ATV Leveling Comparison - madBeavis/PimpMyAtv GitHub Wiki

Foreward

Not much hard data exists to quantify how well ATVs perform against each other, so I decided to at least make an attempt to quantify it. I didn't need anything super scientific, so I didn't go too overboard, just enough to reach clear conclusions.

The initial test was done in 2020, so xp per spin was 250. It is currently at 300xp per spin now, so one should be able to get to level 30 xp of 2 million in under 40 hours if done in an optimal manner. Do watch for events that increase the xp amount per spin and/or increase lucky egg time, so make accounts then.

For the comparison, I have the following ATVs:

  1. x96 mini
    1. s905w cpu (Quad-core Cortex-A53)
    2. GPU: ARM Mali-450MP5
    3. 2gb/16gb
    4. residing in my computer case build but disabled relay control
    5. ~$35 from amazon
  2. Tannix tx9s
    1. s912 cpu (Octa-core ARM Cortex-A53)
    2. GPU: ARM Mali-T820MP3
    3. 2gb/8gb
    4. in stock case, set vertically on the top of my computer case build to provide airflow through case to reduce cpu temps
    5. $31 from amazon
    6. obtainable at time of this writing
  3. U2C
    1. s912 cpu (Octa-core ARM Cortex-A53)
    2. GPU: ARM Mali-T820MP3
    3. 3gb/16gb
    4. in stock case, set on top of computer case build
    5. $40 from amazon when i found them
    6. seems to be discontinued

Setup

I run a separate instance of MAD for leveling with a separate database. Both my normal and leveling instances of MAD were run locally on an i7-8750h laptop with 16gb ram and 500gb WB Black NVME ssd. MAD was git pulled prior to start of the run. My intention was to not stop MAD during the runs or do anything that would slow things down.

ATV rom 1.2.7 (32 bit) with current PD 1.3.3.10 and RGC 1.10.1. Pogo was 175.2. As mentioned elsewhere, devices were provided with some cooling. Temps observed during the run were about 40c for the x96mini with the s912 variants between 45c and 55c. I did not run the add-on to speed up quests and leveling. I run NFS on all my ATVs. Minimal interaction with devices via scrcpy was done beyond checking xp.

Walker setup was fairly standard. It only takes 5 areas to reach 30 on my NYC setup. Routes are quick calculated (this was prior to ortools upgrade). I wanted to minimize backtracking for any missed stops, so i set each walkerarea to rounds of 1. You can find the fences and pokestop data on the code portion of this wiki.

2021 setup. I upgraded my pc to a mini pc that has a Ryzen 9 5900hx, 32gb ram and 1tb Samsung 980 ssd. The only atvs I using for leveling is my two remaining U2C s912 variants since they exist outside of my main setup and as such, not utilized as daily drivers. Since I am lazy, I am running 32 bit madrom on both. Still on a separate instance for MAD.

Leveling area

I was needing to make some new accounts for new devices, so I figured why not test devices them leveling up. Hard to compare devices in much other than stability when they reside in the suburban hell that mine spend most of their time.

So I setup 1x U2C and 3x tx9s to leveling in some of the well known leveling cities. I only run calculated route runs in London, Tokyo, Paris and New York. After a day, I checked on their progress and to my surprise, the tx9s in NYC was already at level 27. I did not do any lucky eggs to this point, but they were lasting for 1 hour each and you can stack them. I loaded up all the eggs and let the devices run until they reached 30 within 2 days.

Comparison of the s912

The second and third rounds were comparing actual device to device. I create the exact same walkers for NYC for the 4 devices and fired them off. The first round was won by the U2C and second round was won by a tx9s. Nothing decisive, less than a hundred stops difference in both cases.

s912 vs s905w

For the next test, I wanted to see how the standard s905w based x96 mini compared to the s912 variants. I created several more walkers for NYC, this time I was running 1x x96mini, 1x U2C and 6x tx9s. My other U2C ATVs hadn't arrived yet and I didn't to pull any more x96 mini off their normal jobs. For this test, I didn't pop any eggs to keep things cleaner.

I fired things up and let them run. I was coming off working nights, so sleep made more sense than times of even intervals, so it is what it is. The tables for timed data are at the end of this page. As you can see though, the s912 variants are close to each other but the s905w lags behind.

I shutdown the test after 48hrs as the fastest ATV had reached what appeared to be a spin limit. I skipped doing eggs to not skew the xp/hr value.

Summary

The following is a summary of how I see things, your mileage may vary depending on your situation. With the prices of s905w going up, some s912 like the tx9s available for reasonable prices and the U2C very hard to find, there is only one clear winner for me.

Some of the takeaways for me are as follows:

  1. The s912 wins the performance race
  2. The extra 1gb of ram in the 3gb U2C didn't yield any benefit over the 2gb of the tx9s
  3. The s905w is out performed by the s912 variants by >30%
  4. All boxes are able to be flashed with the MADrom without issue
  5. you may not want a rack of s912, but having one on hand to level could prove useful. ymmv depending on your situation, i.e. for my suburban hell i won't likely see a difference, but if you are in a dense urban area a s912 could help

Differences from my normal leveling:

  1. I pop eggs, they are useless for workers once they reach level 30, so might as well use them. Leveling mode seems to handle the hatching screen well, but could result in reboots. I think the xp, stardust and chance at relevant monsters like dratini from 10k eggs is worth the risk. If you decide to sell accounts later or hit gyms with the accounts, decent monsters only make them more valuable.
  2. I would stack all 6 lucky eggs at say level 27. Should get you to 30 in under two days based off the math.
  3. I had run coords prior for the walkerareas. Coords has the potential to lead to backtracking when the route is finished to retry missed stops. I will be doing round of 1 for all further leveling endeavors.

A few random observations:

  1. From what i have seen prior to placing the s912 variants on the discharge fans at the top of my computer case, the s912 variants handle the 80c temps far more stably than the x96 mini
  2. tx9s is a monster in regards to size, U2C is similar sized to x96 mini
  3. scrcpy is stable and fluid on s912 variants

Further investigation / just wondering out loud (cooldowns probably mask though):

  1. I would like to see if someone with a say a fiber level of internet would level faster
  2. Is running local better than running on vps to speed up leveling