GPUs - Steam Hardware Survey Oct 2024

[mackerel]

Latest Steam Hardware Survey has a glitch in its GPU numbers. Assuming this is only a scaling error and that relative positioning is not affected, this can be corrected. And I have done. Notable changes after correction are:
1660 Super: 2.91% (+0.97). No idea what is going on here. Did someone find a boatload of them and set them up in a game cafe?
Arc: 0.19% (+0.19). A new entry as Arc finally makes it on the list. But this raises more questions. Reporting threshold is 0.15%. Could it have jumped that much in one month? Did that many Lunar Lake laptops end up in gamer's hands? Or maybe there were that many tech reviewers? Another possibility might be that Steam didn't separate out Arc before this month, and it was lumped in with "others".
7700 XT: 0.13% (+0.13). Another new entry, marking the 2nd RDNA3 dGPU to show after the 7900XTX. This is below the normal reporting threshold so only made it thanks to the glitch. This does further imply the 3rd placed RDNA dGPU must be below 0.12% share otherwise it would also similarly have made the list.

Most of the rest didn't really change much. Right at the bottom are:
4060 4.05% (-0.53)
4060 Ti 3.31% (-0.35)
3060 Ti 3.29% (-0.28)
Again don't know why these dropped. These do have some numbers behind them, but compare to 3060 for example at 5.76% (-0.10), that hardly changed. Why those in particular?

 

[EarthDog]

I honestly have no idea how this data works. I thought it was random monthly across all users who respond, but, I don't seem to get one that frequently so I feel like it's a subset of users who are asked and respond.

 

[freakdiablo]

I honestly have no idea how this data works. I thought it was random monthly across all users who respond, but, I don't seem to get one that frequently so I feel like it's a subset of users who are asked and respond.

I feel like there's some randomisation on who gets asked, but it's also cyclical. I get asked maybe once a year per machine it's installed on - so I wouldn't be surprised if they pulled ~8% of Steam installs each month, then just use the results independent of who actually agreed.

 

[mackerel]

As long as it is random who gets selected, a subset is adequately representative. I was never good at stats but there's ways to determine how many you need to ask for what level of confidence. Based on Steam's own stats, recent peak online users is 38M, and currently 28M. 3rd party sites claim monthly active users well over 100M. You don't need to ask everyone to get a good picture. With such a large user base, I'd guess 0.1% per month is still plenty. Again, the key is it has to be random to avoid any bias. There is a small risk that people who decline to submit when asked could skew results.

I get asked maybe once a year per machine it's installed on

No hard numbers but I feel that could be about my kinda rate. Obviously there is a selection for active users, as opposed to all installations. If you never log in, you're not going to get picked.

 

[EarthDog]

Like I said, feels random, so, generally a good idea...and could also easily be the reason why it's changed.

I see DX8 cards dropped off the planet on this dataset, lol.

EDIT: That doesn't bode well for extrapolation, these glitched numbers, lol

 

[freakdiablo]

I see DX8 cards dropped off the planet on this dataset, lol.

Well, my Etch a Sketch doesn't support W11 so it can't run the latest steam client either lol.

 

[mackerel]

EDIT: That doesn't bode well for extrapolation, these glitched numbers, lol

To expand on what I did, I have been keeping a spreadsheet of the individual listed GPUs for some time. If you add up all the October results, you get about 130% overall. If you then divide the shown October results by 1.3 to bring the total back to 100%, those are the numbers I used. Compared to September, the majority of GPUs have not significantly moved in the corrected set. Thus I believe the overall dataset is probably ok. That just leaves the 1660 Super as an outlier for some unexplained reason.

I wonder what those DX8 and older GPUs are. We've had DX12 hardware for over a decade now (software for over 9). I think a fair chunk of it might be ancient generation Intel iGPUs but I'm not motivated enough to find the DX version for them.

 

[EarthDog]

I'll circle back next month... sans anomolies and see what's shakin'.

 

[DaveB]

Arc: 0.19% (+0.19). A new entry as Arc finally makes it on the list. But this raises more questions. Reporting threshold is 0.15%. Could it have jumped that much in one month? Did that many Lunar Lake laptops end up in gamer's hands? Or maybe there were that many tech reviewers? Another possibility might be that Steam didn't separate out Arc before this month, and it was lumped in with "others".

That was me?

I went back to my Arc A750 GPU in a cheap AMD build when I needed to sell my 13th Gen Intel/4070 Ti Super setup to get a new refrigerator for the beach house.

And yes, I am active on Steam so I count!

 

[Voodoo Rufus]

1660 Supers were good cards for the money when they were new. I'm sure they're fine for e-sports and 1080P still.