Larry Litmanen :
To me Steam data should be very accurate. It's not like they ask people what they have, Steam simply sees what specs the user has on any given month and reports it.
Steam doesn't check the hardware that everyone is using each month. They only sample a portion of their user base, occasionally popping up a message asking if a user would like to submit their data to the hardware survey for that month. So, the results could vary depending on how those users are selected, and how likely various demographics are to select "yes" or "no" to the request to submit a profile of their system information. Some systems tend to rarely get requests to take part in the survey, while others get asked fairly often, and only Valve knows how these systems are selected.
Larry Litmanen :
I would say Amazon sales data is not accurate because most people do not buy CPUs, they buy ready made systems. So a sale on Amazon simply means that people who build their own PC chose this or that CPU, that segment of people is not significant compared to overall PC user base.
Amazon's sales numbers could be considered to be reasonably accurate within the context of custom-built desktop systems, at least as much as Steam's survey results could be considered accurate within the context of the subset of systems that have Steam installed. Of course, Steam's results only reflect the segment of PCs that are used for gaming purposes, so that eliminates many other demographics. Most computers in the world are not running Steam, after all.
Larry Litmanen :
So when i look at Steam data it shows a MASSIVE decline in the last few months for AMD, they have less than 10% of the market for CPU and GPU respectively. Steam is obviously not an exact measure, not everyone games but i think it is very accurate, numbers on Steam are what they are. If you have an AMD CPU that's what the survey shows.
When I look at the
Steam data, it shows me that something either went massively wrong with their survey system in recent months, or the demographic of systems running Steam has massively changed. Just looking at the highlight graphs at the top of the results should show you that. To start with, the market share for both AMD graphics cards
and Intel's integrated graphics got cut in half in a little over a month. Did half of those with AMD or Intel graphics suddenly feel the urge to get an nvidia GPU in October? Of course not. A wild swing in installed hardware simply isn't going to happen that quickly, especially when there haven't been any major new product lines released in recent months. For the most part, we're still on the same graphics hardware generation released a year and a half ago, with only minor revisions from both companies since then. And at the same time, you see AMD's CPU market share following an identical trend despite them having a compelling CPU lineup.
If you look a bit further into the rest of the results, you can get a better idea of what's going on here, and it doesn't actually involve AMD's sales decreasing. To start, have a look at the market share for operating systems in use. Back in September, over 50% of Steam users were running Windows 10. Now, that number has plummeted to less than 24% within two months. Over the same period of time, the market share for Windows 7 has surged from 35% up to 70%. Did everyone suddenly decide to drop Windows 10 to go back to Windows 7? Again, obviously not. The market share for Mac and Linux also dropped by more than half during that same period of time.
So, what changed? Have a look down at the language data. Over the last few months, the percentage of Steam users who speak Simplified Chinese has increased from around 17% up to 62% of Steam's user base, while English-speaking Steam users have dropped from around 40% down to just 17%. Basically, Steam's active accounts have more than doubled in the last few months due to a massive influx of Chinese users. So, the change in results reflects the hardware being used by Steam users in China, more than anything. Plus, a very large portion of PC gamers in China play games at Internet cafes, so it's likely that survey results are getting submitted numerous times for each of those public computers, and the results might now be reflecting the hardware being used in Chinese Internet cafes more than anything. Or maybe a bunch of them might even be bot accounts used for grinding gold in free-to-play MMOs or something. It's difficult to say without further data.
Either way, the data no longer reflects the hardware being used by gamers in the western world as much as it did a few months back. I'd love to see regional filtering added to the Steam Hardware Survey in a future revision, much like the existing ability to filter by operating system. As it stands now, the results are simply not relevant for most uses. If a developer wants to optimize a game for the hardware and software their target audience is using, the current data won't be very useful to them unless their primary target market for the game is China. The same goes for analyzing trends, as in this article. If the demographic being surveyed can potentially change drastically from one month to the next, there's little way of telling how much the results are actually changing, and how much they only appear to change due to changes in the Steam user base itself.