to pop from Dodges right now.

I'm sorry, but I am from the tldr crowd.

How should we spec our DPS come 85 and how will our DPS rotation pan out?

Thanks

A nitpick, but are you sure rend refreshes work that way? I could have sworn rend did damage on initial application, but if you refresh rend before it falls off, it just refreshes, it doesn't give you the initial application damage.

Given that, I could see with the glyph reducing tclaps rage cost it possibly being a DPS increase, given at that point tclap is basically spending 5 rage to gain 1 thunderclap worth of damage, and potential splash damage.

I had given up on it being a DPS increase given how pitiful thunderclap hits for, but given how restricted rage is, I can almost see the argument for it beating out incite except at very high gear levels (where HS might be used more frequently so you gain more incite procs).


Also thanks for the rundown on the difference between a model and a sim.

That's the way I had it initially, but I've changed the behavior to just refresh now. I have also fixed the activates methods for TC, CS and OP, which shifted the stat values again. Haven't posted yet as I wanted to be sure of what I did.

BTW, Rawr4 is up as of today. There are still lots of bugs with lots of models, but the core and a couple of Models are really polished

Well, the Sim is applying Rend almost immediately after it falls off, just so the initial tick of damage is counted. As Rend hits harder than Thunder Clap and costs less rage, it should be a DPS gain and a DPR gain.

To be fair, I have not set SimCraft to simulate Blood and Thunder, because I honestly don't know how to. I don't know if SimCraft has the innate data of Thunder Clap already.

Again, Incite will always be a DPS increase, period. There is no situational use where it might not. Your HS will crit more often. Heroic Strike is used quite often in SimCraft, so I assume the same is true of actual raiding under similar conditions.

The Rend cycle used by SimCraft allows for 6 ticks per Rend, every time. There is no refresh, ever. It's always "Rend fell off, apply new Rend". The difference here compared to B and T is that B and T allows for 5 ticks of Rend on the first 15 second cycle, then 6 ticks for every 18 second cycle after that as long as you continue to refresh it. Instead of the first tick of Rend gained by Rending whenever Rend falls off of the target, you get Thunder Clap's damage. After 30 seconds (not counting for latency when reapplying Rend), you will end up with 12 Rend ticks on the target using the Reapply method. You will end up with 11 Rend Ticks and one Thunder Clap using the refresh method, because you will not gain the extra tick you would otherwise have from reapplying.

That may be very convoluted to read, but it's pretty easy to see in practice. I don't know how to make a good looking table, so here is a makeshift one:

15 Rend

12 Rend

9 Rend

6 Rend

3 Rend

0 Rend

(reapply)

15 Rend

12 Rend

9 Rend

6 Rend

3 Rend

0 Rend

VS

15 Rend

12 Rend

9 Rend

6 Rend

3 Rend

3-0 Thunder Clap

(refreshed)

15 Rend (this is the same as the 0 tick of the earlier Rend that was refreshed)

12 Rend

9 Rend

6 Rend

3 Rend

0 Rend

SimC has very rudimentary aoe and tanking support at the moment which means that while TC is modeled, it is not part of a default action priority list. Since TC is not used, we failed to implement Blood and Thunder.

If there is any immediacy to this need, please PM me or open up an Issue and I'll make sure it gets resolved.

ded, while I got your attention(maybe), I was playing around with absurd amounts of mastery rating and noticed that the increased opportunity strikes didn't increase the uptime on Colossus Smash. A difference of 10000 mastery rating didn't improve the uptime(~50%).

Are opportunity strikes supposed to proc sudden death, or is there a hidden ICD on the talent?

The relevant testing was posted a few pages back.

Ah, thank you. So, it doesn't look like SimCraft is correctly modeling SoO -> SD. Which would up Arms simulated dps even more.

Another SimCraft question, I've tried modeling activating Inner Rage just for things like overpower, but I'm having trouble canceling the aura afterward. Is there a way to do this?

Dedmon, thanks for the update. Downloaded the new one and confirmed that increased mastery lead to increased colossus smash uptime.

Another question for you: do you know who's maintaining the Arms simulation? Simcraft gives it one of the highest possible dps of all the classes/specs, but if you go to worldoflogs.com, you can see Arms is one of the worst. The boss I'm using for comparison is Argolath, which seems like a pretty straight forward stand and dps fight. Is there something I'm missing?

Simcraft seems a little off for Arms, especially for Deep Wounds where I see a huge difference with my own results in bossfights (Simcraft says 13,5% where most of the time it's 7-8%).
But I don't agree that Arms performs poorly, I have no problem to compete with other classes in our raid and I still have to get used to this playstyle. The reason that you see few Arms Warriors that perform well on WoL is that Arms is harder to play than Fury and there are far less players who play this spec (or played it in Wotlk).

Another important thing to note is that SimCraft is in full 372, T11 gear, so the stats are much higher than what we are seeing at current levels of gearing. Arms will of course be MUCH better with that level of gear, due to it's reliance and focus on Colossus Smash, and I believe the stats will more closely correlate with those of SimCraft. I haven't personally tried a SimCraft at the heroic blue level, but that should be something to do to see what the breakdown is.

Of course, that's in a perfect world with perfect awareness and no bad judgement calls. Arms is somewhat unforgiving, which is one reason why the parses can't keep up to a sim for the spec.

I assume you're referring to more gear means more mastery mean more sudden death. Haste helps a bit too, more rage for more heroic strikes.

I don't buy it.

If you take the T11 suit in simcraft and add a couple zeros to the mastery stat on an item, and then subtract the resulting difference in opportunity strike and deep wounds, that's pretty much all the increase in dps.

The interval for colossus smash went from 11.12 to 9.87. Other abilities, not affected by mastery, like slam, saw dps increases from the increased uptime, but not much. ~1.25s increase is decent uptime improvement, but it doesn't look like it makes a big difference for arms dps.

I still think there's something wrong with the simulation itself. Anduryondon might be right about deep wounds, I've noticed some large difference between simcraft and my logs.

Judging by current logs, one cause appears to be low levels of Critical Strike among Arms warriors. Look at the discrepancy between Execute criticals. Fury warriors generally enjoy percentages around 25% or even higher; while Arms players generally struggle to break 10%, and some are actually hitting Execute a dozen or more times without a single critical. Given Execute's correspondingly higher contribution to Fury's damage done within a short time period, that can't be insignificant.

Edit: I'm avoiding broader conclusions; just trying to interpret the numbers as of now.

There are fewer SimC developers than there are modules, so most of us specialize in one or two and help out on all the rest. My main responsibility is the underlying engine (ie: make it as easy as possible on the other devs) and picking up any slack.

Ideally, I would love for people to open Issues on topics just like this one you are discussing. The comments that refer to particular details not "smelling right" are invaluable.

Deep Wounds and Ignite benefited significantly from the the do-not-defer-on-dot-refresh mechanic change. While I support the aura-delay-related "munching" that occurred in 3.x I believe there are new client-server-related problems that reduce in-game DPS of which I am not aware.

If you do open an Issue, please include some logs against which we can compare simulated results. We are also very, very interested in optimizations to the default action lists. You mentioned some concerns with Slam. If you click on the name of the abilities in the table, you will see some underlying information related to scaling that we automatically extract from the client files.

If you prefer to continue the conversation here, I will keep abreast of developments.

A tiny bit off the ongoing topic, but useful information nevertheless.

When doing the 2nd encounter in Bastion of Twilight, you can charge the bosses while in melee range and get the Juggernaut effect. Nice addition to have better chances at keeping Wrecking Crew up as well as a rage booster.

Ded,

simcraft is an excellent tool, so I'd rather do it your way. How do I go about opening an issue? I assume this is something done on your google code site?

The ONLY discrepancy I've been able to find so far is when you import someone's profile from battle.net or JUST the Arms T11 BiS profile, the arms DPS results seem inflated.

However, if you import the whole warrior T11 profile(2H/1H Fury and Arms), the Fury results are the same as if you had simulated them singularly, but the Arms results are much more inline(about 20% less).

For example, I get 29227 DPS from the singular Arms simulation, no monkeying around with the default action list. If I batch simulate all three, I get 23684. I don't know how, but there seems to be some flat DPS multiplier when simulating Arms dps only. Not sure what else it would be.

Though, still a bit concerned Deep Wounds numbers don't exactly line up either.


EDIT: just found it. Stuck this line of code in the Arms-only simulation:

optimal_raid=1

And got 23683 DPS. No idea why removing that would inflate dps,

SimC had a bug managing Blood Frenzy when Savage Combat was turned off. (Setting optimal_raid=1 turned on Savage Combat, occluding the bug.) This has been fixed in r6081.

I will push one more release before Christmas. I would like to dive into Deep Wounds prior to that. Any logs exemplifying the disconnect between game/sim sent to me via PM, email, Issue will be investigated thoroughly.

I believe you are on the right track with Deep Wounds. I just imported my Armory profile and compared 1000 iterations in SimC. I compared the results to a kill parse of roughly-similar length from this past raid week (Valiona and Theralion): World of Logs - Real Time Raid Analysis

Below you'll find a comparison of the damage outlay. The first number is from my parse, the second predicted by SimC.

Overpower - 20.6% vs. 15.5%
Mortal Strike - 15% vs. 10.4%
Heroic Strike - 14.1% vs. 9.6%
Melee - 12.4% vs. 11.1%
Opportunity Strike - 12.2% vs. 13.3%
Deep Wounds - 9.8% vs. 14.9% (side note - no DW tick crits? Rend tick crits appear fine; my DW contribution is usually closer to 7.5%)
Rend - 6.5% vs. 5.2%
Colossus Smash - 3.7% vs. 4.1%
Execute - 2.2% vs. 6.3% (was experimenting here somewhat; would probably Execute more in the future).
Avalanche - 2.1% vs. 0.5%
Slam - 1.4% vs. 9.1% (I almost completely de-emphasize Slam).

Now, it's possible that I simply played poorly and made horrible decisions. However, at the time of this post this parse ranks favorably compared to other Arms parses and the damage outlay is very similar to the top parse for that encounter (here: World of Logs - Real Time Raid Analysis, no Slam usage at all, more Execute usage and 11 TotT).

Just perusing the various 25M Fury and Arms leaderboards in WoL, it's clear that something is off as the latter trails significantly in all fight types and environments.

I really dont' think you should de-emphasize slam, in my Valiona/Theralion parse it's 8,5% damage and I'm 1000 dps ahead from your reference parse without any TotT usage (10-man).