Steam's hardware survey just flipped: new top GPU, Windows 11 crashes, Chinese takes the lead

[midian182]

In brief: The latest Steam survey has been released, and Valve has really shaken things up. After admitting last month that VRAM on some graphics cards was not reported correctly, there's been a complete change in the GPU section: the RTX 5070 is suddenly the top card, and RTX xx60 and RTX xx70 GPUs now hold the top 11 spots.

Valve said last week that it fixed an issue where VRAM on some graphics cards was not reported correctly in the Steam survey.

While there is a dedicated VRAM section, it seems whatever Valve has changed affected most of the survey as a whole.

In February, the RTX 5070 jumped from fifth place with a previous 2.87% share to the top of the chart with a nearly 10% user share, replacing the now second-place RTX 4060 in the process.

Now, the first non-RTX xx60 or xx70 card on the list is the RTX 3050 in 12th place.

Most popular GPUs among Steam survey participants during February

Unsurprisingly, the RTX 5070 also saw the month's largest gains, increasing by 6.55%, while other cards made huge gains in February. For comparison, the best-performing GPU in January was the RTX 5060, which increased just 2.5%.

Best-performing GPUs among Steam survey participants during February

Another huge change is in the OS category. According to February's survey, the number of Windows 11 users fell 10% to 56% overall, while Windows 10 was up 12.4% to reach a 40% share.

This is a reversal of the previous trend of gamers, rather than regular users, embracing Windows 11. A recent report from Statcounter showed that 72.78% of Windows desktops were running the latest version, while Windows 10 was down to 26.27%.

Things were less turbulent in the CPU section. February was the second consecutive month in which AMD lost some CPU share as Intel made gains. It was only a minor decline, 0.39%, but two falls in a row has slowed Team Red's momentum in this category, and raised questions over whether it can close the roughly 14% gap on Intel this year.

Another major difference is in participants' system RAM. 16GB was the most popular amount last month with a 40% share, but after jumping a massive 19%, 32GB is now the most popular – what memory crisis?

Simplified Chinese, meanwhile, shot up 30.7% to become the most popular language on the platform (54%) as English fell 14.7% to second place.

Ironically, VRAM, the section Valve said was incorrect, barely changed last month: 8GB is still the most common amount despite falling 0.2%.

We've seen drastically different Steam survey monthly results before, including in March 2025 and October 2023 – things usually return to normal a month later.

Huge month-to-month swings like these are usually sampling noise, not the PC market changing overnight. The survey is opt-in and only captures a subset of active users, so if the prompt reaches a different mix of regions or usage patterns (like more internet cafés one month, more home PCs the next), the percentages can jump.

Small changes in Valve's weighting, normalization, or device classification can amplify this, and short-term events like sales, new hardware, or a hit game pulling in a specific crowd can skew the snapshot further.

Permalink to story:

Steam's hardware survey just flipped: new top GPU, Windows 11 crashes, Chinese takes the lead

 

[winjer]

I bet that the increase in systems with 32 GB of RAM are people panic buying, before things get even worse.

 

[themastergoose]

I see Valve still isn't counting anyone with RX 9000 series cards then?

 

[yRaz]

The thing about it being opt in is that if the sample size is big enough it should compensate for any error.

And the 5070 is the new 60 class card anyway.

 

[ScottSoapbox]

I don't believe these numbers.

Primarily the 5070 at the top spot. That GPU sold so poorly that it was the first to go below MSRP and in the AI crunch panic buying the last to go above MSRP.

No way has the 5070 outsold everything else.

 

[Theinsanegamer]

We see this every year. Whenever the steam algo hits china a bunch of old hardware and software shows up. Then the next month it all drops again.

Really they need to go to a poll of every user every month, with an opt out option, so we can get a 100 percent accurate view of what people have.

 

[Burty117]

Really they need to go to a poll of every user every month, with an opt out option, so we can get a 100 percent accurate view of what people have.

I've been confused why they don't do this, seems logical to me.

 

[Lew Zealand]

My favorite metric on Steam Survey disaster months like this is:

30% of all people stopped speaking English and Russian and Spanish, etc and happened to start speaking Chinese. Put another way: the number of people in the survey speaking Chinese magically went up 129%. LOL

Yeah we can safely ignore everything else this month, it's all completely wrong. But still useful for something: you can now see what the typical Chinese PC has in it.

And hey Steam: Your data was screwed up February 2025 as well, maybe check what's wrong with your Feb survey settings.

 

[ScottSoapbox]

We see this every year. Whenever the steam algo hits china a bunch of old hardware and software shows up. Then the next month it all drops again.

Really they need to go to a poll of every user every month, with an opt out option, so we can get a 100 percent accurate view of what people have.

1. That would be annoying so more people would opt out.

2. It doesn't actually matter. Companies selling the products already have 100% accuracy on sales. It's only forum members that worry about the Steam survey.

 

[Beerfloat]

I see Valve still isn't counting anyone with RX 9000 series cards then?

It took 11 months after the release of the first RDNA4 series videocards to earn one of them a separate listing on the survey. And on the 12th month it sank back into the 'Other' category.

The level of fail of this generation is turning embarassing even by AMD standards.

 

[James321]

It took 11 months after the release of the first RDNA4 series videocards to earn one of them a separate listing on the survey. And on the 12th month it sank back into the 'Other' category.

The level of fail of this generation is turning embarassing even by AMD standards.

That what happens when you always wait for Nvidia to dictate the market.

AMD could have beaten Nvidia to the market by 2 Months. But decided to wait for Nvidia to release their GPU's first. I was going to buy a 9070XT, but then Nvidia ended up releasing their GPU's first and AMD price was ridiculous as they screwed that up as well. So bought Nvidia, probably for the best anyways.

 

[scoffer]

That what happens when you always wait for Nvidia to dictate the market.

AMD could have beaten Nvidia to the market by 2 Months. But decided to wait for Nvidia to release their GPU's first. I was going to buy a 9070XT, but then Nvidia ended up releasing their GPU's first and AMD price was ridiculous as they screwed that up as well. So bought Nvidia, probably for the best anyways.

I have to say that you got that exactly right. AMD could have made a killing, I think. But as usual...f****d it up. Greed, I suppose. I'm not an AMD -hater. I've got a 9070xt myself.

 

[Gravitytr1]

We see this every year. Whenever the steam algo hits china a bunch of old hardware and software shows up. Then the next month it all drops again.

Really they need to go to a poll of every user every month, with an opt out option, so we can get a 100 percent accurate view of what people have.

No.

 

[RaXelliX]

I see Valve still isn't counting anyone with RX 9000 series cards then?

They have been counting them since last summer.

That would be annoying so more people would opt out.

It would still be a bigger sample size because less people would opt-out than do opt-in with the current model.

It took 11 months after the release of the first RDNA4 series videocards to earn one of them a separate listing on the survey. And on the 12th month it sank back into the 'Other' category.

RDNA4 has been present in DX12/Vulkan category since last summer.
The threshold to show up in the main table is higher.

AMD could have beaten Nvidia to the market by 2 Months.

For what?

Nvidia had literally zero stock of all the 50 series card when they launched. It was comical.
And AMD outsold everything they produced prior to launch.
Launching two months earlier than they did would have simply resulted in cards being sold out faster and people willing to buy AMD being left empty handed.

I was going to buy a 9070XT, but then Nvidia ended up releasing their GPU's first and AMD price was ridiculous as they screwed that up as well. So bought Nvidia, probably for the best anyways.

So essentially you're ***** and moan about 9070 XT price that was up to 200 above MSRP , but were more than willing to pay 250+ for 5070 Ti I presume?

Nvidia's 50 series launch was a joke. Stock was nonexistent. 5090's were never seen and cost 3000-4000. 5070 Ti's were regularly 1000+ and 5060/5070 class were not much better either.

Add to that the horrible drivers they took months to fix and that they somehow screwed up again with r590 releases.

I bought 9070 XT last year even tho it was above MSRP and I've had a smooth sailing so far. Drivers have been very solid. Had 5070 Ti been at the same price (it was not, always +150) and had standard 8pin that I dont have to babysit the I would have at least considered it. I mean I came from 2080 Ti and and I had been using Nvidia for the past 9 years at that point.