Latest in Gaming

Image credit:

The state of DPS in early Dragon Soul

Brian Wood

Patch 4.3 is well under way with all the buffs and nerfs that came with it -- including the extra 10% AP for melee classes from the raid buff improvement on top of all the class-specific changes. We've now had enough guilds do enough clears of Dragon Soul that we can begin to take a look at the DPS numbers that the classes are putting out and see just where they stand in 4.3. This is also a nice time to take a snapshot of the data, because the rogues do not yet have their legendary weapons, though of course more casters than ever have the 4.2 legendary.

As always, we draw our data for the analysis from the wonderful, which in turn pulls its data from World of Logs, where raiders upload parses of their actual raids. This gives us the absolute best data source that we have access to. These numbers are exactly what thousands and thousands of real raiders are actually capable of putting out in the actual raid, averaged over all of the bosses this tier.

So did patch 4.3 bring about a new state of improved balance to the game, or are we going to see more chaos in the DPS spread? The only way to find out is to sort through all the data, and then make pretty graphs out of them.

Gallery: Early Dragon Soul DPS Analysis | 8 Photos

About the data

Before we dig into the graphs, we have to discuss the methodology for data collection and make sure we understand where the numbers are coming from and what the advantages and disadvantages of these sources are.

All of the data here are coming from the median DPS of the top 100 parses, or in one case, the median DPS of all parses. I prefer to look at the median rather than average for DPS parses, because it's pretty common to have outliers up on top that really skew the average. For any given data set (25-man heroic Ultraxion, 10-man normal, and so on), I'm taking the median DPS of the top 100 parses for each spec. Once I have all of those, I then calculate the median DPS of all specs. Then each spec is charted as its percentage above or below that median.

This gives us a really easy picture of how the classes are performing against each other and how far away they are from where they should be (at the median).

Advantages of top 100 parses

When we look at the median of the top 100 parses, we know that we're seeing players who have very good gear, are very skilled, have all raid buffs, and are also enjoying good RNG, to boot. It's a good representation of how much DPS a spec is capable of putting out in actual in-game content.

Looking at the top 100 parses has three big advantages:

  • We're comparing apples to apples. We're looking at players with similar gear, buffs, skills, and luck levels across all classes -- specifically, very high levels of each.
  • In some fights, certain classes have odd jobs like kiting, interacting with mechanics, etc. With the top 100, we can be pretty certain we're looking only at examples of players who were not doing anything to hamper their DPS. These are the lucky guys who didn't have to kite or pull the lever.
  • Because the top 100 is, in fact, the best performance players have produced for their class, we are looking at a true potential of the class in real content -- much more accurate than any spreadsheet or simulation. In general, class DPS is balanced around the potential (after all, if you're doing substantially less, you don't need a buff -- you just need better skills, gear or buffs). Don't get me wrong, average DPS matters too, but that's more about balancing spec difficulty and scaling than DPS output.
Comparing the classes rather than specs

The primary disadvantage of looking at the top 100 parses is that it disproportionately favors the best DPS spec of any given class. The very best players in the very best gear are typically going to play the very best DPS spec for the fight. If there is a spec lagging behind, far fewer of the best players are playing that spec, and so its numbers fall even further behind, giving a skewed picture.

I'm doing something slightly different this time than I did for the Firelands DPS analysis, based on a reader comment. Rather than comparing all the specs in the game, we're instead mostly going to look at just the best spec of each class and use that to compare the classes against each other.

After all, I don't care if frost mages are doing 400 DPS and arcane is doing 300 DPS; if fire mages are doing 50,000 DPS, then we're all going to say that mages are overpowered. And hey, as much as the arcane mages might complain, they can just respec to do crazy damage. And in reality, that's just what happens.

It's more important that each class has at least one spec that's competitive than it is that all of their specs are competitive (though we surely want that, too). And a class that might look like its DPS is in the middle of the pack when you look at all specs could turn out to be dead last when you compare it just against the best spec of each class.

Of course, druids and shaman don't have the luxury of just respeccing to the best DPS spec -- they require entirely different gear sets, so we'll be looking at both shaman and druid DPS specs, and then just the best spec of each other class.

What is the goal?

One thing to consider before you look at the graphs is exactly what the design goal of the DPS spread should be. Given that every boss has different mechanics that can skew the balance in favor of one class or another, exactly how much deviation is acceptable?

If you draw a line down the middle of the DPS and take the median, how far off of that median is OK, and how far off means that a buff or nerf is needed? Is 10% acceptable, or is that too far? What about 5% off? How close is it realistic to expect the actual raid numbers to be? If a class is 2% below the median, does it need a buff? If it's 2% over, does it need a nerf?

I don't have a good answer here, but what I can say is that the spread so far for Dragon Soul is much, much narrower than it was for Firelands. Let's take a look at the actual numbers in part 2.

From around the web

ear iconeye icontext filevr