DiabloFans

Disclaimer: This post is full of statistics and hypothetical data analysis of hypothetical data, but it's all we have right now. If you don't care about stats you might just look at the pictures and draw your own conclusions, or just jump to the end for a TL;DR!

As you might've seen I recently updated the Difficulty Overview for the PTR:

http://d3resource.com/difficulties-ptr.png

The big change is obviously the one to GR keys. Now you get a guaranteed drop from Normal rifts, which completely changes all other values and makes us wonder about the new scaling. I've played a bit on PTR myself and watched a lot of videos to get some data, but the only data point that I can call relatively reliable is T13, with 2 keys guaranteed and a 25% chance for a third key. That gives me two data points:

* 1.00 keys on average on a Normal rift
* 2.25 keys on average on a T13 rift

For the 15 difficulties in between, we can now extrapolate. I took four approaches to that:

* Naive scaling: Basically increased the drop chance for 10% for each difficulty, and 5% at the beginning and end to make it fit.
* Likely scaling: This is based on the previous scaling for drop rates I've observed over 20 years of playing Diablo (drop rates are rarely linear but usually favor certain difficulties). For example, it rewards players playing the highest difficulty.
* Linear regression: Simply taking the first and last point and drawing a linear regression line. Note that monster HP increase takes a big step after T6 (since it changes from 3 GR levels per difficulty increase to 5 GR levels per difficulty increase), so this is sort of flawed, but previous drop rates didn't care about that either...
* Constant scaling: This is a formula based on the assumption that the key drop rate increases as much as the monster HP, keeping a constant reward structure for switching to a higher difficulty.

The result looks as follows:




Now, while we don't know any specifics about the GR key drop scaling, we know exactly the HP scaling (since it matches to GR scaling numbers and I've personally confirmed this on PTR to be correct). What we can do now is plot each of those GR key drop rates with regard to the monster HP increase. This is what it looks like (I've started at T2 because the pre-Torment scaling is odd for some data, so T1 -> T2 is the first increase I look at):



As you can see, the constant scaling is constant. The "likely" scaling is Blizzard's usual policy and would reward players to play on T13. The initial scaling I assumed as well as a linear regression would mean: stay the hell away from T13. You can also observe the drop from T6 to T7. Now, before you tell me "Blizzard would not overlook this T7 drop", look at the same graph for 2.4.1 on live servers:



If you're purely after keys, doing T7 rifts instead of T6 rifts is among the stupidest things you can do. It has actually been like this for three entire seasons. You can also see why T10 rifts are sort of worth it, and T9 is not. Basically it's T6 > T10 > T8. (Further down I present evidence for the T7 gap still being the case in 2.4.2.)

Now, that relative increase still doesn't tell me as much, as it's hard to measure a 2% or 4% increase between difficulties on your character. Instead, what we can do is estimate the number of rifts per hour. In the TX challenge, the fastest solo clear is 23:13 minutes for 10 TX rifts (http://us.battle.net/d3/en/forum/topic/20418543014). That's obviously not representative, but it's also done solo, so I assume that fast groups of paragon 1000+ players can clear about 20 TX rifts per hour, resulting in 30 keys per hour.

Therefore, let's take a look at how many rifts you have to clear on each difficulty to get 30 keys:



You can clearly see the major difference: the reward structure has become much more "flat". Currently, you have twice the time to finish a T10 rift to get about the same number of keys as on T3. For well-geared character, that is possible. On PTR, even with the most "high difficulty rewarding" scaling, you'd only have 30% more time. Not impossible, but much less likely. This can be both a good and a bad change: Purely for getting keys, you might also have the option to run lower difficulties. That means you can run more builds. For a Crusader, it's Bombardment or bust; with my 6m sheet DPS full ancient Roland's Crusader I had an epic 3 minute fight against a yellow elite on T13 - I'm sure a perma-Steed Charge build speedrunning low Torments would've gotten me a few keys in that time.

This is reminiscent of the time when we had to run bounties to get Rift Key Fragments: While most people ran it on high difficulties to get more cache legendaries (which they didn't need), the fastest way to get those keys was actually running Normal difficulty. Unfortunately almost no one realized this, despite my efforts to even make a compeition out of it. Many, many, many players are completely going to screw themselves over though if they decide to run T13 for keys in 2.4.2. By all the data we have right now it clearly will not be the most efficient to get keys.

But wait, there's more than keys! What about the other stuff? So I took a look at other things you get from rifts - gold (important for empowering, and potentially even a rare commodity with the huge nerf in gold rewards); legendaries; and last but not least Death's Breath.

Again, both in 2.4.2 and in 2.4.1:





Since the numbers for monster HP and gold are *not* guesstimates but actually correct, you can clearly see how the T7 gap is still a thing in 2.4.2. The gold amount is relatively stable so will probably offset with faster rifts when you go on lower difficulties. We don't have accurate numbers for DB drops yet; but someone who wants to make a case for higher difficulties could argue that more Death's Breath and more legendaries make higher difficulties more rewarding (especially considering that GRs don't drop any Death's Breath). Once you're at a point where the probability of gear upgrade is minimal, or you actually don't *want* an upgrade (because you've already augmented everything), it's safe to say that going lower is better though.

So with that... let's get to the summary.

TL;DR

* The new GR key drop scaling in 2.4.2 means you should definitely wait longer before going up a difficulty.
* There's still a gap between T6 and T7 for some drops (especially gold); stay on T6 for a bit longer and try to jump to T8 or T10 directly. Avoid T7 and T9.
* By far the main reason to go higher is legendaries+gold+Death's Breath, not GR keys.
* Staying on lower difficulties (and enjoying the higher build variety) is a viable option.
* When you augmented your "final" gear and are not looking for upgrades anymore, you can't go low enough. Maybe we'll even see insanely fast passively-killing Normal builds?



That's it! As usual, to draw any more conclusion I NEED MORE DATA. Unfortunately absolutely *no one* has provided me any data of their PTR runs. If you run any rifts on PTR, any difficulty, I'd be so happy for your input. Even if it's just like "hey, I did 5 rifts on T7 on PTR yesterday, got 1 key twice and 2 keys three times". If 10 people do that, we'd have a good sample to figure out which curve is the right one.

Quote from AtACarnivorus»

Very interesting analysis. I did overlook this thread indeed and only stumbled upon from the 'new' T13 thread.

I started testing on T13 yesterday and will continue to do so the next days, so you can expect some data from my runs! I'll gladly help figuring out the right curve

Thank you very much! That would be lovely. I ran quite a bit of PTR lately, even bounties to figure out the cache legendary drop rate (not 100% sure but quite sure it's 2 per cache on T13). But lately not even any streamers are running PTR anymore, at least I haven't seen anyone, so no data coming in.

Also did quite a few test runs on "undergeared characters" (took mediocre rolled non-ancient set pieces mixed with some crappy legs to simulate season start) and I really wonder if it's worth it to go above T6 on the first few days. You just want keys and run as many GRs as possible... so I guess only raise it up if the difficulty becomes "boring", but unless you're paragon 1000+ and have GG gear I don't think that'll make you hit T13.

Hm, not really @anything specific. The most consolidated information I have on T13 droprates (keys, DBs, organs I'm 95% confident in their accuracy). T13 Keywardens I haven't tested much but I don't think anyone is too fuzzed about this either. I think what would be most interesting is to get any solid data on any mid-Torment drop rate (especially keys) since the entire scale between Normal and T13 is just extrapolated, and every data point in the middle would help us to shape the curve. As you can see in my models the T6 key drop rate ranges from 33% to 80% for a second key. I actually think from my runs it felt even lower, think I almost always got 1. So even things like "I did 10x T6 runs and got 13, 14, 15, 16, 17, 18 keys" would already be helpful data!

But first and foremost just "play your way" (ugh) and if you can just keep track of GR key drop rates and on which level that'd be awesome data already.

Thanks to both of you, you're awesome!!!!

Here's the result so far when I compare this data to the extrapolation used:

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

It looks like NickFudu got a bit lucky whereas AtACarnivorus had a bit low drop rate. Overall the trendline suggests that the mid-Torment drop rates are a bit higher, though that does not really align with my own data (which rather suggests it's lower). In any case, the spread of the data shows how difficult it is to come to a conclusion. However, the great thing about your tests is that it shows: pure RNG will be more relevant for your first few dozen rifts than the exact drop rate. Until then, we can only collect data and further hope to converge to one of the trendlines!

Thanks again, really appreciate your testing.

Really? Wow that is strange... then I have to assume it's a flat 2 DBs on T11 and that would either mean that T10 changed as well or there's a huge increase from T10 to T11. I guess need to check that again... Thanks!

Quote from lolz4»

I think all the statistics concerning "optimal" db farming have to be taken with a grain of salt as it might be better to run some tiers lower but with the sage set yielding another db per drop. However it is always great to have more "raw data" from the game, so a big Thank You for that.

Oh absolutely 100% agree on that. Probably one of my hidden (or not so hidden) agendas with this analysis is to show that always running the highest difficulty is a bad idea. Not just because of efficiency, but it also dismisses so many builds that are potentially more fun. People who are looking for a T13 capable build in 2.4.2 at paragon 800 will struggle, be inefficient, and have only 1-2 builds at their disposal. At the same time, just staying on T6, experimenting with fun builds, you will probably get the same amount of stuff (and more keys) with much less frustration. :-)

The reason is hidden in the charts+analysis in the first post. There's a sudden increase in monster HP after T6 (because up to T6 every Torment increase is +3 GR levels equivalent, and after T6 it's +5 GRs). The drops however just continue to increase as before, so don't keep up with the increased difficulty. As I said in the first post, with keys even dropping 100% on Normal now and the entire curve being flattened, there's no reason to bump it up to T10. Yes, a paragon 2000 player in a cookie cutter build can probably clear T10 just as fast as T6. But 99% of players who play T10 would be much more efficient on T6 or lower (even now and especially in 2.4.2), plus having the benefit of being able to not play cookie cutter builds. For example, a DH could run with Danetta's and random speed burst items - but there's hardly any DH build that is really good enough for T13 below paragon 1000.

There are two reasons why people should go above T6: 1) When bored about running a lower difficulty and craving for a challenge (though I'd ague then you should just run GRs, which is the point of doing most efficient Torment rifts in the first place) and 2) when you're combining Rifts and bounties in the same game. That however requires people to be able to do split bounties on T10 which is not something anyone should be doing on the opening weekend.

Thanks. I see a trend...

It definitely looks as if the curve is much flatter, based on your mid-Torment drop data. That means that it's even less worth it to up the difficulty - on average your drop rates are about 10% higher on the mid-Torments then what I extrapolated.

I think T6 vs T7 would be my main focus. Why? Because "rift bounties" might be interesting on the first few days (3 people running bounties an 1 person doing a rift in the same game), but you won't be able to do this on T10 or even T13 on the opening weekend fast enough - so T7 would be the perfect level in terms of bounty mats. However, the gap between T6 and T7 contradicts this. But all others are interesting as well.

I'm hoping I have some time as well and might do a few runs as well, in particular to track Death's Breath as well. I wonder if they changed the scaling below T10 for that, otherwise the DB increase on T11+ is sick.

If you could track DBs for the first 10 elites or so and check if they roughly are in the correct ballpark, that might already be enough to hint at a change or not (T6 has 75% chance for a DB drop, T8 has 100% chance for one and 15% for a second). You could even just check how many DBs you have at the end of the rift, scroll up chat to see how many elites you killed (assuming you pick up every DB), add one for the RG (I think?) and that would be enough to give us an idea if they changed it or not. Because if they did, I assume T6 DB rate to be at 100% or close to it and not just 75%.

I'll try to run some T10, T11, T12 because something is happening there that's strange... for T13 I could always just review PTR videos, tons of stuff there which is quicker than running it myself ;-)

OMG don't go too crazy on this, seriously! You've already done so much, don't go out of your way. See if you have something to test (like some Sprinter speed run builds or T6 Season starter builds or so) that you can combine with Torment farming. Don't farm 10 hours of Torment rifts just for figuring out some decimal points ;-)

Quote from AtACarnivorus»

1st results:

• T6: 18 runs, 24 keys, 193 db 1 hour
• 1st 10: 14 keys (4 doubles), 114 db (1 odious collector) - 33 min
• 2nd 8: 8 keys (no double), 79 db - 27 min

Doing the same on T8 now incl. counting elites killed.

Edit: T8 finished:

• T8: 16 runs, 25 keys, 231 db 1 hour
• 1st 10: 15 keys (5 doubles), 147 db - 115 elites killed - 34:20 min
• 2nd 6: 10 keys (4 doubles), 84 db - 69 elites killed - 25:40 min

Got some double db drops and even 2-3 double db drops from RG on T8 here.

Okay that is some crazy data that I'd like your opinion on. The DB drop rate is pretty much exactly 1.25 per elite on T8 - currently in 2.4.1 it's 1.15, and 1.25 on T9. 184 elite killed are not a small sample anymore that is to be taken lightly - so if you say you're 100% confident in this data, I'm gonna adjust the DB drop rate. This would also explain the scaling on T11/T12, but I don't see the reason for this change. (Maybe DB are now 100% on T6 or T7 - used to be 75% and 90%, respectively.) However, you say that the RG also dropped DB - is that factored in in the DB/elite count?

Oh and regarding the key drop data, this is now slightly converging towards a 0.2-0.3% increase over my data. Which actually makes sense looking at it - for the mid Torments I assumed many 1.X8 drop rates, so I can just bump those up by 0.2. Also makes them look nicer.

Awesome stuff! Another thing - one hour for 16 rifts is insanely fast, and looking at the amount you got from it I'm not sure you would get more on T13. This hints at this as well:

Quote from Shodokan123»

Doing T13 even when you are fully augmented will likely not be an optimal use of time if you are aiming for pure efficiency.

Absolutely 100% this. I think rather than focusing on finding T13 builds, we should think about builds that optimize speed (and FUN) and just see which difficulty they work on. I'd say that if you're dying in a GR or if it takes more than 3 minutes on average to close it, go lower.