http://i.imgur.com/Qkz0kfz.png
g402 with gyro @ 800 dpi which is suppose to be smoothing free under 2000 or so dpi
Well I would like to say that logitech acknowledges the existing of "smoothing". So they obviously must have a test of their own along with MIONIX releasing test firmware for the avior to specifically fix their "smoothing" that has been reported

https://imgur.com/irDoD7r
g100s

http://imgur.com/98fMoZK
Roccat XTD optical

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

kinzu v2 400 dpi

http://imgur.com/19GZTOx

Kana V2 (white)

(what is smoothing ?)

http://postimg.org/image/bw811xx8j/
abyssus 2014

http://i.imgur.com/AevzM9K.png
g100s lowest dpi on some cm storm pad

http://i.imgur.com/BB3FGdM.png

Logitech g502 800 dpi

A4Tech X7 cloth mat

http://s27.postimg.org/k3c248cxv/Untitled.jpg

Zowie FK1 @ 800 dpi

this will so lead to nothing...

http://gyazo.com/f8507fff8e697a9b432f3908bc4a037d
g100s @ 1000dpi

http://gyazo.com/60b6973588ff58cf07f2ac24495c9283
kana @ 800dpi

http://i.imgur.com/Iz9tqc1.jpg?1
g400s @ 800dpi with CPL mousefix(i know, markC was inconsistent for me, so more like 600-700dpi)

Results so far:
Time: Non smoothing wins
Accuracy: Smoothing wins
Targets/s: Non smoothing wins

Smoothing:
Time: 30.5
Accuracy: 95.58
Targets/s: 2.86

Non-smoothing:
Time: 34.17
Accuracy: 95.2
Targets/s: 2.91

I put Abyssus 2014 into the non-smoothing catagory, not sure if thats right or not though Oo

those numbers don't really say much if you do not include an estimate of the uncertainties i.e. the standard error

It's not right. None of the mice tested have been "smoothing" free.

Almost all non-MLT04 mice have some kind of "smoothing" feel (which is likely caused by a few different factors such as input lag/click latency, and artificial processing of mouse movement) to them.

Also, it's most noticeable in 3D FPS games, not 2D flash/desktop environments.

The original FK was the only mouse I tested that felt significantly off in the desktop environment.

http://imgur.com/J6ZgfY5

G100s.

Not entirely convinced that this means anything lol (can't have the same sensitivity that you use in games, for one), but I tried my best.

Might aswell submit my own with the Aurora:
http://i.imgur.com/5gDkKiR.jpg
Tried to get 1 minute+ but gave up :/

Yeah, it's pretty hard to get 1 minute. I guess the less you try to go for accuracy the more time you could potentially last.

World record is 4:34 apparently (rofl), but since you can easily slow down the game and edit the video to make it look legit, who knows if it's real or not: http://youtu.be/ntaSNtXp4cQ

I finally made an account. I didn't expect to stay here this long, but I guess this is the beginning of the end for me. :)

I appreciate your effort, but I do not believe a test of this nature tests the correct variable. The primary variable being tested here is the difference in skill between individual test subjects. The variance in this single variable alone will cloud any possibility to analyze the data for other variables, including the one we are actually trying to test: differences between the devices' sensors.

A more suitable test—though more difficult to execute—would be a double-blind, where a single subject tests two different sensors. The goal of a double-blind executed in this fashion is to test the subject's perception of a single variable. It could then be repeated with multiple individual subjects to establish a pool of data that could potentially be useful in drawing some conclusions. In order to do this, we would have to equalize as many variables as possible except the single variable we are trying to test. An example would look something like this:

Take two Logitech G400 shells, remove the internals of one and replace it with the internals of a Logitech G400S. Test the four clicks of both mice for feel. If they are not consistent, we will have to go through different batches of top shells and Omron swtiches until we find two pairs that are consistent enough for there to be no perceptible difference to the user and replace the switches on the PCBs. At this point, we may as well also remove the scroll wheel and the side buttons. Finally, weigh the two devices and add weight to the lighter device if there is a discrepancy, making sure also to keep weight balance in parity. Add labels underneath each device, "A" for the G400, and "B" for the G400S.

The reasoning behind these specific choices: "Smoothing" is the single variable we are trying to test, so we are attempting to equalize as many other variables as possible. We choose rev. 1 and rev. 2 S3095 sensors—at 800 CPI @ 1000 Hz we are hoping these two sensors perform as identically as possible except in the area of "smoothing." We also have to take the the physical characteristics of both devices into account. This is why we place both devices in the same outer shell, equalize the weight, and try as best as possible to get the clicks between the two devices to a parity where there is no perceptible difference to the test subject. If we keep the scroll wheel and side buttons, we would have to test them for parity as well—we remove them as variables for the sake of keeping the test as simple as possible.

The hope is that every perceptible variable to the user has been made equal except "smoothing" between the two sensors. I must caution that even this extreme example may not be perfect—there may still be other minute differences between the two sensors besides "smoothing" that could theoretically invalidate the results.

We then devise a hooded box with an open front panel to place over the mousing area so the subject cannot see which device he is using. This is to completely remove any possibility that any cosmetic differences between the two devices could be seen by the subject and invalidate the double-blind. For example, pretend that during one of the trials, the subject's fingernail accidentally scratches the side of one of the devices leaving a visible mark. This mark could potentially allow the subject to identify the device in later trials, compromising the integrity of the double-blind.

The subject will select a single game to use for this test. It can be whatever game he wants, but it should be a game he is extremely familiar with. For sake of example, let's say it's Quake. All settings are configured to the subject's preferences, and in-game sensitivity is adjusted to match the subject's preferred X cm / 360 at the S3095's native 800 CPI.

We first allow the subject to establish a subjective baseline. He is allowed to play Quake with both device A and B for as long as he desires, and is allowed to switch between them as he pleases. At all times, he is aware of which device he is using. This is so the subject can get a feel for both devices and establish a subjective baseline for how he perceives the two devices to perform. When the subject is finished, the administrator of the test asks the subject whether he can perceive a difference in "smoothing" between the two devices. The subject answers yes or no and the administrator records this answer. We are now ready to perform the double-blind test.

The double-blind procedure itself is simple:

The subject exits the room to a place where he cannot hear or perceive in any way what is happening in the test environment. The administrator plugs either device A or B into the test system, places it onto the mousing surface, and places the hood over it. The administrator creates a new row on his datasheet labeled "test 1" and records in this row which device—A or B—is plugged in. The subject enters the room and plays Quake for an allotted period of time—say five minutes. After the allotted period has ended, the subject is asked which device he thinks he was using, and he answers either A or B. This answer is recorded on the datasheet in the same "test 1" row next to the entry which lists the actual device. The subject then leaves the room and "test 1" ends. To reset the test, the device is wiped down, unplugged, and replaced randomly with one of the two devices—flip a coin. The administrator starts a new row on his datasheet (test 2) and the process is repeated however many times deemed fit. An example of what the final datasheet could look like:

First column: Actual device | Second column: Subject's answer

Test 1: A | B
Test 2: A | A
Test 3: B | B
Test 4: A | B
Test 5: B | B

This is just an example with made-up entries. In a real test, we would repeat the process as many times as possible to get a large enough sample size, perhaps 20-50 repetitions—obviously, more is always better. After the test is finished, we judge the subject's accuracy in the double-blind against his yes or no answer from the baseline phase. The variable being tested here is the subject's perception of our test variable. We asked the subject during the baseline phase if he perceived a difference in "smoothing." We then put this assertion to the double-blind test by having him switch between two devices where (we hope) "smoothing" the only perceptible difference. If the subject answered yes in the baseline phase and had a high rate of accuracy in the double-blind, the evidence would support the idea that this subject can indeed perceive a difference in "smoothing" between the two sensors. If the subject's accuracy in the double-blind was poor, we can assume the subject cannot perceive any difference.

The test can subsequently be repeated with as many individual subjects as desired to establish a pool of data from which possible conclusions could be drawn. All we would need is enough people who consistently scored high in this double-blind to come to the conclusion that there is a real-world, perceptible difference between these two sensors that is being measurably observed—we are hoping that the source of this difference is "smoothing," as we have tried as best as we can to eliminate all other variables. Again, the test is not perfect—it is the best I could come up with for this example.

Obviously this test would be difficult to prepare and execute. I only spell it out like this to illustrate my original point: I don't believe the test you've devised is testing the correct variable.

http://i3.kym-cdn.com/photos/images/original/000/284/529/e65.gif

thanks for sharing your masters thesis with us

I didn't read more than like 2 sentences but yeah this data isn't going to be useful at all and im surprised bst even thought to try this.

I agree, thats the best way to test it, but its probably never going to happen. This is just a bit of fun really... yeah maybe its a waste of time. But its better than nothing, no one else has come up with other any test at all. There are threads like this on ESR just for fun, but without taking any data, just people posting their results, this is just like that but I just happen to be doing something extra.

I still think its interesting to see if people who are sensitive to smoothing and/or use a mouse without it, score higher.

If theres a big difference, then it would suggest that smoothing is something that affects everyone, I know it doesn't prove it, but it suggests it. If the difference is very small, then it suggests that even though it affects some people, it doesn't affect others.

I think the more people that participate, the more accurate it will be, with regards to skill. At the moment with just a handful of people, someone who gets 120 seconds in could skew the results, but it would be watered down with enough data.

We will technically be able to execute a similar test though when we get our hands on the Aurora PCB and shove it in to a WMO.

I assume your intention is to use two WMO shells, one with WMO internals and another with the Aurora's? As stated before, you would have to attempt to equalize as many other variables as possible. If there is a discrepancy in weight—which there most likely will be without modification—that alone would be enough to tip the subject off to which mouse he was using.

Another consideration that comes to mind is tracking. The MLT04 has an extremely low PCS in comparison to the A3090. You would have to make sure the subject never exceeds this PCS in testing or he will instantly know which mouse he is using. Also, I have no personal experience with this, but my hunch is that even excluding any "smoothing," the MLT04 and A3090 would have completely different tracking characteristics that could differentiate the two to a perceptible user. Click latency could be a possible issue as well.

http://utmalesoldiers.blogspot.com/2013/02/114.html

This is why in my example I chose two mice that are as similar as possible—down to even the sensors they use—in order to isolate the "smoothing" variable as much as possible.

I do think your idea is still worth a shot. You really don't know until you try with these things. As long as everything is well-documented—including which variables were attempted (and not attempted) to be equalized—there are always conclusions that can be drawn from data; even if that conclusion is, "Not enough variables were equalized," that is still helpful for any future endeavors.

I understand it's for fun. I just feel there are too many variables acting simultaneously for the data to be useful in drawing conclusions about any single variable. For instance, click latency is something that has a huge range of variance among different devices.

http://utmalesoldiers.blogspot.com/2013/02/114.html

I agree, thats the best way to test it, but its probably never going to happen.

If I'm not mistaken, did you not recently launch a gaming mouse product? I think of all users in these communities, you would be the most capable of pulling something like this off—is it not within your ability to build two prototype devices that are identical in every way except rev. 1 and rev. 2 A3090 sensors? ;)

Granted, it's time and resources. I wonder, what if you started a thread at OCN to crowd-source the funding? I think there's a chance it would gain traction. If time is an issue, you could even raise the money just for the test devices, have them built, and then shipped to someone else decided upon by the community who would carry out the experiments. At least worth starting a thread over there about it, imo.

On second thought, rev. 1 and rev. 2 A3090 sensors do not share a native CPI step, so they wouldn't be the best candidates. To my knowledge, there aren't two sensors available to the open market that share extremely similar tracking characteristics and native CPI steps that only differ in the area of "smoothing." The rev. 1 and rev. 2 S3095 sensors seem like the best pair for this purpose, but as far as I'm aware those are proprietary to Logitech.

now to make everyone feel bad about how they did :

https://www.youtube.com/watch?v=F0DD4YlgbKQ&a...jipCrfdNlw

http://www.esreality.com/post/2640619/input-l...pid2675541

not sure if this is how it works, but it might be possible to actually measure it if so. noacc hasn't responded to that thread in a while though.

http://www.amazon.com/Casio-Exilim-Ex-ZR400-D...00B7D7P2K/

here's a 1000FPS (at low resolution) camera for $170 if anyone wants to test it and then return it the next day :-)

http://i.imgur.com/MFBIA5C.jpg

deathadder 2013 @ 1800dpi 1000hz windows sens 6/11

did this after 3rd try ;) never played this kind of game before :)

3rd try = not following directions

Thats ok, maybe I should have explained it better, you can try as many times as you can only post one result. I was just trying to avoid people wanting to update their results because they improved, would be too much effort.

Anyone who wants to submit new results because they submitted their first results, reply to this message with the new results and I'll update it.

http://imgur.com/U5TStpX

Yeah, as expected, if you relax the accuracy it's very likely that you last longer. So I wonder what that means for bst in terms of overall results.

http://imgur.com/okS6qB7

Ime 3.0

Kind of funny how the average accuracy is exactly the same.

Smoothing:

Time: 25,28,33,33,24,40,55,37

Average: 34.375

Acc: 100,91.5,96.3,96.3,94.4,95,95.4,97.8

Average: 95.8375

Targets/s: 2.77,2.83,2.91,2.9,2.76,3,3.19,2.96

Average: 2.915


Non Smoothing:

Time: 47,19,25,43,38,33,51,52

Average: 38.5

Acc: 92.1,90.6,98.2,95.5,94.8,100,97.7,97.8

Average: 95.8375

Targets/s: 3.09,2.67,2.78,3.04,2.97,2.89,3.14,3.14

Average: 2.965

http://i.imgur.com/Jsz8Ej1.jpg

Kana v2 white 800 dpi

there are infinite other (and way more important)influences, which affect the outcome of this test, than smoothing.
so plz...

But you're talking about things that exist for both sides, there isn't any reason why people with smoothing would have more or less influences than people without. Both sides are a mixed bag, but the one thing that separates them all is smoothing.

Put it like this, if you tested the entire world on the 100m sprint, and you split them according to how much fruit they eat, and there was a huge difference between people who ate 5 a day and those who ate none, would you say "oh but its a useless test because theres infinite other influences"? I agree it wouldn't be a perfect test, but it would be a good indicator imo.

just no.
you also had the entire world to participate in this test, to show any indications.

Put it like this, if you tested the entire world on the 100m sprint [...]

You are correct that random variables would affect both groups equally. Through your example however, you are also pointing out a key factor: sample size. The smaller the sample size, the greater these random variables may affect the results. You said so yourself in an earlier post that a single result that deviates highly from the average could easily skew the numbers of one group substantially. The cause of this outlying result could be any random variable other than the one we are testing for. For example, if someone scores poorly using a device with extremely poor click latency, this single variable from this single test result is already messing with your data.

http://utmalesoldiers.blogspot.com/2013/02/114.html

The assumption then, is that someone in the opposing group may also have a device with extremely poor click latency. Even assuming that this happens, what if this person is much more skillful with his high-latency device and scores much closer to the average than your outlier did? As you can see, there are simply too many variables acting simultaneously for us to really know what is happening.

In my opinion, you can only draw meaningful conclusions from survey-like tests of this nature when you are able to obtain an extremely large sample size. Only then is the data potentially representative of both groups.

g100s
http://i.imgur.com/PhyNnho.png

wmo @ 1000Hz

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

I think this whole thread turned into a competition to see who does better in a flash game. lol

This game reminds me of this one, btw:

http://youtu.be/MwTX3bQO_d4

Pretty similar.

I did it once and got 2:30, but I didn't save and just tried it again, w/e. Have not played fps games in a while.

https://i.imgur.com/Ikjzakd.png
g9x(no shell or cartridge), 800 dpi, razer goliathus control.

edit: Tried it again today, did 2:39 on the g100s although I probably could do better on the g9x. My in-game aim is definitely better on the g9x.

Played for solid 4-5 min this was the best.
http://prntscr.com/5426kd

Tbh after the 1st 3-4 trys I just got worse.

g9x I believe 400 dpi on puretrack talent. Haven't played fps ~1.5years.

g100s on 1000dpi no drivers no internal weight

http://puu.sh/cJzpv/544025fd95.png

For fun:

http://imgur.com/tT2DFdv

Zowie FK2014, 1150 dpi, 1000hz

http://puu.sh/cK1bo/9a4846e723.png

W M O
M
O

This wasn't necessarily my best time - I got 96.x% accuracy earlier with about 130 targets, but this was at least 30 targets more than my previous rec.

http://puu.sh/cK1rz/52c4207e3f.png

Deathadder

I'm used to both mice. The shapes are too wildly different for me to be able to tell if there's any virtual difference between the two, but this made it really evident just how awkward the Deathadder feels compared to the WMO. It's really big and heavy and unwieldy, puts my hand in an awkward position, and doesn't move as easily or cleanly.

Also, I was amazed by how well I did. My aim is notoriously bad, but I did alright here.

The deathadder is a significantly better mouse than the wmo in every way.

Maybe, maybe not, all I know is I prefer the WMO, but I use the DA when I need side buttons.

so what is smoothing? name implies movement filtering, which should be VERY bad, or is it just some kind of jitter correction?

Think of prerendered Frames, then turn the concept around, a set amount of Frames are gathered to create a smoother(less jittering/jitterfree) cursor path.

source? since that would mean lag, although depending on FPS specs, i seem to remember years ago IME3.0 etc stated 6000fps with would mean 2-3frames gathered and smoothed, would still be less than 1ms lag.

(1000ms/6000fps)*3fps=0.5ms

While i cant give you a specific source for several reasons, i will instead refer to cpate´s Post on ocn instead. http://www.overclock.net/t/1499731/after-some...t_22639253

And yes thats how it works, altough with certain sensor models its not just 1 or 2 frames, early 9800 even had casual bobs complaining about "lag" with their Qpad 5k l.e. on their forums.

below 2000dpi no smoothing. From a logitech guy? is good enough for me since 800dpi is more than fast enough.

Just to make it clear the "adaptive smoothing" counts only for AM010.

57 secs WMO 1000hz 6th try

http://i.snag.gy/Kfz6v.jpg

Logitech G5 (M-UAH113A), no drivers, just plug 'n play.