Monday, September 21, 2009

Haste Rating Is Better Than I Thought

Edit: I changed to wording in my Gemming section a little bit, to better indicate my meaning.

Ok, I messed up.

When I recently redid my gear list I used a valuation for Haste Rating that I came up with several months ago, without redoing the analysis. I was lazy. I felt confident in the number, because logically lag should devalue Haste Rating a little more significantly then the other stats. Also, my numbers were a little more extreme then those found by SimCraft, but they agreed to an extent. So, I ran with it.

Well, because of reader questions, I decided to take another look at SimCraft and rerun my model to see how things went.

My New Valuation:

Here are my new values:






















Stat12.5% Haste With Lag10.5% Haste With Lag
Hit Rating1.8211.721
Spell Power11
Haste Rating0.8161.034
Crit Rating0.7820.795


When I ran the SimCraft numbers, the results were very similar. So, I feel very confident in this valuation. However, I do want make one exception.

WiseEclipse - I haven't modified my model to include this addon, but logically it should devalue Haste Rating. Lets think about what WiseEclipse does. Basically it makes it a lot easier to proc Solar Eclipse with Starfire. If your proccing Solar Eclipse quicker then you will be casting Starfire less often and Wrath More often. Since Haste has a more significant impact on Starfire then it does Wrath, one can conclude that anything that reduces your Starfire usage has the potential to reduce your value of haste. How much I don't know.

Why Was I So Wrong?

In short, I don't know for sure.

It has been a while since I made significant modifications to the guts of my model, but it is possible that I made some changes a couple of months ago that I don't remember now. If I had criting Moonfire ticks procing Natures Grace that might have had an impact, but that seems a little extreme.

One thing I know I did change is the starting stats in the model. In my first run I was using 14% Haste from gear, about 3300 Spell Power, and a little less crit chance. When I ran it a second time I matched it more closely to my in game experience. I upped my Spell Power to 3450. The crit chance was increased about 1%, and the haste from gear was lowered to 12.5%.

I didn't think this would affect the valuation that much, but it could have. Haste is a very sensitive stat when it comes to scaling. If you have to much of it then the value of your other stats will shoot up. If you have to little of it the value of Haste Rating will shoot up. If I had to guess that is what happened here.

The Gear Guides:

I have updated both of the gear guides to compensate for this change. I apologize to everyone that made a choice based upon my recommendations that now turns out to be less then optimal. I guess one good thing is that My best in slot choices stayed relatively unchanged.

Gemming:

First things first, Spell Power is still the primary thing you should gem for. I know that my numbers say that Haste is a little better then Spell Power if you're below the haste cap, but don't read to much into that number. It assumes a level of Spell Power with a lot of Spell Power gems your gear. If you replaced some of that Spell Power with Haste rating it would throw off the balance, and probably make Spell Power more valuable. Also, my values are not absolute. When two values are close it means that the two stats are close in value. It does not necessarily mean that one is better then the other.

That said, I know one question will come up at some point. "Should I Gem for Crit Rating or Haste Rating?

To be completely honest I'm not sure at this point. It could really go either way. Haste generally scales better then Crit Rating, but it is very important to not go to extremes with either stat. As I mentioned before WiseEclipse probably devalues it a little since it limits the amount of Starfire casting you will do. I also think that the 2T8 set bonus devalues Crit Rating a little, but that is in dispute.

I'm going to be very vague and suggest that you try and keep the two in balance. If you think your Crit Chance is a bit low and you Haste a bit high then focus more on Crit Rating. If the reverse is true then focus more on Haste Rating.

What I really need to do is figure out how to make SimCraft use my stats and post a tutorial on the blog so that you can find out what your own valuations are.

11 comments:

lissanna said...

I'm ending up with about 500 haste rating in my moonkin gear with ToC gear (that seems to only have haste on it, GRRR). I think the abundance of good pieces of gear that have haste on them at higher item levels means that we end up having to gear with a lot of haste (most of the time, it's either spirit or haste that you are choosing between, and spirit is likely worse). I think the lowest I can get my haste rating without gimping other stats seems to be about 450-ish haste.

Hamlet said...

Overall, those are pretty similar to my values, except that I have haste rating as bring weaker past the softcap. For people who are over the haste softcap but under the Lunar crit cap, I see crit as being distinctly stronger.

---

WiseEclipse doesn't mess with things too much; it makes crit a little better. Overall, it's less of a DPS increase than most people assume, I think, and doesn't rework the stat table all that much.

Macc said...

Thanks for this post! I have been gemming for spell-power/crit rather than spell-power/haste. My current haste is at 550, with my crit at roughly 30% and 34% with Idol of Lunar Fury proc'd. I'm also at 2474 spell-power unbuffed, out of moonkin form. Do you think I should continue gemming the way I am or change the gems up a bit?

Maestro said...

I've been waiting for this post for awhile now. I had a feeling we were underestimating haste but I just haven't had the time to squeak out some numbers because I'm working on some glyph modeling.

@ Lissanna
I wouldn't worry about being over 400 haste too much. It's not ideal, but it generally requires a good deal of sacrifice to drop haste values. With the new spiritless itemization, its not so easy to pinpoint your haste value and still find significant upgrades. As Graylo mentioned, his BiS did not change very much and this is exactly the kind of behavior that I would expect.

Anonymous said...

random question to do with haste, seeing as 3.2.2 has just came out i didnt change any gear but b4 3.2.2 i had 440 haste and now i have 436 i know its only 4 haste but i want to know where it disappeared to if anyone knows lol

thanks

Erdluf said...

I'm not sure that 2t8 devalues crit that much (at least before the lunar cap).

It increases your average crit (which favors haste) but it strongly values Eclipse uptime (which favors crit). My copy of Hamlet's spreadsheet shows the crit/haste valuation going up (slightly) when adding 2t8.

Graylo said...

@Hamlet

When ever I make a post like this I find that some people go to extremes. They see that I say Haste is a tiny bit better then Crit and then immediately go with the conclusion that Haste = Good, Crit = Bad.

I know that WiseEclipse won't move the valuation a lot, but logically it will devalue Haste a little. I am just trying to remind people that there are other impacts out there that they need to think about and shouldn't move to extremes unless the evidence is clear.

@Erdluf

Similar to what I said to Hamlet, I don't think 2T8 has a huge impact on the valuation of crit, but an addtional 15% crit to Lunar Eclipse does devalue crit rating. Of course this is offset by the 15% increase to damage provided to Solar. Really how you value Crit Rating in regards to Eclipse probably depends a lot on your starting stats. Yours might show a slight increase. Mine might show a slight decrease. In my opinion those results are very similar and indicate that balance is in order. Again, what I'm trying to prevent by providing exceptions is to prevent people from moving to extremes.

Talsh said...

Thanks for the update - this is a much more intuitive stat valuation, at least from my perspective. Does this change your thoughts on swapping to 4T9.245? Or is it still optimal to wait until 4T9.258?

QuestionC said...

Simcraft has gotten pretty easy to use now. Here's a little tutorial. First you need to make a simple 1-line simcraft file:

Grayolo_armory.simcraft
armory=us,garona,graylo save=Grayolo.simcraft

This simcraft file will pull your character's data from the armory. The 'save=' option is not mandatory, I'll explain it later.


Drag and drop Grayolo_armory.simcraft on to SIMCRAFT.BAT. This will execute the batch file using Grayolo_armory.simcraft as the input file.

This will create Grayolo_armory.txt and Grayolo_armory.html output explaining your character's sim. This will also create Grayolo.simcraft which contains your character's essential simcraft data (stats, talents, and rotation) pulled from armory. Grayolo.simcraft and Grayolo_armory.simcraft are both valid simcraft files and can be used interchangeably unless your armory profile changes. Grayolo.simcraft generation is caused by the 'save=Grayolo.simcraft' option.

SIMCRAFT.BAT does not generate scale factors. For that we will use SCALE_FACTORS.BAT

Drag and drop Grayolo_armory.simcraft on to SCALE_FACTORS.BAT. This will take much longer, about 140x as long as running SIMCRAFT.BAT. SCALE_FACTORS.BAT will also output Grayolo_armory.txt and Grayolo_armory.html which contain scale factors.

How Simcraft calculates scale factors.
Let's say we are calculating the scale factor for haste. Simcraft first calculates your expected DPS. Simcraft will then add 150 haste rating to your character, simulate again. Simcraft takes the difference between the two, and divides by 150. The result is your scale factor. For Hit, the default delta is -100. For all other caster stats, the default delta is +150.

Options - I put the scale factor options first, other options later.
center_scale_delta=1 - Setting this will cause the scale calculation to be based on the difference between your character's DPS with haste+75 and your character's DPS with haste-75 (as opposed to haste+150 vs standard). This makes the sim take twice as long, but the results are less skewed by diminishing returns.

scale_haste_rating=50 - Changes the delta simcraft uses for calculating haste from 150 to 50. A smaller scale factor will increase the scale's sensitivity to variance, but will reduce inaccuracy caused by diminishing returns on the stat. 150 is IMO overkill for scale factors unless you have turned down the number of iterations. The similar scale options are scale_intellect=, scale_spirit=, scale_spell_power=, scale_hit_rating=, scale_crit_rating=.

scale_lag=1 - Simcraft takes lag into account for its simulation, but does not by default generate a scale factor for lag. This will enable scale factor generation for lag. This is a boolean option, it is not the same as scale_haste_rating.

iterations=20000 - You can speed up your simulations by turning down the number of iterations, at the cost of increased variance. This is set in the .BAT files, so you have to edit it there.

wowhead=14311012 - Instead of armory=, you can make simcraft look up a wowhead profile. In this case, 'http://www.wowhead.com/?profile=14311012'. You have to Save your profile in wowhead to get a profile #. This option is useful because you can change talents and gear in your wowhead profile.

optimal_raid=1 - Makes simcraft assume you have full raid buffs. This is set in the .BAT files, so you have to edit it there. This is on by default.

threads=2 - Simcraft plays nice with multi-core processors. If you have a quad-core processor, you could double the speed of your simulation by setting 'threads=4' SCALE_FACTORS.BAT. This is set in the .BAT files, so you have to edit it there.

Anonymous said...

Just curious, but do you take into account moonfire dot now critting into your statweight?

Thanks for your blog I'm an occasional visitor :)

Whisperdoom (Lightbringer) said...

Lol and I was chewed out by many including Graylo for promoting haste over crit... Makes me think I should write a blog on moonkin someday. Too bad I hate number crunching.