Diablo® III

CM/WW DPS Simulator (updated website)

aww my bad, thought i uncheck MW/BM. But it should be 116 w/ scoudrel and 112 w/o him.
Reply Quote
Ok, so it loads your crit chance as 54% (that's the value the API gives in the entry for your overall stats), and your profile page says your gear has +46% crit chance, add the 5% base and that's 51%, so it must be including the scoundrel bonus in the crit chance it's giving. So you'd have to change the crit chance value or uncheck the scoundrel crit bonus. Maybe later I'll change it to load the crit chance from the sum of all the items instead of going with what the API gives as the overall crit chance so that doesn't happen.
Edited by Steve#1523 on 1/22/2013 4:05 AM PST
Reply Quote
Really interesting site this, but something seems off.
If I'm hitting 1 target it says I'd be doing 1.27mil dps, which is about right since I do about 1.35 from testing.

But then if I'm hitting 2 targets it says I'm doing 2.96m dps, which just seems like its alot too much. Haven't done any testing on the matter, but it definitely feels more like 1.8ish mil.

EDIT: Ahhh.. I thought it meant my DPS per target, but if 2.96 is my overall DPS split between both targets then it definitely seems right and makes sense. Awesome site btw, thanks!
Edited by Niko#2983 on 1/22/2013 8:56 AM PST
Reply Quote
Well I haven't done any testing against more than 1 target, because I have no idea how that would be possible, but if you mean fighting 1 target compared to fighting 2 targets and hitting both with each WW cast, it makes sense that it would be somewhat more than double the 1 target DPS, because you're getting a lot more CM procs which gives you a lot more Chain Reaction and Diamond Shards casts. It would be close to exactly double (slightly less because of Storm Armor base spell damage and getting slightly less WW casts in because of the extra Frost Nova casts) if you were hitting both with WW and casting an equal amount of Chain Reaction and Diamond Shards as you do vs 1 target.
Edited by Steve#1523 on 1/22/2013 7:02 AM PST
Reply Quote
Well, I've never read definitively where Mag Weapon is better than timewarp in EVERY aspect (I thought there were benefits to both), but it seems this tool puts that question to rest. I guess the benefit of the bubble has always been in my head (or I'm doing something wrong here).

Mag Weapon:
Average freeze uptime = 73.03%
Average seconds per Frost Nova cast = 1.42 seconds
Average Frost Nova casts per fight = 21.14
Average seconds per Diamond Skin cast = 1.38 seconds
Average Wicked Wind casts per second = 1.82
Average active twisters = 9.57
Average Chain Reaction casts per second = 1.65
Average global cooldown uptime = 92.24%
Average CM procs per second = 8.56 (91.39% from WW)
Average APoC gain per second = 94.70
Average AP spent per second = 96.67
Average Arcane Power = 74.64 (out of max 113)
Average life gain per second = 11220.79 (74.62% from LoH)

Time Warp:
Average freeze uptime = 69.56%
Average seconds per Frost Nova cast = 1.52 seconds
Average Frost Nova casts per fight = 19.73
Average seconds per Diamond Skin cast = 1.47 seconds
Average seconds per Slow Time cast = 5.22 seconds
Average Wicked Wind casts per second = 1.69
Average active twisters = 8.91
Average Chain Reaction casts per second = 1.55
Average global cooldown uptime = 92.32%
Average CM procs per second = 7.96 (91.34% from WW)
Average APoC gain per second = 88.12
Average AP spent per second = 90.10
Average Arcane Power = 74.88 (out of max 113)
Average life gain per second = 7783.31 (100.00% from LoH)
Reply Quote
I was a little surprised too, I was thinking "eh the cast animation's not a big deal if it's just every 5-6 seconds", but after thinking about it, that's using up like 6-8% of your total cast time depending on APS. I've only looked at a few different configs for this but Time Warp is significantly worse than Force Weapon in all of them.

I did a set of test runs for one Time Warp configuration after adding the skill to the program, the simulation matches the test data very closely:
http://d3cmww.com/cmww?configid=54592&testid=54359
http://d3cmww.com/test_results

The spam_slow_time option is on the site also, to simulate spamming Slow Time constantly along with the other skills rather than waiting a while to recast it, and that lowers the DPS a lot more. So definitely don't do that.
Edited by Steve#1523 on 1/22/2013 8:05 AM PST
Reply Quote
is it clojur? looking at the source code, i think it can only it be
Reply Quote
01/22/2013 08:04 AMPosted by Orion
is it clojur? looking at the source code, i think it can only it be


No it's Common Lisp. I've never used Clojure but seeing as they both use Lisp syntax they probably look pretty similar.
Reply Quote
I was a little surprised too, I was thinking "eh the cast animation's not a big deal if it's just every 5-6 seconds", but after thinking about it, that's using up like 6-8% of your total cast time depending on APS. I've only looked at a few different configs for this but Time Warp is significantly worse than Force Weapon in all of them.

I did a set of test runs for one Time Warp configuration after adding the skill to the program, the simulation matches the test data very closely:
http://d3cmww.com/cmww?configid=54592&testid=54359
http://d3cmww.com/test_results

The spam_slow_time option is on the site also, to simulate spamming Slow Time constantly along with the other skills rather than waiting a while to recast it, and that lowers the DPS a lot more. So definitely don't do that.


Not to mention the Life gain benefits. ~3500 lps boost... nuff said. I'm going to test it, but I think I'm already a mag wep convert.
Edited by Jobewan#1365 on 1/22/2013 8:34 AM PST
Reply Quote
+1 for not supporting IE, people need to get out of 1999.
-1 (for your troubles, not the actual site) for using Lisp. Looks like it worked out, but man do I feel for anyone that has to program in Lisp. (not supposed to be an actual insult, just kidding around :P ).

overall great site, very functional, overall great job. Props!
Edited by Kieble#1634 on 1/22/2013 8:46 AM PST
Reply Quote
When i try to load my character it says failed. So i was wondering if i have to put all of the info in manually?

Never mind I got it spelled the name wrong lol
Edited by Joel#1120 on 1/22/2013 8:57 AM PST
Reply Quote
would love a conflag/firewalkers/stormcrow option , and all the meteor variants.

storm crow proc + conflag better than evocation?
storm crow proc + conflag + molten inpact > than eb?

of course, its up to you.. just nice to have.
Reply Quote
^ conflag/molten +1
Reply Quote
Better yet... Why not an "easy" button... You load your profile, click the easy button, and all of the variant build stats are graphed via a color chart. :D
Edited by Jobewan#1365 on 1/22/2013 10:53 AM PST
Reply Quote
01/21/2013 04:41 PMPosted by Steve
Also, what spell priority do you use?


It's slightly complicated. I had to mess around with it a lot while trying to get all the results to match the test data.

It sends all the skill commands at once in a single frame, and then sets a variable that represents the next frame in which commands will be sent, with the frame delay until the next command send frame generated with a simple random formula that has an average delay equal to the value of the avg_command_interval parameter.

When it sends commands, it takes the static list of all skills and generates a new list with the order of the skills randomized, and then goes from the start to the end of that random sequence and attempts to cast each skill (all of this taking place in a single frame). Even though they're all sent in the same frame the order of the cast attempts still matters a lot, casting one skill first can make it unable to cast a later one, because of the global cooldown (cast delay) or not having enough AP.

That's the basic framework, and within that there's just one more thing it does, which I posted earlier:

So I put in code for a random chance when sending a WW command to send a Frost Nova command first, with probability equal to (2.35/APS)^4.0


I don't know what it is but there's something weird about how the game processes spammed commands, to get simulation results to match up with testing you have to put something in that heavily favors Frost Nova being cast rather than Energy Twister when it's possible to cast both.


That seems overly complicated. Does that mean you can only use 1 skill at a time, or does it enable you to cast WW+DS+EB all at once? If i'm understanding you right, it randomly goes through each ability and casts them if possible. If the first picked is FN but you can't cast it, do you still cast WW or do you assume the FN triggers a white attack?

Have you tried using a basic pririoty like FN > WW > white attack, then EB+DS if available? It seems a lot simpler than picking a random skill to cast, but you do make a good point that the game seems to act really oddly when you spam skills.

There definitely seems to be some large differences between your simulator and mine, though the dps results seem fairly similar. Also, does your simulator start the fight from scratch, or are you just doing a full windup calculation, i.e., assuming you're already in full windup at time = 0?

Finally (I know, that's a lot of questions), when you use CM and ApoC coefficients of 1, I know you mentioned that the FN casts don't simulate correctly, but does your dps still end up being approximately correct?
Reply Quote
That seems overly complicated. Does that mean you can only use 1 skill at a time, or does it enable you to cast WW+DS+EB all at once? If i'm understanding you right, it randomly goes through each ability and casts them if possible. If the first picked is FN but you can't cast it, do you still cast WW or do you assume the FN triggers a white attack?


The only one I have that triggers a white attack is WW, and only when it fails due to insufficient AP (the only other way it can fail is if the global cooldown is active, which prevents the white attack also). I didn't try to find out if any of the other skills can trigger white attack.

It casts multiple skills in one frame if it can, so for EB and DS the order doesn't usually matter, except that if WW is before EB and AP is low at the time then after the WW cast succeeds and spends the AP there might not be enough left to cast EB... and that could happen in reverse, EB before WW, EB uses up the AP and causes WW to fail and trigger white attack, though during the approximately 95% of the time that the global cooldown is active the WW cast would fail anyway.

WW and FN can't be cast in the same frame because the first one immediately triggers the global cooldown which causes the second one to fail.

Have you tried using a basic pririoty like FN > WW > white attack, then EB+DS if available? It seems a lot simpler than picking a random skill to cast, but you do make a good point that the game seems to act really oddly when you spam skills.


I'm pretty sure that wouldn't work, I actually did try always attempting FN before attempting WW (currently it does always do that for APS<=2.35), but at higher attack speeds it gave far too many FN casts at the expensve of WW casts. I have no idea why it changes that way with attack speed.

Also, in configurations where available AP is a limiting factor (such as http://d3cmww.com/cmww?configid=54433&testid=54359), the relative amounts of WW and EB casts can have a large effect on the results, and I think setting either of them to always be attempted first would skew the balance too far one way.

I think from a standpoint of trying to match how skill spam works in the game, it makes more sense to attempt casts in a random order, at least with the AutoHotkey script I was using that's sending the commands way faster than the client can process them.

But mostly there's not much of a logical explanation for why it should work this way instead of some other way, the command spam algorithm and the proc parameters were the two things I had any freedom to change without violating game mechanics, and there's more freedom in what I could do with the command spam, so I tried a whole lot of variations on a few different fundamental algorithms, and this was the one that ended up matching up with a large number of test configurations simultaneously.

Also, does your simulator start the fight from scratch, or are you just doing a full windup calculation, i.e., assuming you're already in full windup at time = 0?


Starts from scratch, full AP no twisters.

Finally (I know, that's a lot of questions), when you use CM and ApoC coefficients of 1, I know you mentioned that the FN casts don't simulate correctly, but does your dps still end up being approximately correct?


http://d3cmww.com/test_results?testid=54362 (I think that page should work even in IE) is the closest I can get leaving the CM and APoC proc multipliers set to 1.0, with 2.5 for SA and a completely unreasonable speed of sending/processing skill commands. The DPS isn't off by nearly as much as the FN casts but it's still pretty bad. And if I just take the set of parameters that I'm actually using on the site and change the CM and APoC multipliers to 1.0, it's a lot worse: http://d3cmww.com/test_results?testid=54363

I'd be surprised if you can be getting the DPS high enough with the standard proc rates for CM and APoC, the 1.15 APoC multiplier adds a lot of DPS especially in my low APoC test configurations, and the lower CM procs will mean much less CR casts
Reply Quote
Thanks for the answers. I appreciate the time taken to respond.

You do have some interesting points about the random nature of spammage. That is definitely not very easy to model. I also agree that the priority system simulated can have a large impact on sim results, especially at lower APS since you are limited quite a bit by AP.

I have my simulator priority set up as I described above, and it seems to work reasonably well at higher attack speeds. I only have a couple vids left from when I did some soloing of the uber bosses, so I looked over a 55s clip at the start of the Rak/Ghom fight for comparison. My simulator with evo and cold snap, 20 APoC, 52% crit, 2.7338 APS, 100ms latency/delay, says I should cast approximately 46-47 frost novas over that period of time, since it's against 2 targets. I actually was able to cast 50-51 over the 55-56s to start the fight. The extra FNs definitely hurt my dps though. I was expected to have an effective multiplier around 11.5 but it was more like 9.11. I should have gone through again and counted the DS and EB casts, but I didn't think about it before posting this. Then again, I haven't actually tested my code for multiple mobs, and it looks like there's something strange going on with my LoH gains, so who knows.

Regarding my simulator dps results, I can get pretty reasonable numbers when not using SA, for various builds, with the right latency. I tested a few cases that BDF posted on the diablofans forums and the results lined up remarkably well. The general problem is I really don't think SA follows the 2.5x proc model very well, or the model I use where the extra proc rate is inversely related to attack speed. My guess is there's a lot of error introduced by the SA model, maybe 10-20%. I just don't have time to test the model more, or the gear to test higher APS to come up with a better model. The 2.5 just kind of puts it in the ballpark for the most part.
Reply Quote
Haven't test it out yet but it looks like a nice job.
I have done some interesting evil things as well. Add me to discuss.
Reply Quote
Hey guys. Can anyone explain this:
http://www.youtube.com/watch?v=VRpaM1PA_z8
The video is apparently a bit older, here's the current profile:
http://eu.battle.net/d3/de/profile/Qiller-2507/hero/333382
This is a simulator config, with his current items and the skills taken from the video:
http://d3cmww.com/cmww?configid=54696

In the video he runs at around 8x DPS. Without shards and just above the 1.7648 APS / 23 ticks breakpoint. Freeze uptime is really low, as expected. But damage is much higher than the simulator says (which is configured with better gear than in the video). And much higher than what I would have predicted.
How is this possible? What am I missing?
Reply Quote
01/22/2013 02:12 PMPosted by Loroese
My guess is there's a lot of error introduced by the SA model, maybe 10-20%.


Just about every time I was doing test runs for a config that included Shocking Aspect, I ended up having to go to 20-30 test runs before I felt like I had a stable average. The DPS for a single run has extremely high variance that never happens without SA. I'm not sure why that should be the case based on the spell description.
Reply Quote

Please report any Code of Conduct violations, including:

Threats of violence. We take these seriously and will alert the proper authorities.

Posts containing personal information about other players. This includes physical addresses, e-mail addresses, phone numbers, and inappropriate photos and/or videos.

Harassing or discriminatory language. This will not be tolerated.

Forums Code of Conduct

Report Post # written by

Reason
Explain (256 characters max)
Submit Cancel

Reported!

[Close]