First, what version are you using? There have been a few changes to DK since our last release and we should have another release out once we make the DK PTR changes. So it's important to note if you're using the latest release or compiling from the SVN.

What do you mean by remarkable numbers? If they're not matching what you expect in game, check the action list, if you're giving it different actions then what you do in-game, there's no way it can be close. The output of the sim is only as good as it's inputs.

NoCS is taken into account and is easily testable by adding/removing a talent point and seeing your misses increase/decrease by 1%.

I don't see HB doing 10 dps less than Obliterate. I see their DPET nearly the same, but that's not the same as DPS, since you cannot and do not use them the same amount of times. Again, if something is clearly wrong, please give us examples/reasons why (e.g. HB hits for 10k in game, 20k in the sim). The profile you used is also stacking mastery.

There are no misnomers in a simulator unless you're losing a low amount of iterations, more iterations will eliminate the effects of RNG. Ten thousand iterations is recommended for testing small changes, but 1,000 will do for larger changes.

Also, you can copy the URL of the charts on our reports and use those here, instead of taking a screenshot and uploading them. That should save you some work.

After some in-game testing we found out that the sim was incorrectly allowing Icy Talons and Improved Icy Talons to affect rune regeneration rates. After removing the benefit to rune regeneration, 2H frost is no longer GCD capped and it's scale factors match the OP. DW Frost is still GCD capped in the sim and gains DPS by switching to Unholy Presence. The scale factors also do not match the OP, with them being Haste > Crit >> Mastery in the sim. We know this goes against the community, but we can only scour over the code so much before players that play a DK daily can tell us what's going on.

The full report is available here: Simulationcraft Results

I should have reworded that. I hadn't experienced competitive numbers (i.e. 16k sub was tested)

Dually-noted

I was stacking Mastery based off of the general consensus of Dual Wield (Mastery outweighs all)


Not to jump to conclusions, but am I reading 2H Frost back on top again?

Right now I'm not able to get consistent results out of the simulator with respect to stat scaling when using scale_delta_multiplier=0.2. The overall dps, dynamic buffs, procs, and gains all stay the same (+/- small deltas based on rng), but scale factors change drastically.

Here's what I did, using the same chardev profile that you used in Simulationcraft Results

using SimulationCraft 406-8 for World of Warcraft 4.0.6 Live (build level 13623)

./simc chardev=33828 name=BiS-372 save=372.simc

and used the following options:
threads=4 ; iterations=20000 ; calculate_scale_factors=1 ; scale_delta_multiplier=0.2 ; scale_only=strength,crit,haste,mastery

Results:
Str=2.6053 Crit=1.0735 Haste=1.0839 Mastery=0.8033
Str=2.6484 Crit=0.9815 Haste=1.0867 Mastery=0.7680
Str=2.4217 Crit=0.9686 Haste=1.0234 Mastery=0.8788

Then I changed run.simc to remove scale_delta_multiplier=0.2 and re-ran it 3 times

Results:
Str=2.5570 Crit=0.9939 Haste=0.9734 Mastery=0.8823
Str=2.5596 Crit=0.9635 Haste=0.9550 Mastery=0.8759
Str=2.5346 Crit=1.0065 Haste=0.9479 Mastery=0.8465

Much more consistent. I also tried playing around with five_point_stencil=1, center_scale_delta=1, and smooth_rng=1, with similar inconsistencies. Kahories, for reference, does not change it's EP values for crit or mastery with a 300 delta (though haste drops significantly due to gcd capping).

Also, as iterations drops, the results are even worse. 1000 iterations, 0.2 multiple (and setting scale_factor_noise=0 so I actually can get output).

Results:
Str=2.8306 Crit=1.3456 Haste=1.1830 Mastery=0.6296
Str=2.6870 Crit=0.3124 Haste=1.4333 Mastery=1.0956
Str=2.4843 Crit=0.8375 Haste=1.1617 Mastery=1.1048

Whereas kahories remains consistent at those levels. At this point, kahories hasn't been touched since January and is missing 4.0.6 metas, and enchants, so I don't exactly want to compare results between them because kahories could be wrong too. However, mastery, crit*, and strength are linear stats, and any simulator that disagrees with that must be wrong. Every time I've done a scaling plot in simulation craft, these stats produce spikey graphs. For reference, here's a plot from kahories that was posted on consider's blog awhile back

http://considerit1.files.wordpress.c...tdwscaling.png

Every stat except haste produces a completely smooth curve, crit being slightly curved due to DR of course. Here's a scaling chart from simulation craft:

http://i.imgur.com/PQbT0.png

I can't explain why the graph is so spikey, but that spikiness does explain why the stat weights vary as much as they do with low deltas, and why they do not in kahories. Of course none of this actually helps us come any closer to what really matters, the actual stat weights. Maybe for next week I'll try reforging all of my mastery to crit and see what happens during raids.

When I saw the report that Kahorie's sim uses only a 50pt stat delta for scale factor calculation I was hugely impressed. That is nothing short of remarkable. In simulation, when the iteration count increase, scale factor variance decreases slower-than-linear. When stat deltas decrease, scale factor variance increases MUCH-faster-than-linear.

The plots from a couple posts back were incredibly smooth. That level of "smoothness" can only be accomplished by immense iteration count or by removing sources of RNG. I may have to peek at Kahorie's code to see what areas were chosen to be deterministic.

For reducing variance in SimC, please stay away from smooth_rng. It requires each module dev to perform a fair amount of work for Cata. I removed that from the GUI because I fear code-rot has set in. . I need to revisit that functionality. Instead, I recommend the use of deterministic_roll=1 to ensure use of the same RNG "seed".

You can also dramatically reduce the RNG of latency affects:

queue_lag_stddev=0.00001
gcd_lag_stddev=0.00001
channel_lag_stddev=0.00001

Simulationcraft Results

In that report, it claims the average obliterate hits for 15362, and crits for 31877. It doesn't say whether that's mainhand or offhand, or average or combined, so it's hard to know what exactly to compare that with. It's low for a mainhand, high for an offhand, maybe works as an average, and is way off for combined. Same goes for the rest of the strikes for DW. It claims that contributes ~30% of the total dps, which implies that it is combined since that is about what you'd expect for total mainhand + offhand obliterate to contribute for dps, but combined main + offhand obliterates hit for close to 30k as you can see on world of logs. Hopefully that's just a reporting bug and isn't affecting the dps calculations.

Nate and Arison covered RNG and all that fun stuff. Again, when linking separate charts, it's extremely useful to link the entire report as well, or at least tell us what version, how many iterations, and any other options you use, so that we can reproduce it and/or see where a mistake was made.

Currently any strike that hits with both hands is reported as one line. We'll get them split before the next release, so it's easier to compare to WoL, but if Obliterate is only hitting for half of the normal value, then we have a problem. Ironically, if that's true, it's only going to push mastery's value further down in comparison.

[e] Reporting is split in SVN. A few strikes have different min-max damage values (Obliterate, Frost Strike), resulting in a slightly different DPS.

It just blows my mind that in the most recent SimulationCraft Spreadsheet (4.0.6 v8 - Tier11 Ilvl372), 2H Frost is pulling a whopping ~2700 DPS ahead of its Dual Wield counterpart, making it the 2nd highest DPS spec on the list, while DW Frost is in the bottom tier.

How can trading [Nervers of Cold Steel]+[Threat of Thassarian] for [Might of the Frozen Wastes] (and wearing the appropriate 2H weapon) result in such an enormous DPS gain?

Also, this pretty much goes against what's said in the OP: If that happened to be true, DW frost would be dead (or just a temporary spec until you get a 2Hander), as there would be no reason pick it over its 2H superior alternative.

I think it's unlikely that they will remove BS from game, just likely they didn't remove abilties like warriors' Strike or shamans Primal Strike which aren't used in endgame anyway. Actually I think it's unlikely that this change will go live since it changes Frost into basically 2 button spam specc with 1 more on a procc (and few more abilities once in a while); I doubt Blizzard's intention was to make rotations vanilla levels of faceroll again.

With helpful analysis from folks in this thread, several inefficiencies and bug-fixes were made to the DK module. Please re-check that link now that they have all been incorporated.

Frost 2h - 25534 dps
Frost 1h - 25354 dps
Difference - 180 dps

Where is this 2700 number coming from? I mean sure Frost has a bigger difference than Fury does between it's two specs (2 dps), but that's pretty good.