Arcane Brilliance: Optimizing your mage's gear through Rawr
Close your eyes.
Imagine, if you will, that you have a mage. Ha ha, of course you have a mage. Sometimes I crack myself up. Now imagine that your mage has gear, but that gear is not optimal. Now open your eyes.
Hello, every mage ever.
Yeah, chances are pretty close to 100% that no matter who you are and how religiously you play the game, your mage, beloved though he may be, does not have every slot filled with best-in-slot gear and does not have every item perfectly gemmed, enchanted and reforged so that your DPS is up to its fullest possible potential. WoW is a computer game -- an old one, sure, but still home to millions of algorithms and formulas and computations and other math terms that I don't fully understand. It would take, I firmly believe, a robot from the future to calculate all of those numbers and variables and turn them into something a human brain can parse accurately enough to actually benefit from.
Well, mages and magesses, I have recently contacted Kavan, one of the main authors of the ubiquitous gear optimization program Rawr, and I can tell you that he is that robot from the future. I only pray he can protect us from whatever other, more evil and technologically advanced robot from the future that might at some point follow him through the wormhole with a directive to kill our unborn children or something, because you know that's what's happening.
What Rawr is and where you can get it
I've mentioned Rawr before now in this column, but usually only in passing. I've never been a huge believer in depending on a computer program for optimizing. It has only been during this expansion that I've come to see the massive value of such programs in augmenting my own very fallible ability to optimize my mage. And Rawr is, in my opinion, far and away the best of such programs.
Though it works for all classes and specs, it is probably known best in mage circles, and for good reason. One of the major authors of the program was a mage, and one of the first completed modules for it was the mage module. It is simply the best gear optimization resource out there for us.
It pulls your mage's data directly from the Armory and instantly calculates a million things you need to know. The heart of the program is the gear optimizer. Once you plug your mage's info into the program and run the optimizer, it will immediately tell you some things that could be improved. Have a piece of gear that is suboptimally reforged for your spec? The optimizer will suggest a way for you to eke an extra tenth of a damage point per second out of that piece of gear. And though the program isn't always going to be perfect, it's almost certainly going to be right more often than my stupid human brain is going to be.
You can run Rawr in your browser through Elitist Jerks (though it may ask you to tweak your Silverlight settings first) at Elitist Jerks Rawr.
You can also download the stand-alone program at Rawr, as well as read more information about it there.
The orgins of Rawr
Kavan has been playing WoW since the closed beta, and like all good people, he played and still plays a mage -- an arcane one, to be precise, long before that was en vogue. He remembers a time when he considered Wand Specialization a DPS increase. This was apparently before he evolved into a math robot from the future. His guild, the Silvermoon Crusaders on Kilrogg, has never been on the bleeding edge of raid progression, but they've seen most of the bosses go down at some point or other, and most importantly, though the players involved have come and gone, the guild is still going strong.
It wasn't long before his inner math wizard began to emerge.
Initially I was just playing as any regular game, but with mathematics and computer science background the theorycrafting aspect was just impossible to ignore. At the time I hadn't found any tools that met my needs, so I started my own Excel spreadsheet. It started with just tracking what gear I had, tallying the stats, then working out the spell data, rotations and so on. I've always loved genetic programming so the opportunity was perfect for creating an optimizer for gear selection based on it. What I've always loved about arcane was the mana aspect. In classic WoW days you only really had Arcane Missiles so all flexibility was dependent on gear. Depending on how much spirit you mixed in you could aim for different fight lengths. I remember having around 15 or so gear sets for different combinations of fight lengths and other options.
At some point around then I also stumbled upon Elitist Jerks forums. I remember trying to get others in my guild to use the spreadsheet, but most were very afraid of numbers so there wasn't much success. EJ was the perfect place to find like minded players. I've never been min/max to the extreme; I always remained constrained by my role playing choices, but the community there was a lot more receptive to my ideas. I remember making the first post around August 2006. Then came The Burning Crusade, arcane became more mainstream, so there were more players interested in arcane theorycrafting. I kept working on my spreadsheet, but it remained more or less used just by myself. Toward the end of 2007 I came in contact with Astrylian, who was at the time just starting his work on Rawr. The main premise of Rawr at the time was the ability to pull gear data from armory and having a paper-doll to play with stats and model results. At the time it was just a program for bears. We exchanged ideas, I really liked the ability for automatically getting gear data and I saw how adding things such as gear optimization could make it even better. It was also written in C#, which I had a lot of experience in so it was a natural fit.
When Astrylian released the first version of Rawr in 2008, Kavan volunteered to write the mage module for it, basically porting his spreadsheet to Rawr.
In Excel I used to use 3rd party linear programming solvers for figuring out which cycles have to be used for optimal dps. Part of transition to Rawr meant I had to write an LP solver from scratch since we didn't want to depend on external libraries at the time, so that was an interesting experience in of itself. I started with very naive textbook version and it was serving the purpose at the time. But then with time I started exploring more complex ideas. Being released from the spreadsheet constraints I started looking at things such as cooldown stacking. Adding more and more things resulted in bigger and more complex models and the initial solver kind of started crumbling under its weight so I started researching more advanced papers on those topics and improving it along the way.
Theorycrafting evolves
And then came Wrath of the Lich King. The way the classes in the game worked was rewritten. Rotations changed completely. Well-used spells vanished, and new ones took their place.
With the release of WotLK arcane theorycrafting kind of matured. Before that, everything dealt with static rotations. The introduction of Missile Barrage added a random element to the mix which started the theorycrafting of dynamic cycles. Initially this was all done on paper, but then later I automated the whole process in Rawr so now everyone can answer these questions using the Cycle Analyzer tool. While Rawr was continually fleshed out to be more friendly for novice users it always had a large bag of advanced options available for theorycrafting purposes. Most of those were added because I wanted to do something and there was no other way to get to that data. Optimal cooldown stacking was one of those questions that I always found interesting. Not just in general, but also in specific cases such as when bosses have vulnerabilities in specific phases.
And that's the thing I love most about Rawr. Though it can be used to great benefit by even the most advanced mages playing the game, it can also be utilized by knuckle-draggers without opposable thumbs like myself. With just a few clicks of the mouse, the basic optimizer will provide newbies with information they can use, but the advanced options are also there for hardcore theorycrafters. In fact, I believe the only class it doesn't work for is warlocks (due to the fact that they can't see the computer screen clearly enough through their eyeliner and tears, and not through any fault of the program).
The game's current iteration ups the ante more than ever before.
Cataclysm was the ultimate theorycrafting heaven. The way Mana Adept operates put the mana management dynamic to the extreme. I can just imagine someone at Blizzard thinking, "here try to optimize this," with an evil grin. Finding the optimal way to take advantage of Mana Adept was one of the most complex theorycrafting questions I've ever tried to solve. Starting with theory of optimal control to get to the basic principles of using cycles in order of increasing mps and then converting the abstract theory to quadratic solvers that can solve for optimal dps in Rawr. The question of optimal cooldown stacking became even harder than it was before and until very recently it was basically impossible to get results in feasible amount of time. The newest versions of Rawr now sport a genetic solver in advanced options that is able to answer that question in a few minutes.Now, before you ask, I have no idea what or who a "genetic solver" is. I'm 90% sure Kavan made that up to confuse me. But the fact of the matter is that Rawr has now matured to the point that it can assist, mathematically, with one of the single most confusing and dynamic mechanics ever instituted in this game: Mana Adept. The mana management meta-game has been the bane of many a mage since its inception. And while Rawr isn't going to hold your hand and make you a great arcane mage, it is an invaluable resource in helping you make sense of Mana Adept's complexities.
Rawr is a living project. Kavan and a few other dedicated and brilliant souls continue to support it and improve it, and Kavan specifically does so out of a love for the game and a deep affection for mages. Seek it out, learn to use it, and watch your mage's potential grow. If it helps you kill that warlock even a hundredth of a second faster, I think we can all agree that Kavan's hard work has paid off.
Thanks to Oldchap, author of CombustionHelper and MageManaBar, for putting me in touch with Kavan, and thanks to Kavan for providing most of the text for this column -- and to both of them for their ongoing contributions to the mage community and for just generally being awesome.
Filed under: Mage, Analysis / Opinion, (Mage) Arcane Brilliance
Patch 5.2 interview with Dave Kosak
Inside an old alt's vault
The latest patch 5.2 news
All of the latest Mists of Pandaria news





Reader Comments (Page 2 of 2)
solecizm Oct 9th 2011 5:04AM
I just love the fact that mages are using maths to win xD
Shine on, you crazy magical nerds.
Agrocasey Oct 9th 2011 6:38AM
The "genetic solver" that kavan is talking about is a genetic algorithm, a type of metaheuristic which can be used to optimise complex problems. I've used this technique quite a bit in my work. I'm pretty sure Rawr also has Simulated Annealing, another metaheuristic as an optimisation option. The process that is used in the genetic algorithm is based on the theory of evolution, while simulated annealing is based on a metallurgy technique called annealing
I used to use Rawr a lot in Woltk for optimising my hunter's gear, however since Cataclysm came out, I don't think they have had a developer specifically for the hunter portion of the program. That is kind of a pity, since with the introduction of focus and mastery to hunters, it would make for an interesting optimisation problem. We do have a resource available to us in femaledwarf.com, but it doesn't have the optimisation solvers that Rawr uses. I'm not sure exactly what techiques femaledwarf uses, but I'm guessing its some kind of simulation or something like that. I'd love to see the two resources come together so we could get the best of the both of them.
bvcasey Oct 9th 2011 6:53AM
I realize this is a wow blog site, and the developers on Rawr probably have their hands full with working with the WoW classes, but I'd love to see a similar application to Rawr become available for Rift, especially now that they allow addons in the game. It would also be pretty cool if a similar program became available for SWTOR once it comes out and enough theorycrafting is available for optimising gear for the various classes in that game.
Moorit Oct 9th 2011 9:29AM
Interesting that this should come up this week. I finally made the decision to ignore RAWR because it was telling me that my tier 11 4pc bonus was so profoundly amazing that it wouldn't let me upgrade anything to tier 12 or firelands offset pieces, even when I had enough VP to get two pieces of tier 12. Anybody know what's up with that. The tier 11 bonuses are good, but are they so good that I shouldn't use anything from Firelands or the VP vendor in those slots?
Albert Oct 9th 2011 2:26PM
I think you must be doing something wrong, my aim is einstini15, ill help you set it up if you like.
nieboh Oct 10th 2011 11:49AM
Clearly I have no idea what I'm doing with Rawr. I just did an optimization with a pretty substantial thoroughness. (After waiting a few hours I gave up waiting and went to bed. When I woke it was done.)
My Optimizer results say the following:
Current: 28552.44 Optimized: 26636.69 (-6.7`% change)
Now, maybe I am misinterpreting this, but it seems fairly self-explanitory to me. I think I will pass and just go on using MrRobot.
nieboh Oct 10th 2011 12:05PM
I perhaps should mention that I did the optimization on DPS Rating only. I don't care about survivability...that's my healer's worry. (just kidding) (only slightly)
bvcasey Oct 10th 2011 11:10PM
What class are you trying to optimise your gear on? I'm guessing mage since this is in a mage column, but if it isn't, I know some of the classes aren't implemented fully in the current version. If it is a mage, I'd suggest posting up an issue on their website.
Also, the genetic algorithm is stochastic in nature, there is no guarantee that it will find the global optimum everytime, and indeed it can produce results that are worse sometimes. The main reason GA is used is that it can produce very good results that are usually within a couple % of the optimum relatively quickly, particularly when you are dealing with problems with lots of variables such as the ones Rawr deals with. If you were to use a deterministic search algorithm that is guaranteed to find the global optimum eventually, you would likely have to wait longer than the end of time to find that solution using today's technology on these particular problems.
nieboh Oct 11th 2011 3:38AM
Yes, this is for a mage. I understand the reasoning behind finding a "good" result near optimum rather than searching for the true optimum, I just don't see what all the fuss is about with Rawr. As I said, AskMrRobot got me to a solution that was as near (actually closer in this case) than Rawr did, with a much more intuitive and user friendly interface.
I am not in any way trying to detract from the usefulness of Rawr, or to tell people who use Rawr successfully that they are wrong. I simply don't think it is for me at this stage.
Also, I do want to point out that others who may be inexperienced with Rawr and will be trying it for the first time on the basis of this artical that, as you say, the solutions obtained by using Rawr are not absolutes.
As an aside, I'm almost tempted to go with the Rawr result despite the overall loss in (theoretical) dps due to one factor alone. With MrRobot's optimization my current spell hit is 16.92%. With Rawr that would increase to 16.96% and I just can't shake the desire to cap hit at all costs.
bvcasey Oct 13th 2011 7:52AM
If I recall correctly, MrRobot bases its calculations on stat weights, which are basically values that represent the relative contribution to dps that each stat provides compared to the main power stat (agi, str, int). They are calculated for a certain level of gear, so the further you are from that level of gear, the worse the solution is to the actual optimum. Also, I don't think it takes into account the effect haste can have on a rotation, especially for hunters.
Rawr on the other hand, is based on a model that actually considers the actual rotation of the class/spec you are playing and calculates the frequencies and damage of the abilities you will use based on your current stats. The dps that it calculates should be closer to the actual maximum dps that your gear will produce than the dps calculated by MrRobot
inthemeadow Oct 12th 2011 2:10PM
I've been using Rawr for a long time now. It has become especially more useful to me for reforging and hit cap juggling with each new piece of gear I obtain. I like the in game addon as well that automatically factors in pieces I keep in my bags and bank. Rawr lets me feel like I've done everything I can do to maximize my potential without being a theorycrafter myself. Keep up the good work Kavan and friends, I appreciate it!