Bad data helps no one. Amazon top seller list and steam surveys have fundamental flaws because they are not designed for the purpose for which they are used here.
He's not alone, anyone not too invested in what the figures show could see that steam is a reasonable statistical tool - huge data set, random sampling, regular surveys. Not perfect, no survey every is, but still very useful. No - I can't be bothered arguing with you over it because you don't want to listen because it isn't saying what you want to hear. Just don't be under the assumption that banging the drum louder then anyone else has made you any more correct.
It's not a random sampling. It doesn't accurately identify hardware in users systems. It only tracks the hardware of those users that do not have certain IGPs, do not have multiple GPUs, and that volunteer their data. It is great for that demographic, if that appeals to you. It is useless for tracking sales data, which people seem to think it can accurately reflect. You don't have to have much experience with data to understand how poor a metric this is for what many users here seem to think it is.
Like your post added anything to the discussion
AMD is back on track, making inroads as planned by its management. Now let's wait and see if Vega can compete too.
Sent from my HUAWEI MT7-L09 using Tapatalk
It does work. See the performance of NVIDIA's GeForce GTX business and the performance of AMD's high end GPU biz. AMD's "share gains" like I keep telling you are mainly in low-end OEM systems, markets from which NVIDIA has pretty much pulled out.
Really now?
Basic statistics (Stats 100, 101, 110 various naming schemes) was a mandatory part of my bachelors degree here in the US - but it could be that its not the same way at every university even within the US
Zinfamous is correct. The Steam hardware survey is by definition not a random sample.
400x 1080
600x 1070
600x 1060
100x Fury
700x RX 480
Not bad for a weeks work!
I say this as a member posting in this thread.
This is a AMD specific topic and should have been posted in the AMD forum. Posting it in the general area was a mistake. Or if done so because of the higher traffic, for exposure/attention, then be prepared for the negative attention that brings with it. Also as the sub forums have more specific posting guidelines,members are subject to stricter moderation as a consequence. Don't like the heat? Stay out of the kitchen, comes to mind.
Latest Mindfactory.de numbers taken from today.
RX460 545
RX470 2025
RX480 3415
GTX 1060 10090
GTX 1070 17535
GTX 1080 7505
Yeah that was kind of my point in creating this thread, just to show that AMD is starting to pick up a little steam towards becoming a better competitor. I don't understand the reasoning behind some posters doing everything they can to disprove the numbers shown here, what incentive is there for that?
Hard to keep a topic positive here lately.
I wanted to say this the day I saw this thread.
Latest Mindfactory.de numbers taken from today.
RX460 545
RX470 2025
RX480 3415
GTX 1060 10090
GTX 1070 17535
GTX 1080 7505
Mindfactory regularly replaces SKUs and I believe quite a couple of ref 480s have already dropped out of their catalog. Also, I believe Alternate was the main launch partner for the 400 series in Germany, they had a lot of 4GB SKUs in stock and generally better stock levels than Mindfactory.Amazing that in a few months the 480 lost over 100 sales. Sounds like those numbers are wrong.
I am interested in the fact that the tool inaccurately reports on the data that it is supposed to report on.
Steam conducts a monthly survey to collect data about what kinds of computer hardware and software our customers are using. Participation in the survey is optional, and anonymous. The information gathered is incredibly helpful to us as we make decisions about what kinds of technology investments to make and products to offer.
Well it pops up randomly but you can also just click Run from your OS (Start button) and copy/paste the following and press enter. Both steam://takesurvey/1/ and steam://takesurvey/2/
Zinfamous is correct. The Steam hardware survey is by definition not a random sample.
This text is helpful for understanding the common statistical mistakes made in the life sciences, but is applicable to other disciplines as well: https://www.amazon.com/gp/product/0199946647/ (Intuitive Biostatistics, 3rd Ed.)
<-- also reads a lot of peer-reviewed journals
It's not a random sampling. It doesn't accurately identify hardware in users systems. It only tracks the hardware of those users that do not have certain IGPs, do not have multiple GPUs, and that volunteer their data. It is great for that demographic, if that appeals to you. It is useless for tracking sales data, which people seem to think it can accurately reflect. You don't have to have much experience with data to understand how poor a metric this is for what many users here seem to think it is.
I read a lot of peer-reviewed articles for a living (Life sciences) that depend on statistical analysis and employing the proper methods for the reported experiments. No flawed method is ever considered a "useful tool." The inherent nature of a tool inaccurately reporting what it deigns to report makes it useless. I haven't once argued that I don't like it "because it doesn't show what I want." Here, you guys are essentially accepting that fact while trying to argue that "it is still useful," then turning my argument into a subjective opinion on how I simply don't like the results.
You aren't going to win this argument against me as you simply don't know my experience with data sets. Obviously, this tech space really isn't my realm, but I know flawed methods when I see them. It is really quite obvious.
I don't care if it shows AMD or nVidia in a more favorable light. I am interested in the fact that the tool inaccurately reports on the data that it is supposed to report on. Whatever the reason that is doesn't matter. Until it is fixed, it doesn't tell anyone anything. I would be making the same argument if it showed "AMD ownership" as significantly higher. I guess you will just have to trust me on that claim, as I have to trust you that you will "instantly buy an AMD card if they are the top performer"
What is it supposed to report on? Is it not reporting what Valve says it is?
Please be specific about what aspects of the reported data are inaccurate to the statement from Valve.
Here's something to think about and which anyone in statistics should know.
When sampling a population the goal of the sample is to have a sample that follows the same distributions of the traits under consideration as in the population - the sample must be representative of the population in order for conclusions about the sample to apply to the population.
The easiest way for one to do this is to take a random sample. That does not mean, however, that the sample must be random - non random samples can give exactly the same statistics as random samples if not chosen with an inherent bias (i.e. there is no correlation between that which defines the sample as non-random and the traits that are under statistical consideration).
Good examples of 'non-random' samples which are statistically relevant are: taking everyone in the phone book under 'S' and asking them their favourite colour (negligible/no correlation between first letter of name and colour choice), voluntary surveys where only those interested participate (i.e pretty much all political polling surveys, a lot of graduate thesis', etc.) where it is assumed that there is no correlation or that the correlation between the likelihood of a response and the traits under consideration is irrelevant, and instead of getting the height of a random sample of people who shopped at a 'big and tall' store you use as your sample people who walked in on a certain week (no correlation between entry time and height).
All those examples are technically non-random but it is difficult to argue non-representative unless there is a correlation between the non-random variables and the recorded traits. Its really difficult to get true random samples in a lot of cases so much research makes do with what is available.
In conjunction with the Steam survey; its not whether the sample is random or not (in general though any kind of sample tends to become more accurate as the sample size increases) it is whether the survey is a representative sample. Most of the complaints/objections about the survey do not seem to be about this but are rather about it not being mandatory, ignoring certain users, etc. (Note: by nature of being a sample that kind of thing is allowed).
We are talking about marketshare: AMD and Nvidia. What about the Steam survey would cause it to not be a representative sample of actual (non-professional) marketshare? In other words, the fact that Steam is voluntary and messes up frequently in multi-GPU systems is meaningless (the survey is a sample and not the population as a whole) unless there is a systematic trend - AMD users less likely to volunteer, survey is more likely to ignore AMD IGP + AMD dGPU over Intel IGP + Nvidia dGPU, etc.
Mining could be a very valid reason for under-reported AMD numbers, another could be businesses and institutions are more likely to buy from one camp (and naturally these computers never have Steam on them). Nonetheless, even if there is known bias, it must be a statistically significant amount of bias (i.e. mining market is large enough to skew results).
The steam survey will be a statistically accurate depiction of the 'non-professional dgpu marketshare' if the following conditions are met.
1) Equal likelihood for a given AMD or Nvidia system to have Steam installed. (Skewed by miners, institutions, etc). Note that while miners could skew this probability there needs to be a significantly number of miners and they must not be counterbalanced by another source (i.e. education using Nvidia). This condition has fair chance of not being met.
2) Equal Opt in likelihood between AMD and Nvidia. In general, do AMD users take the survey as frequently as Nvidia users? Perhaps, Nvidia with its good reputation has more of its users on average 'brag' on the Steam survey? However, in general there isn't much reason for this condition not to be met.
3) Steam, when it messes up reporting, does so with equal likelihood between Nvidia and AMD. With multiple GPUs, igps, etc. how does Steam report? Does this method of reporting result in proportionally more AMD systems not being reported as AMD systems? Not sure on how well the survey meets the requirement.
In general, it doesn't matter that Steam is imperfect as long as Steam is equally imperfect between AMD and Nvidia. All you are looking at is a marketshare percentage after all
I remember my statistics course. First words the instructors said (paraphrased), "Lots of people don't truly understand statistics, even those who use them daily and publish often make mistakes".
I guess it also depends on your field - I'm in physics and there are a ton of methods, which are known to be wrong, even completely fundamentally flawed, yet are used everywhere because they give a close enough answer that it doesn't matter that they are flawed.
I'm not going through the various forums of users showing how it inaccurately reports iGPUs as users' dGPU, how it ignores multi GPU configurations, how the voluntary nature is not an actual statistical random sample--I did this at length in the other thread in response to Shintai's refusal to honestly address the very same criticisms. The fact is that it is broken and actually doesn't report hardware accurately. Here, let me bold this again, because every one of you is completely dismissing the criticism and inventing an argument that no one is making:
The steam survey does not accurately report hardware that is in users systems on a consistent basis.
This criticism is
completely brand-agnostic. If you don't like me pointing out these flaws, then you have some issue or agenda to mischaracterize the actual criticism that everyone makes. I don't get this, because a flawed tool is a flawed tool. It is unacceptable to be making any argument with this other than the select demographics that it seems to accurately report. The fact of the matter is that if it can't accurately track some hardware, it can't accurately track all hardware. On top of this, members here are trying to report this broken tool as a way to show actual sales data--at the same time these people un-ironically criticize AMD for selling units because of mining.
See above:
IT IS ACTUALLY BROKEN. It doesn't report the hardware in systems as it claims to.
You do understand that no real statistical sample can be taken as valid if the reporting is actually wrong, right?
Please tell me you didn't waste all your time with that post while completely ignoring the actual flaws that we are addressing with the steam survey.
This isn't an issue of p-hacking, it's an issue of the steam survey essentially blurting out a bunch of neighbors like 1897346389750932484656209348 and calling those random, and half of you guys call that random and take the survey's word for it.
(hint: if you do know anything about statistics, you know that pile of numbers is not what a truly random sample looks like)
Let me be more direct: do you think that a program that counts the iGPU on your intel i7 as the GPU, over whatever GPU you have plugged into your PCI slot, as an accurate measure of your hardware? What about a program that ignores your multi GPU setup and counts those 2 or 3 or 4 cards in your system as one card? Is that good data for telling us how many GPUs are out there in the world?
Everyone that I know in my position would say"Gee zin, that's a bad tool. Let's not use that tool."
Let me be more direct: do you think that a program that counts the iGPU on your intel i7 as the GPU, over whatever GPU you have plugged into your PCI slot, as an accurate measure of your hardware?
A flawed tool can be better than no tool at all. (Or worse sometimes). Most methods of gathering data are flawed to some degree.
Hit the nail on the head. How about Mr Zinfamous present better data then, since he has so much noise to make?