Ok so I ran those tests.
Dps/point for haste (my gear by pressing ep calculator button)
2.04
dps/point for haste (by subtracting 50 in av dmg tab and 10k runs)
1.96 ran again also 1.96
dps/point for haste (by adding 50 in av dmg tab and 10k runs)
2.11
So we have three different dps/point values 2.04,1.96,2.11 that when compared to other dps/point values have enough of a difference that it seems we cannot make an accurate assessment of which is better. What mean by this is let's say my white hit is 2.01 dps/point by using the EP calc button. Do I accept the 2.04 or 2.11 value and gem haste or accept the 1.96 value and gem hit or should I run 10k runs with +/- 50 hit and compare that to the 2.11/1.96 values?
My logical thought on this and I may be wrong is to do a double EP test to get a value when adding to a stat by 50 ex. 2.11 and when subtracting a stat by 50 ex. 1.96 then compare re difference in which stat you would lose vs which would be gained.
The changes are so minor though that I don't think I will be noticable unless you take the time to compare every change.
Ps. Forgot to add that the EP test most SE(dps/pt) were 0.01-0.06 while the 10k runs with +/- 50 has SE(dps/pt) of 0.111 on all three.
I'm going to run some 10k's on hit tonight and compare those.
I have added most racial bonuses, except for the Blood Elf one. I also fixed a bug with horde names and trinket procs. Fixed some data entry errors brought up on the previous page. All links have been updated.
Gmpotu, the purpose of these tests was to try to find out which hypothesis best explains volatility of EP values, built in RNG or variability of EP estimates as you change the stat. Let's introduce some notation not to lose track of your tests. x(A) is dps/point value of haste at A deviation of haste from your base gear. We are trying to find out x(0). We know that x(-200)=2.04 with confidence interval of (1.98, 2.10); x(-50)=1.96 (1.74, 2.18); x(+50)=2.11(1.89, 2.33). First thing, as you can see increments of 50 do not give you very precise estimates of dps/point of haste. SE of 0.11 is too large for trying to precisely estimate dps/point values of around 2. Second, x(-200) is more precise, but it might be scewed because we are measuring the dps/point at a point 200 haste below our gear level.
I am guessing changes in dps/point when increasing or decreasing haste are second order but to be sure, you can try this. Use same gear as for the tests above, but this time reduce haste by 200 in the EP adjustment range in the gear page. Then rerun EP calcualtion (sorry I know it takes long). Then you will have an estimate of x(-400) which has same good precision as that of x(-200). The you can see how far it differs from x(-200). If x(-400) does not differ much from x(-200), then variations in x(-50) and x(50) are largely due to RNG.
You should also try to plug your gear into Aldriana's spreadsheet which does not have the effect of RNG. While his modeling is different, you can still get a pretty good feel for how much and in which direction dps/point values change as your haste decreases by 200. If you can't extract dps/point values from Aldriana's spreadsheet, just observe how much EP values change when you reduce haste by 200.
I ran the EP calculation again this time with -200 haste on the gear page.
dps/pt 2.13 this time. So
x(-200)=2.04
x(-400)=2.13
x(-50) =1.96
x(+50)=2.11
I'm not sure what to think now. I'll try running ald's sheet tonight.
On a side note I did the same test with hit.
x(-100)=1.94
x(-50)=2.12 run 1 ,10k iterations
x(-50)=1.99 run 2 ,10k iterations
x(+50)=2.04
x(-100)=1.85 -200 haste
You should also try to plug your gear into Aldriana's spreadsheet which does not have the effect of RNG. While his modeling is different, you can still get a pretty good feel for how much and in which direction dps/point values change as your haste decreases by 200. If you can't extract dps/point values from Aldriana's spreadsheet, just observe how much EP values change when you reduce haste by 200.
Let me know what you find out.
I put my gear into Aldriana's sheet and it was almost 1000 dps difference. I don't think his sheet takes into account ToTT on cool down being applied to you from another rogue or use of saronite bombs. I'm also not sure if he implements using the new T10 - 2 set bonus because even disabling bombs and ToTT on CD your sheet is still about 150-200dps higher but that may be an odd RNG thing.
Aldriana's sheet does not take into consideration that you are receiving the tricks buff, only whether or not you are using energy to give it. It does take into consideration the set bonus.
I have some questions regarding the usage of haste cooldowns, especially the right timing of those. The topic was brought up here Ensidia. So my understanding is that blade flurry, heroism/bloodlust and slice and dice synergize due to the multiplicative stacking of their effects. I wanted to simulate the effects with different settings in the dps strat section, but apparently there are no options regarding cooldowns in it. Is there any possibility to test severeal cooldown strategies with this tool? The only ressources I could find where an wowwiki article about haste and the old Rogue Theorycrafting Think Tank article.
...I wanted to simulate the effects with different settings in the dps strat section, but apparently there are no options regarding cooldowns in it. Is there any possibility to test severeal cooldown strategies with this tool?...
You can see how the current haste cd's are used in the combat tab, but it's not exactly that user friendly. If you don't mind meddling with the formulas, you can change the timing yourself to accommodate your needs.
It seems like the default option matches BL with hyperspeed + speed potion and when Hyperspeed expires, Troll racial with BL. I don't know why Hyperspeed and Troll racial aren't used at the same time, but maybe you start to overlap envenoms too much. Changing these to all be used at the same time, increased my dps slightly (~40 dps).
EDIT: I have not noticed any problems stacking Berserking and Hyperspeed Accelerators together, unless something was changed recently.
No the spreadsheet is not designed to let you choose when to use haste related cooldowns, such as bladeflurry, shard of the crystal heart, troll berserking, and haste pot. It is assumed all those buffs are stacked, and used along with heroism on the 10th second and then as soon as the cooldown runs out.
There is one exception. Since I think it was mentioned that troll berserking cannot be activated at the same time as hyperspeed accelerators, the simsheet is instructed to activate berserking on the 22th second. Either way the timing of the cooldowns is set in stone and they are timed in a way to stack additive and multiplicative haste buffs.
Regarding the underlying theory, nothing has changed since the time when it was determined that stacking haste buffs is beneficial. It is a pretty straightforward math. The part regarding stacking multiplicative buffs can be found on wowwiki. But it is also true that using additive buffs while multiplicative buffs are active is better than using them separately.
For example: Let's say you have 10% haste from gear. You can use a potion with haste rating which will give you another 10% additive haste, and you can activate a buff that will give you 40% multiplicative haste. Each buff lasts 10 seconds. You can either use them simultaneously or let the potion last first 10 seconds and the 40% buff last the second 10 seconds. Let's say you have only a 2.6 speed weapon.
Case 1 (Stack). First 10 seconds: Haste = (1+10%+10%)*(1+40%). With that speed you will do 10/(2.6/haste)=6.46 attacks. Second 10 seconds: Haste = (1+10%). You will do 4.23 attacks. The total number of attacks in 20 seconds is 10.69.
Case 2 (Separate): First 10 seconds: Haste = (1+10%+10%). With that speed you will do 10/(2.6/haste)=4.62 attacks. Second 10 seconds: Haste = (1+10%)*(1+40%). You will do 5.92 attacks. The total number of attacks in 20 seconds is 10.54.
So the benefit of haste when you stack additive and multiplicative buffs is higher. Normally however, the benefit you get from this timing is not that big. So the rule of thumb is that stacking is not worth it if you delay the buff for so long that you are not going to have time to use it again. Only if you are sure that the delay is not going to lower the number of times you can use a certain haste buff, then you can try to delay and stack some buffs.
There is one exception. Since I think it was mentioned that troll berserking cannot be activated at the same time as hyperspeed accelerators, the simsheet is instructed to activate berserking on the 22th second. Either way the timing of the cooldowns is set in stone and they are timed in a way to stack additive and multiplicative haste buffs.
Though it is true that on-use trinkets lock you out of the hyperspeed accelerators it is not the case with Troll Berserking nor Orc Blood Fury. I am able to run Adrenaline Rush, Troll Berserking, Hyperspeed Accelerators and Blade Flurry all at the same time.
I have implemented a new module that allows you to measure the impact of target switching and fight interruptions on your dps. As a user you can enable the module in DPS Strat tab and choose the following three parameters:
Travel time = time it takes a rogue without run speed buffs to reach the add (1-10 seconds). Any run speed buffs, such as Fleet Footed and Cat's Swiftness, reduce the time it takes you to reach the add and then back to the boss. Add dps time = time spent attacking the add (0-30 seconds) Time between adds = intervals between add spawn times (50-180 seconds)
When you enable fight interruption, first add is programmed to spawn 20 seconds into the fight (hardcoded). Then there is first travel period furing which you reach the add, followed by add dps period, followed by second travel period and finally back to boss dps. The cycle repeats itself and the length of the cycle is equal to time between adds you choose. If you choose 0 add dps time, then there is only one travel period up to 10 seconds and you go back to the boss. Thus by setting add dps time to 0, you are effectively simulation a small fight interruption.
Details of modeling:
1. Right before each run phase, provided you have the energy and DP stacks for it, your combo points are spent on either an eviscerate or an envenom depending on your dps settings. This is done so that your combo points are not wasted and you get an envenom buff when switching to a new target.
2. The add is assumed to not be sundered, but its armor is fully exposed one second into add dps time. When you go back to the boss, it still has full sunders.
3. It is assumed that the add has all other debuffs as chosen by you in the Inputs tab, in particular, it is assumed that when you start dpsing the add, it is already poisoned.
4. While you are running and dpsing the add, your DP stacks on the boss are not wiped immediately, DP continues ticking and if you get back to the boss fast enough, you will have a chance to refresh DP and not allow it to fall off. Regular tick reset rules still apply, so if you refresh DP after 4 DP damage ticks, the next DP tick will be reset. Similarly, while you are running to the boss, DP continues to tick on the add until it runs out.
When referring to simsheet results, I want to introduce new notation regarding the interrupted fights. (5/30/50) interrupted fight is a fight with 5 second travel time, 30 seconds add dps time, and 50 seconds between adds. In terms of examples, I'd say closest you can get to Putricide fight is (5/20/50), Jarraxus is close to (2/6/50), and Blood Queen is somewhat like (10/0/120).
P.S. I changed Troll Berserking to start at the same time as other cooldowns. I also moved the starting time of Shard of the Crystal Heart to occur after Hyperspeed Accelerators.
P.P.S I know the file gets bigger and runtime increases, it's the price I have to pay for added functionality. I am constantly thinking of ways to reduce the running time.
I recently plugged my combat gear into the simsheet to play with target switching. My gear is mainly itemized towards mutilate, so I don't have a lot of arpen in my gear. I assumed that since ap vs arpen has been close for my combat build, arpen would be ahead for any fight with switching. Since there will be a few times where poison stacks have to be refreshed, my assumption was that arpen would pull ahead due to poison not scaling with arpen.
I'm a little stumped by the results I saw. 20/51/0, 1000 iterations for each. The only 'gear/input' changes made were the gems used. Red gems (including JC ones) were either all ap, or all arpen. I used my gear obviously, but otherwise nothing changed. The only 'dps strat' change was either a 1 or 0 for target switching.
Regular dps with all ap gems in red sockets: 11110.
Regular dps with all arp gems in red sockets: 11128.
Target switching dps (5/30/50) with ap gems: 9252.
Target switching dps (5/30/50) with arp gems: 9248.
Based on the results I saw, my assumption was completely backwards. I'm a little bit confused by this.
Looking at your results, it appears that your armor pen and ap based setups produce about the same dps. I would be surprised if 18 dps difference were significant at 1000 iterations. The target switching mode does not change that balance. So basically what you are seeing is the two setups are equal in both modes.
As to which one should be theoretically higher and by how much, it depends on interruption parameters. Don't forget that with haste-based gear you stack up DP pretty fast and resume your poison damage, but at the same time your DP continues to tick on the boss adding that much of AP based damage to the mix. So it's not entirely clear that target switching favors arpen setups.
P.S. I am finishing analyzing BiS lists with Lich King gear, so expect some results over the weekend. I will also analyze various target switching modes.
Gemming: 20arp x 15, 34arp x 3, Nightmare Tear, 10hit/10exp x 2. Optimal gemming is based on achieving armor penetration hard cap, while staying under the crit cap and almost reaching the expertise cap. Armor penetration food.
Professions: JC/Engineering
Alternatives: Other items, professions and gemming choices produce very close dps results for combat (within 20 dps). In particular:
TAJ+Engineering vs Herkuml War Token+Blacksmithing
Fal'inrush, Defender of Quel'thalas Heroic vs Windrunner's Heartseeker Heroic (requires regemming)
Band of the Bone Colossus Heroic vs Ashen Band of Endless Vengeance (till you reach exalted)
Gemming: 20haste x 9, 34haste x 3, Nightmare Tear, 20ap/10haste x 7, 20hit x 1, 20ap/10hit x 1. Optimal gemming is based on staying under the crit cap and almost reaching the expertise cap, while maximizing haste and AP as secondary stat.
Professions: JC/Engineering
Alternatives: Very few alternatives within 20 dps. In particular:
Agility/haste vs AP/haste (requires adding hit gems)
All results were obtained for a 5 minute fight without target switching.
It would be very amusing if it were intended, in your sheet there actually is a "Steakthrower", although the real name is Stakethrower...dont know if you're aware of that.
The file is now updated and can be downloaded from the front page. In the new file you will find all LK loot after the recent adjustments. I also added Battle Squawk buff in the Inputs page.
Target Switching
Based on the recent simsheet, I ran an analysis of target switching. In general the effects of target switching on gear choices are not very big. For each spec, I have testes several points of contention and tried some common sense alternatives. I tested these alternatives for three interrupted fight modes: (5/20/50) Putricide, (2/6/50) Jarraxus, and (10/0/120) Blood-Queen.
Mutilate
Spec: 51/13/7 vs 51/18/2 - the latter is firmly ahead by more than 100 dps
Tier boni: choosing 2/5 with head, shoulders, and chest offpieces is still lower than 4/5, however the gap is smaller
AG vs AP: AG/haste is slightly stronger than AP/haste (change requires adding 2 hit gems), however, haste remains the main stat for mutilate.
Herkuml vs WFS Heroic: the latter is still ahead by more than 30 dps
Combat
Spec: 5/5 improved poisons and 0/3 vile poisons is ahead of 4/5 and 1/3, by the order of 30 dps
Tier boni: the gap between 2/5 and 4/5 is smaller, but the latter is still ahead by at least 70 dps
TAJ+Eng vs Herkuml+BS: for target switching in all modes, Herkuml pulls ahead by less than 20 dps
Adding a DPS table below:
Simsheet
Mutilate
Combat
Base
13855
13668
Blood-Queen
10/0/120
12579
12230
Jarraxus
2/6/50
12308
12476
Putricide
5/20/50
11078
11146
The effect of fleet-footed and cat's swiftness varies by the type of fight. For a Putricide type of fight, Fleetfooted is worth up to 286 dps for two talent points. For combat Cat's swiftness is worth 40 dps on top of Icewalker enchant. However for a fight like Jarraxus, Cat's Swiftness is a downgrade of 40 dps relative to Icewalker.
If you are seeing crit reduction in AvDamage tab, you are looking at an older version. In the latest version, crit depression is directly incorporated in your effective crit rate. Also in your gear tab, 4.8% is part of your crit cap formula, which now equals to 100%-D-M-G+4.8%.
Got the new version and it worked with the 4.8%. I didn't realize there was update since the letter stayed the same.
The new version gave the expected results similar to Ald's sheet. Is there a way for the gem slots that it could autoselect the bonus if you have the right colors matched? Clicking through each gem and then clicking yes or no with how small the lines in the sim are seems tedius.
Thanks for the excellent sheet and all the hard work.
Got the new version and it worked with the 4.8%. I didn't realize there was update since the letter stayed the same.
The new version gave the expected results similar to Ald's sheet. Is there a way for the gem slots that it could autoselect the bonus if you have the right colors matched? Clicking through each gem and then clicking yes or no with how small the lines in the sim are seems tedius.
Thanks for the excellent sheet and all the hard work.
Clicking yes/no only affects the listed gems, such that yes at a blue socket forces only purple/green/prismatic gems to show in the list. The socket bonus is automatically included if the gem matches.
Are you sure that the offsets helm + shoulders (Geistlord's Punishment Sack (Heroic) (Cultist's Bloodsoaked Spaulders (Heroic) isnt the bis for mutilate? According to my calculations, running the 2 offset is ahead by 50-60 dps.
If it is more than 13855 by a statistically significant margin, PM me other details regarding gear, gemming, rotation and professions. From what I have tested 13855 was the highest mutilate result (within statistical significance) and it involved the set that I have posted as BiS above. This result is repeatable within 10 dps, which is the size of the confidence interval.
I just ran 2/5 with your suggested offset pieces and got 13827 with haste/ap gemming and 13804 with haste/ag gemming.
Based on the recent simsheet, I ran an analysis of target switching. In general the effects of target switching on gear choices are not very big. For each spec, I have testes several points of contention and tried some common sense alternatives. I tested these alternatives for three interrupted fight modes: (5/20/50) Putricide, (2/6/50) Jarraxus, and (10/0/120) Blood-Queen.
Mutilate
Spec: 51/13/7 vs 51/18/2 - the latter is firmly ahead by more than 100 dps
Tier boni: choosing 2/5 with head, shoulders, and chest offpieces is still lower than 4/5, however the gap is smaller
AG vs AP: AG/haste is slightly stronger than AP/haste (change requires adding 2 hit gems), however, haste remains the main stat for mutilate.
Herkuml vs WFS Heroic: the latter is still ahead by more than 30 dps
I'm getting significantly different results for non-Heroic gear levels. For example, with my current gear I am seeing 51/13/7 ahead of 51/18/2 on Putricide/Jaraxxus/Blood-Queen, and surprisingly Saurfang as well (@300 iterations each, ~70-80 dps on most). I'm also seeing 51/18/2 with Opportunity behind 51/18/2 with RS on most of the interrupted fights, which contrasts with some other conclusions on target switching. Can anyone confirm getting similar results with BiS outside of hard modes?
Are you using the latest version of the spreadsheet? There were two changes that affected target switching in the latest version and the version number was not changed. Specifically, I made it so that all DP stacks were wiped on the add when the add phase was over, which assumes the add is dead. When you first reported your results re: 51/13/7 vs 51/18/2 while target switching, the DP stayed on the add. There was also a bug with TAJ when used with target switching. If you are getting these results using the latest sheet, perhaps you need to PM me with your specifications, such as gear, spec, etc.
01/27/10, 1:28 PM
Similar Threads
