Steam Hardware Survey - GPUs - Feb 2024

[porina]

 

Now we're in March the Feb data has been out for a bit and I took a look at it. Before I get into the numbers, I know, there is uncertainty in how they were obtained and they can't be taken as absolutely representative of anything. Still, it is the best data set I'm aware of for PC gaming. I'd welcome any alternate sources which could supplement this.

 

One thing in particular is I've noticed occasional odd months where NV share spikes a little. Feb 2024 might be one of them. It is not as big as other spikes but it does feel like there is something there. NV overall share went up 2.7 percentage points from the previous month, with the three preceding months spanning less than one percentage point change. So this month's data probably isn't the best to look at for longer term trends especially comparing red vs green.

 

The biggest gains were in the mid range NV GPUs of recent generations, with the 3060, 4060, 3070, 2060, 4070 taking the top 5 spots in percentage point change. While new sales of 4060 and 4070 would be expected, there can't be that many older gens entering the market?

 

Of note is a new entry of the 4070 Super, which means with less than 6 weeks of sales it already has a presence bigger than all individual current gen AMD GPUs except the 7900 XTX.

[Alex Atkin UK]

Always take this with a grain of salt, given the survey is only issued randomly to people.  I've had it maybe 10 times in 19 years of Steam use and some of those have been on devices with only an iGPU.

 

I'm not sure I've ever been asked on my gaming PC as its turned on the least due to being dedicated to gaming vs my daily desktop.

[PDifolco]

I'm amazed how few people have 6600/6700/6800 AMD cards, which are by far the best low/mid range GPU in value...

And more people have a 4090 than a 4080  

[Blasty Blosty]

20 minutes ago, Alex Atkin UK said:

given the survey is only issued randomly to people

Oh I never knew that, guess that's why I didn't get invited this time

[porina]

21 minutes ago, Alex Atkin UK said:

Always take this with a grain of salt, given the survey is only issued randomly to people.  I've had it maybe 10 times in 19 years of Steam use and some of those have been on devices with only an iGPU.

 

I'm not sure I've ever been asked on my gaming PC as its turned on the least due to being dedicated to gaming vs my daily desktop.

I'm not great at statistics. According to Steam themselves the recent peak daily active user count is 35M. According to a 3rd party site monthly active users could be 120M. What is an appropriate sample size to get meaningful data? Even if they picked 0.1% (1 in 1000) daily active users, that's still 35000 samples. Vast majority will not see the survey, but the sample size is probably still plenty.

 

If anyone is good at statistics can provide a better analysis of the required sample size, that would be interesting. I count 113 listed GPUs since the start of 2013, which includes some not currently listed. It does not count any GPUs that have never been listed, so the true number of options may be somewhere higher than that. GPUs never listed include all current gen AMD except 7900 XTX, and all Arc. It gets a bit more questionable if we include mobile. At least NV make it pretty clear.

 

2 minutes ago, PDifolco said:

I'm amazed how few people have 6600/6700/6800 AMD cards, which are by far the best low/mid range GPU in value...

And more people have a 4090 than a 4080  

Judging by people on forums like this is probably a biased sample, and a specific subset of the "people who use Steam" set.

 

All listed RDNA2 GPUs total 3.2%. Value is not just "fps/$". AMD GPUs only look like good value if you apply constraints to what you factor in to that value. Based on many posts on this forum, they only look at "raster fps/$".

 

When the 40 series was released, there was a common comment "if you can afford a 4080, why not spend a bit more and get the much better 4090"? I guess that happened. Plus, there may be dual use buyers who need the 4090 for work, but also use it for play since they already have it. You could say similar for the 7900 XTX. Why aren't the lower models outselling it?

[PDifolco]

5 minutes ago, porina said:

I'm not great at statistics. According to Steam themselves the recent peak daily active user count is 35M. According to a 3rd party site monthly active users could be 120M. What is an appropriate sample size to get meaningful data? Even if they picked 0.1% (1 in 1000) daily active users, that's still 35000 samples. Vast majority will not see the survey, but the sample size is probably still plenty.

 

If anyone is good at statistics can provide a better analysis of the required sample size, that would be interesting. I count 113 listed GPUs since the start of 2013, which includes some not currently listed. It does not count any GPUs that have never been listed, so the true number of options may be somewhere higher than that. GPUs never listed include all current gen AMD except 7900 XTX, and all Arc. It gets a bit more questionable if we include mobile. At least NV make it pretty clear.

 

Judging by people on forums like this is probably a biased sample, and a specific subset of the "people who use Steam" set.

 

All listed RDNA2 GPUs total 3.2%. Value is not just "fps/$". AMD GPUs only look like good value if you apply constraints to what you factor in to that value. Based on many posts on this forum, they only look at "raster fps/$".

 

When the 40 series was released, there was a common comment "if you can afford a 4080, why not spend a bit more and get the much better 4090"? I guess that happened. Plus, there may be dual use buyers who need the 4090 for work, but also use it for play since they already have it. You could say similar for the 7900 XTX. Why aren't the lower models outselling it?

I'd rather think people are just ignorant or enthralled by NVidia ads...For anything below a 3080/4070 raster is the only thing that counts, lower tier cards suck at RT. And for 1080p cards DLLS is just bad as well (too low render res)

And even on hi tier cards most games have no noticeable RT if any

Your point makes sense for the 4090 tho, at least on release, now 4090 has reach scalper level pricing

 

[porina]

6 minutes ago, PDifolco said:

I'd rather think people are just ignorant or enthralled by NVidia ads...For anything below a 3080/4070 raster is the only thing that counts, lower tier cards suck at RT.

Things have moved on since the introduction of RTX. Current gen console games are featuring RT more often with one title where there is no option to turn it off. Their hardware RT performance is pretty low but devs have got better at using it over time. I think it only a matter of time before we see a mandatory RT requirement game on PC as it reduces workload for them to drop support for legacy paths. We already passed more the 50% of Steam Hardware Survey having a hardware RT capable GPU (everything RTX, RDNA2 or newer). This will likely be a minimal/basic implementation and it will take longer for fully path traced global illumination to be commonplace.

[PDifolco]

29 minutes ago, porina said:

Things have moved on since the introduction of RTX. Current gen console games are featuring RT more often with one title where there is no option to turn it off. Their hardware RT performance is pretty low but devs have got better at using it over time. I think it only a matter of time before we see a mandatory RT requirement game on PC as it reduces workload for them to drop support for legacy paths. We already passed more the 50% of Steam Hardware Survey having a hardware RT capable GPU (everything RTX, RDNA2 or newer). This will likely be a minimal/basic implementation and it will take longer for fully path traced global illumination to be commonplace.

That's for consoles, but won't happen on PC in the next 5 years

RT is still mostly a NVidia gimmick with very few games really making advantage of it (sayr CP2077 aaaand... dunno  )

Plus console players play on big TV monitors from afar, where upscaling isn't that of an issue