Steam Hardware and Software Survey (May 2017)

Any particular items you want to discuss or just felt the need to drop a link and run without participating in a discussion?
 
Two views of the same marketplace:

- market share reports describing boards moving off of shelves today
- steam survey showing the shape of the active installed base
 
Most people here aware about Steam survey, but probably many are not aware that Unity has also started collecting similar data since some times ago. Geared more toward indie games and casual players, it paints quite similar pictures with what Steam survey reveal.

On windows platform in particular, NVidia dominance over AMD + Intel is even more overwhelming. Which is funny, when you would naturally think that on the lower to mid end spectrum of hardware install base IGP should eat more into discrete graphics market share.

http://hwstats.unity3d.com/pc/index-win.html
 
Most people here aware about Steam survey, but probably many are not aware that Unity has also started collecting similar data since some times ago. Geared more toward indie games and casual players, it paints quite similar pictures with what Steam survey reveal.

On windows platform in particular, NVidia dominance over AMD + Intel is even more overwhelming. Which is funny, when you would naturally think that on the lower to mid end spectrum of hardware install base IGP should eat more into discrete graphics market share.

http://hwstats.unity3d.com/pc/index-win.html
  • Windows 7: 77.8%
  • Windows XP: 10.3%
  • Windows 10: 7.7% WTF It was free for a year
 
  • Windows 7: 77.8%
  • Windows XP: 10.3%
  • Windows 10: 7.7% WTF It was free for a year

is this a side effect of piracy affecting both surveys? No one wants to admit it, but they should have put that question on the steam survey.
 
People would rather run virus infested software that they are used to than try something new.
As I understand it, if we are referring to IwannaCry, the majority of affected users were Win7 users. W10 should have been affected too, but IIRC W10 forces patching which they've been accused of malpractice for. Where with win7 users could disable it, and truthfully a great deal of pirated users would have done exactly this because software patching would have discovered that they were not using a genuine copy of windows. I would know, this was once upon a time for me when I was in the university years.

and so the merry go round goes...
 
is this a side effect of piracy affecting both surveys? No one wants to admit it, but they should have put that question on the steam survey.

No, see below

It's interesting how the subset of the Unity gaming market is different from the gaming market at large.

When looking at the Steam survey.

Windows 10 - 51.01%
Windows 7 - 36.75%
Windows XP - 0.88%

When looking at all OSes (to be comparable to Steam Survey)

Windows 10 - 32.1%
Windows 7 - 49.2%
Windows XP - 6.5%

Wait what? That doesn't line up with the Windows hardware stats.

Oh wait, I think I see where the discrepancy is. There's also a Windows Store section. I'm going to guess that a lot of Unity titles are sold through the Windows Store, and thus, any Windows 10 Unity installations that are bought through the Windows Store are considered part of that instead of Windows hardware.

You can see this through their Player (user) platforms.

Windows - 63.8%
Windows Store - 35.5%
OSX - 0.6%
Linux - 0.1%

Of course this makes their Windows hardware GPU numbers almost nonsensical as Windows GPU hardware share is split between Windows (Nvidia dominated) and Windows Store (Intel dominated). But at least they still have the All category.

NV - 47.9%
Intel - 39.7%
AMD - 11.8%

Compared to Steam

NV - 62.41%
Intel - 15.65%
AMD/ATI - 21.53%

That actually shows a lower share of NV GPUs than the Steam Survey. But that makes sense when considering that the Windows Store has a greater percentage of people using Intel GPUs.

Regards,
SB
 
Last edited:
That actually shows a lower share of NV GPUs than the Steam Survey. But that makes sense when considering that the Windows Store has a greater percentage of people using Intel GPUs.
I think what matters regarding discrete shares between AMD and NVIDIA are the ratios, different survey will target different demographics (Unity targets people with iGPUs, while Steam aims for the AAA dGPU market), as such percentages will change, but ratios likely remain stable. In the Steam survey NV to AMD ratio is 3:1, in Unity it's 4:1. So at least we can now somewhat validate the 3:1 ratio of Steam.

In fact, if you only calculated the Windows results + Windows Store numbers in the Unity survey, the market share will be like this:

AMD 12%
NVIDIA 39.5%
Intel 48.5%

The ratio is now 3:1, just like Steam numbers.
 
I think what matters regarding discrete shares between AMD and NVIDIA are the ratios, different survey will target different demographics (Unity targets people with iGPUs, while Steam aims for the AAA dGPU market), as such percentages will change, but ratios likely remain stable. In the Steam survey NV to AMD ratio is 3:1, in Unity it's 4:1. So at least we can now somewhat validate the 3:1 ratio of Steam.

In fact, if you only calculated the Windows results + Windows Store numbers in the Unity survey, the market share will be like this:

AMD 12%
NVIDIA 39.5%
Intel 48.5%

The ratio is now 3:1, just like Steam numbers.

I'm not sure how you are getting those numbers as non-Windows Unity numbers only account for 0.7%. Excluding them isn't going to change the numbers much.

Ah, I see, you just added them then divided by 2. That isn't going to result in correct numbers as Windows Store represent just 35.75% of Windows machines in the Unity hardware survey and not 50%.

Doing the math it comes out to 48.50% for NV for Windows machines. That number goes up compared to overall numbers due to how much Intel dominates OSX and Linux.

Regards,
SB
 
Last edited:
Ah, I see, you just added them then divided by 2. That isn't going to result in correct numbers as Windows Store represent just a 35.7% of Windows machines in the Unity hardware survey and not 50%.
Didn't notice that to be honest, but it's adjustable ..

With that in mind, it will look roughly like this:

AMD 12%
NVIDIA 49%
Intel 39%

So the same ratio of 4:1 as before.
 
One thing that is particularly interesting with the Unity Hardware survey is the mobile hardware numbers.

Android - 86.2%
iOS - 12.3%

Most iOS users are on the latest major revision of the OS.
  • 60.3% on 10.x
  • 27.0% on 9.x
  • 7.8% on 8.x
  • 4.9% on 7.x or older
Android is a bit of a mess in comparison.
  • 1.4% on 7.x
  • 25.0% on 6.x
  • 38.5% on 5.x
  • 35.1% on 4.x or older
Android 4.x is really old now. I think the Android situation illustrates quite well how unlikely it is for a mobile game developer to target the latest generation of GPUs and their features. I suppose this is why PowerVR isn't in high demand on Android devices as it's less likely for advanced features to get used there than on iOS. That in turn makes it easier for Android phone makers to settle for a GPU that is good enough. That in turn reinforces to software developers that they can't target more powerful or capable hardware. Which in turn...vicious cycle.

If there was a large market on phones for premium (20, 30, 40 USD or higher game prices) then there could be an opportunity to break that cycle by focusing on a "core" market of gamers. But with the low tolerance of mobile users to high priced premium software, developers must rely on their title working on the greatest number of headsets possible.

It's no wonder that NVidia abandoned the mobile SOC market. There really isn't much demand there for a high powered GPU outside of Apple. And even there, I wonder if they'll continue to push the mobile envelope WRT GPU power as hard as they did in the past with their partnership with PowerVR? I suspect they will, if for no other reason than bragging rights, but there's a possibility they may settle for doing just enough to remain at the top of the mobile SOC world.

Regards,
SB
 
Steam Hardware Survey is definitely very bogus and we've known that for a while.
And it's not the proportion between GPUs of different brands that makes it completely unbelievable, it's the percentual change between each month that frequently puts very old GPUs taking large jumps in marketshare. This month is no different.




On windows platform in particular, NVidia dominance over AMD + Intel is even more overwhelming. Which is funny, when you would naturally think that on the lower to mid end spectrum of hardware install base IGP should eat more into discrete graphics market share.
Most of the simpler indie games using Unity are probably being sold within the windows store, where Intel has 79% of marketshare, AMD has 12% and nvidia has 8%. This does fit within the lower to mid-end spectrum with IGPs that you mentioned.
On the other hand, they also claim that Intel SM4.0 GPUs have a whopping 47% of the market. Considering the fact that only the later chipsets for the Core 2 Duo/Quad had this configuration, I find it really hard to believe that these are found in so many Windows 8+10 machines nowadays.

And I have no idea how they're counting the "regular" Windows machines and how/if this data is crossed with UWP machines (maybe "Windows" = "Total Windows" - UWP ?), but it's also a bit fishy that the GTX 750 takes 22% of the whole GPU market in March 2017.


There really isn't much demand there for a high powered GPU outside of Apple. And even there, I wonder if they'll continue to push the mobile envelope WRT GPU power as hard as they did in the past with their partnership with PowerVR? I suspect they will, if for no other reason than bragging rights, but there's a possibility they may settle for doing just enough to remain at the top of the mobile SOC world.
GPU performance in Android SoCs will probably keep increasing in performance because of cardboard/daydream VR games.
The bad news about this is that regular games in Android (e.g. late-generation console ports) will keep being a complete mess regarding gamepad support and online ecossystem. The good news is high-end SoC makers will probably focus much more on sustainable performance instead of one-time runs of antutu and GFXBench.




I think the Android situation illustrates quite well how unlikely it is for a mobile game developer to target the latest generation of GPUs and their features.
Yes, Android is a mess regarding software updates. You either stop getting software updates after less than a year (two if you're lucky enough to purchase a phone that ends up being very popular) or you lose tens of hours trying to understand how to change your ROM into one that is more recent but voids your warranty, and eventually kill your phone in the process.

I very reluctantly just purchased an Android mobile phone and I hope it'll be the last. I can't wait for Microsoft to get their shit together and launch a full legacy-compatible Windows for phones. I'll also wait to see how ARMv8 SoCs emulate x86 software, but ideally I'd wait for Intel and AMD to start trying to develop ULP SoCs with x86 cores.
 
Last edited by a moderator:
Steam Hardware Survey is definitely very bogus and we've known that for a while.
And it's not the proportion between GPUs of different brands that makes it completely unbelievable, it's the percentual change between each month that frequently puts very old GPUs taking large jumps in marketshare. This month is no different.

Until we understand the reason(s).

Possibility one. Selling second hand, recycled and repaired hardware to emerging markets is observable since years. The number of NICs grows and there is a growing number of people able to affort those goods. Hence, computer originally used for for example office needs, which are available in massive quantities through these channels, are repurposed in these markets, in which the people also have a growing desire for entertainment and leasure, as their economic situation becomes more stable.

Here's a possible alternative. In Dota 2 there are boosters, people that create an account to bring to level X, and then sell it. I assume this is the case for all MMOs, why not when it's money. Which means one machine possibly has say 100 accounts originally associated with it. The boosters are mostly russians and chinese. Which again relates to NICs and that decent computers are super expensive in these regions etc., and that they are channeling second hand stuff in various ways into the country (see above).

The results are just counter-intuitive, but certainly not wrong, like "iron can float". (edit: uhm I don't want to go into the discussion that yes according to the meaning of used semantics and physical laws iron actually can float, sorry)
 
Until we understand the reason(s).

Possibility one. Selling second hand, recycled and repaired hardware to emerging markets is observable since years. The number of NICs grows and there is a growing number of people able to affort those goods. Hence, computer originally used for for example office needs, which are available in massive quantities through these channels, are repurposed in these markets, in which the people also have a growing desire for entertainment and leasure, as their economic situation becomes more stable.

Here's a possible alternative. In Dota 2 there are boosters, people that create an account to bring to level X, and then sell it. I assume this is the case for all MMOs, why not when it's money. Which means one machine possibly has say 100 accounts originally associated with it. The boosters are mostly russians and chinese. Which again relates to NICs and that decent computers are super expensive in these regions etc., and that they are channeling second hand stuff in various ways into the country (see above).

The results are just counter-intuitive, but certainly not wrong, like "iron can float". (edit: uhm I don't want to go into the discussion that yes according to the meaning of used semantics and physical laws iron actually can float, sorry)

Actually, this is a very important observation considering one very important change recently for Steam.

Steam recently (in the past couple years) have officially released in China. China is the largest market in the world for 2nd hand and refurbished computer gear. At the recycling place I worked at in Japan, one of the major sources of income was salvaging computer components to be shipped to China. Everything from 486's and up would get shipped there to be assembled into working computers for the Chinese market.

Throw in Japan, (another relatively recent market for Steam) which while having many computers has a very small percentage of high end gaming computer and that's going to skew numbers as well.

Considering that Steam randomly surveys anyone that uses their service, it shouldn't come as a surprise then that there will be a resurgence in some really old computer hardware.

Regards,
SB
 
Two views of the same marketplace:

- market share reports describing boards moving off of shelves today
- steam survey showing the shape of the active installed base
Steam Hardware Survey has been shown over and over again to be FUBAR at least when it comes to graphics cards.
The most obvious giveaway:
No, a single HD 8800 -model sold pre-installed in OEM-machines for a year is NOT used more than the several HD 7800 -models sold retail for several years and OEMs for year.
 
No, a single HD 8800 -model sold pre-installed in OEM-machines for a year is NOT used more than the several HD 7800 -models sold retail for several years and OEMs for year.
Again, people here and elsewhere have given logical reason as to why there are counter intuitive results for old hardware in particular, like the sudden uptake in the second hand market, people switching sides (from NV to AMD or vice versa), market share shifts .. etc.

Nobody is arguing the 100% accuracy of Steam numbers, it's a random survey after all, but it has access to the largest number of hardware and users (more than 13 million), and it is very useful in showing general trends and hardware uptake rate. In the end, Steam is not the only one, If you don't like it there is Unity and there is JPR. And all three of them agree on most things.
 
Sorry, with that kind of a sample size and with statistics being a thing (clustering could work wonders here as well IMO), there's little excuse for the survey's lack of accuracy. Far more can be achieved with less resources. It they can't do it properly, they should remove this RNG. That will remove the random talk it spawns as well
 
Back
Top