Radeon RX 9070 XT finally appears in Steam Hardware Survey — RDNA 4 flagship surprisingly lands just behind RTX 5080

palladin9479 said:

Steam currently has a sampled population of 120,000,000 ~ 132,000,000 clients and every client has the capability to participate in the survey, so unlike social studies where you can only do a few thousand at a time, they can potentially do millions of samples.

palladin9479 said:

It's much much lower then that, less then 1%.

Handy calculator for figuring out confidence and error rate.

https://uk.surveymonkey.com/learn/research-and-analysis/sample-size-calculator/
Steam currently has a sampled population of 120,000,000 ~ 132,000,000 clients and every client has the capability to participate in the survey, so unlike social studies where you can only do a few thousand at a time, they can potentially do millions of samples.

At 120,000,000 population and 99% confidence level you only need 66,528 random samples to achieve an error bar of 0.5%. If they wanted a 0.1% margin of error they would need 1,641,339 random samples. 1.6 million sounds like a lot, but they have over 120,000,000 to chose from, meaning they only need slightly over 1% of their population to participate in the survey.

That makes the Steam Hardware Surveys one of the most accurate statistical analysis of a randomized population on the planet. If you were to tell any medical or social scientist that they could gather near perfect data from over a million randomly sampled people from a population of over 120 million, they would consider it a perfect scenario. Yet because people have brand loyalties and preconceived beliefs of what's "correct", they mentally refuse to accept anything that would contradict those beliefs.

Tell some tech enthusiast, which we all are on here, that 51.8% of "PC gamers" are using 1080p monitors, 21.2% use 1440p and only 5% use 2160p, and they would could you a liar insisting that 1080p is old news and "everyone" moved on to bigger resolutions. Tell them that over half of the PC gaming population is using GPU's with 8GB or less of VRAM, and they would again call you a liar because, as they know, all games now require 12GB or more. Present them with data and they would just call your data wrong, never stopping to question their preconceived notions and participation in a self selected community that is significantly deviated from the median.


Caveat to some definitions because this is the land of "well acktually". The Steam Hardware Survey collects near perfect statistics from the population of personal computers with the Steam client installed. This doesn't include consoles, work computers, business laptops, workstations and other personal computers that might not be running the Steam client. Since Steam is the largest and most popular gaming distribution platform on the planet, it is reasonable to assume that most people falling into the "PC gamer" category would have it installed and therefor be included in the surveyed population. This is what is known as "proxy data", when one set of statistics are used as a proxy to measure something that would otherwise be difficult to measure directly. It's a very common practice in social, medical and environment sciences.

https://en.wikipedia.org/wiki/Proxy_(statistics)

palladin9479 said:

Previously the 9070XT was included in the "other" category because there wasn't enough of them surveyed to hit the top list. Notice there aren't any 9060's listed, not because they aren't in existence but because their surveyed quantity is too small to hit the named list.

dion_ said:

A card that slowly accumulates until it crosses the display threshold appears AT that threshold and then climbs. You can read the threshold straight off the chart: the RX 9070 debuted in December at 0.22%, then sat at 0.16–0.18% for months, and there are named entries down near 0.17%. So a genuine threshold-crossing debut looks like ~0.17%. The 9070 XT did not just cross the threshold — it went from a dash to 1.33% in one update, roughly eight times the threshold, smoking a stack of RTX 30/40 cards. A continuous quantity cannot step from <0.17% to 1.33% in a month. So either it grew ~1.2 percentage points in thirty days (which contradicts your own "low-volume vendor accumulates market share slowly" thesis), or it was already somewhere between 0.17% and 1.33% and simply wasn't being named properly. It was far more likely sitting inside the "AMD Radeon(TM) Graphics" bucket, which lost ~1.5 percentage points month-to-month, which is a substantial drop.