Nvidia's Ampere GPUs Steam Ahead to 1 Percent Theoretical Market Share

News
By Jarred Walton
last updated

There sure are a lot of bots and miners running Steam!

RTX 30-series GPUs
(Image credit: Tom's Hardware)

Valve has just updated its Steam Hardware Survey with results from January 2021. If you believe the numbers for videocards, Nvidia's RTX 30-series GPUs now account for over 1% of the total gaming market on Steam. These are some of the best graphics cards, but demand is so high (and supply is so low) that finding one for sale is virtually impossible, and prices are much higher than the launch MSRPs.

I've followed the Steam Hardware Survey for a long time, wondering at the statistics behind the data. The past few months give me (even more) reason to suspect it isn't a proper random sampling of users, which means no one should attempt to draw any meaningful conclusions. Valve has never revealed any details of how the survey gets conducted, but I suspect (based on being sampled on three different PCs all within a day or two of each other, all of which were using a 3080 card for testing) there's a higher chance for it to ask for someone's hardware details if it doesn't recognize the graphics card. This means new cards like the RTX 30-series are much more likely to get included. However, that's just a guess, and it's possible Valve is actually doing a proper random sampling and simply hasn't made that fact public. Still, without a clear explanation of the methodology, we shouldn't take these figures as any true indication of the distribution of various GPU models or other hardware, even among Steam users. Regardless, the numbers are still interesting and fun to gawk at, wherever they come from.

The biggest news is that the GeForce RTX 3080 now sits at 0.66% of PCs surveyed. That's up 0.18% from December, which was up 0.25% from November. The RTX 3080 has been on the charts for three months now, steadily gaining share, and is closing in on the GTX 1660 Ti. The other two Ampere GPUs to show up on the charts are the GeForce RTX 3060 Ti, sitting at 0.27%, and the GeForce RTX 3090, which has 0.27% and theoretically has more hits on the survey than AMD's previous-gen Radeon RX 5700 (the non-XT model).

AMD's RDNA2 Radeon RX 6000-series GPUs are not yet part of the list, though there's still the nebulous 'Other' category with 9.29% of all GPUs. Presumably, the Radeon RX 6900 XT, Radeon RX 6800XT, and RX 6800 all fall into that category, with less than 0.15% of the total each. Interestingly, Nvidia's GeForce RTX 3070 also fails to show up on the chart, so for now at least, the 3090 appears to be ahead in terms of Steam use.

(Image credit: Valve)

The top of the charts is also interesting. The GeForce GTX 1060 remains the most popular card, but its total share dropped 1.61%, falling below 10% for the first time in I don't know how long. (For the record, I actually do know how long: The last time the GTX 1060 was listed at below 10% was in September 2017. GTX 1060 supposedly peaked at just over 20% in December 2019, if you're wondering.) It could be all the users with Pascal are finally upgrading, or maybe just the Internet cafes have finally decided to move on from the 1060. Or it could simply be a normal variation in the sampling, as the GTX 1050 Ti use is up 0.61%, and GTX 1050 is up 0.44%.

Also of note is that several other reasonably popular GPUs have shown large dips this past month. The RTX 2060 is down 1.16%, 2070 Super is down 0.53%, and RTX 2070 is down 1.1%. The generic label of "Nvidia Graphics Device" is also down 0.68% (probably because formerly 'unlisted' GPUs like the 3060 Ti and 3090 are now showing up separately).

We're perhaps getting too far into the weeds, given we don't know the actual collection policy and statistical accuracy of any of the data. At best, this could be a random sample of Steam users from the past several months. At worst, it's a biased sample of Steam users. Either way, it doesn't account for any hardware that's not used with Steam. Still, it does bear at least some semblance to what we'd expect to see in the gaming market.

For instance, AMD CPU usage is up 3% last month relative to Intel usage, which is down 3%. AMD's total for CPU use is now 28%, the highest it's ever been. That makes perfect sense, as Intel's desktop CPUs have been stagnating on 14nm, while AMD's Zen 3 (Ryzen 5000) series CPUs are now at the top of our best CPUs list and lead our CPU Benchmarks hierarchy. Meanwhile, 1920x1080 still claims the lion's share of resolution usage, at 67% of the total, with the second most popular resolution being 1366x768 (yuck), and 2560x1440 usage represents just 7.5% of the users surveyed.

Let me close by once again calling on Valve to do the right thing and provide a clear statement on the statistics behind the survey. If it's a random sampling, tell us so we (and more importantly, game developers) know we can put more confidence in the numbers, and tell us (approximately) how many PCs were surveyed. And if it's not a proper statistical analysis, then fix it. Thousands of undergrad statistics students could explain what needs to be changed. It would also be great to allow numbers nerds like me to get the full list of GPUs, even for those with only a 0.01% share. And as long as I'm making wishes that are unlikely to be fulfilled, please fix all the PC component shortages, especially on the new video cards. Because hey, dreams are free!

Jarred Walton
Jarred Walton

Jarred Walton is a senior editor at Tom's Hardware focusing on everything GPU. He has been working as a tech journalist since 2004, writing for AnandTech, Maximum PC, and PC Gamer. From the first S3 Virge '3D decelerators' to today's GPUs, Jarred keeps up with all the latest graphics trends and is the one to ask about game performance.

See more GPUs News
28 Comments Comment from the forums
  • TimmyP777
    Was this written by a scorned child who cant land a GPU? We are talking ONE PERCENT, with each card being a FRACTION OF ONE PERCENT. How in gods name do you extrapolate an entire article about this?
    Reply
  • JarredWaltonGPU
    TimmyP777 said:
    Was this written by a scorned child who cant land a GPU? We are talking ONE PERCENT, with each card being a FRACTION OF ONE PERCENT. How in gods name do you extrapolate an entire article about this?
    One percent of all Steam PCs is a potentially massive number. Considering there are several hundred million Steam users every month, 1.16% of that could potentially mean over 3 million RTX 30-series GPUs sold. Except, we don't know how Valve actually gathers the statistics, so it could be a far lower number than that as well.

    There are only 22 GPUs total on the survey that have 1% market share or more. Most of those GPUs cost less than $200, and the only GPU with an MSRP of $500 more in that category is the RTX 2070 Super (2.31%). The 3080 and 3090 are extreme GPUs, but they're already about 1% of the market (according to Steam's questionable statistics).

    The 3080 launched at $700 and has been selling for far more than that at most places, so 0.66% is actually very impressive -- it took nearly a year for the RTX 2080 to hit that percentage, and RTX 2080 Super broke 0.66% in July 2020 (one year after launch). For the 3080 to hit that mark in four months, even while people bemoan the shortages, suggests the supply may not actually be as bad as we think.

    GPUs with 1% or less share include every AMD GPU except RX 580, RX 570, Vega 8, and RX 5700 XT. Note that the 5700 XT JUST BROKE 1% FOR THE FIRST TIME this month. So AMD's best previous-gen GPU has never had more than 1% until now.

    The RTX 2080, 2080 Super, and 2080 Ti are all sitting at less than 1% right now. In fact, 3080 is now nearly equal to RTX 2080 Ti on the list -- a 2.5 years old previous top-of-the-line GPU.

    So, sorry if you don't want to think about this stuff. Don't resort to name calling next time.
    Reply
  • spongiemaster
    JarredWaltonGPU said:
    One percent of all Steam PCs is a potentially massive number. Considering there are several hundred million Steam users every month, 1.16% of that could potentially mean over 3 million RTX 30-series GPUs sold. Except, we don't know how Valve actually gathers the statistics, so it could be a far lower number than that as well.
    Valve reported 120 million monthly active users in 2020, the first time they cracked 100 million. Not sure where the several hundred million number comes from, but it isn't accurate according to Valve themselves.

    https://www.statista.com/statistics/733277/number-stream-dau-mau/
    The 3080 launched at $700 and has been selling for far more than that at most places, so 0.66% is actually very impressive -- it took nearly a year for the RTX 2080 to hit that percentage, and RTX 2080 Super broke 0.66% in July 2020 (one year after launch). For the 3080 to hit that mark in four months, even while people bemoan the shortages, suggests the supply may not actually be as bad as we think.
    According to some Oracle data engineer who wrote a script to track sales, 50,000 Ampere GPU's have been sold on Ebay and StockX alone. In comparison, less than 3,500 RDNA2 cards have been sold on those 2 sites. So who really had the paper launch here? Unless scalpers are just getting stuck with RDNA2 cards because no one wants to buy them at inflated costs.

    https://www.techradar.com/news/profiteers-resold-nearly-50000-nvidia-rtx-30-series-gpus-on-ebay-and-stockx
    Reply
  • InvalidError
    I doubt the GPU has any bearing on sampling selection. I have a lowly GTX1050 and get asked if I want to submit system stats for the survey every 3-4 months. The GTX1050 hasn't been new for a very long time and neither has my i5-3470. Going by CPU and GPU results, a huge chunk of Steam users are running ancient PCs and laptops. You wouldn't see that if survey picks were biased toward unknown presumably new hardware.
    Reply
  • JarredWaltonGPU
    spongiemaster said:
    Valve reported 120 million monthly active users in 2020, the first time they cracked 100 million. Not sure where the several hundred million number comes from, but it isn't accurate according to Valve themselves.
    https://www.statista.com/statistics/733277/number-stream-dau-mau/
    That's the whole problem: Valve refuses to divulge the underlying sampling methodology and statistics behind the survey. Is it only users who logged in last month? Probably. How many was that? You link to a site that shows "Number of monthly active users of Steam worldwide from 2017 to 2020" and lists 120 million for 2020. Is that average per month, maximum over the sampling period, or something else? We don't know. January 2021 could have been higher or lower as well. But fine, I gave an estimate of the total number of Steam users / accounts. Even if it's only 120 million active per month, that's still about 1.4 million RTX 30-series GPUs -- theoretically.

    When people claim it was a paper launch, and the hardware survey suggests potentially over a million RTX 30-series GPUs are in the wild -- including roughly 800,000 RTX 3080 cards alone -- I would unequivocally call that a reasonable launch. But the number could be far lower than that, and I suspect it is. Until (unless) Valve explains the data better, it's really just a rough figure that we can look at and hope it's accurate, even while suspecting it's not.
    InvalidError said:
    I doubt the GPU has any bearing on sampling selection. I have a lowly GTX1050 and get asked if I want to submit system stats for the survey every 3-4 months...
    I've been sampled a lot of times over the years, on multiple different PCs. At one point in the past four months, I had at least three of the six or so PCs I use during testing queried about their hardware in the same week. That is, in my experience, a far higher number than it should be. It stood out to me as unusual, because I've tested multiple PCs within a month for years. Every so often, one gets asked to participate in the hardware survey, but in this case it was all within a day or two. Random chance? Possibly, but every PC had a 3080 in it for testing at the time. Maybe it was my user account that got "selected" ... but it wasn't every PC I tested on. Which ultimately leads to the same problem: We don't know what Valve is doing to gather this data.

    Yes, obviously it samples lower spec PCs. And the nearly 10% in the other category suggests either there are tons of really old GPUs floating around, or a bunch of different hardware identifiers -- neither of which really makes sense, but I digress. Ideally, Valve is doing random sampling of a reasonable survey size out of all users that log in within any given month, without any regards to geography or other factors (plus some stuff to ensure the same PC, e.g. at an Internet cafe, doesn't get counted multiple times as separate hardware). If that's the case, then the data presented is actually reasonably accurate -- within a few percent deviation most likely. That would mean AMD's market share for GPUs continues to be abysmal, no matter what anyone else suggests. But if that's not the case, only Valve (and other gaming platforms like Epic, GoG, Ubisoft, and Origin) are likely to have the true data, and the others don't share it.
    Reply
  • Maxxrox
    Nvidia's Ampere GPUs Steam Ahead to 1 Percent Bogus Market Share

    "Bogus?" Based on what evidence? Gut feel? For a "numbers guy", that's both surprising and disconcerting.

    The past few months give me (even more) reason to suspect it isn't a proper random sampling of users, which means no one should attempt to draw any meaningful conclusions.

    Assertion of malfeasance or wilful interpretation of data based on suspicions? Love me some evidence-based journalism. Who is your editor and why did she let this article go to print on Tom's Hardware?

    However, that's just a guess

    Yes. Yes it is. Please structure your title, thesis and language accordingly.

    While I agree that transparency on reporting is something to be desired, Timmy is absolutely accurate in the statement that you sound like a scorned child. The journalistic merit of the article was pretty well-buried; the pettiness and frustration, however, was on full display.

    So, sorry if you don't want to think about this stuff.

    I'll leave that there.
    Reply
  • InvalidError
    JarredWaltonGPU said:
    Ideally, Valve is doing random sampling of a reasonable survey size out of all users that log in within any given month
    Since the survey requires little more than a click from the end-user, I wouldn't be too surprised if the prospective sample size was close to 100% of active accounts in aggregate over the course of a year.
    Reply
  • Chung Leong
    JarredWaltonGPU said:
    The 3080 launched at $700 and has been selling for far more than that at most places, so 0.66% is actually very impressive -- it took nearly a year for the RTX 2080 to hit that percentage, and RTX 2080 Super broke 0.66% in July 2020 (one year after launch). For the 3080 to hit that mark in four months, even while people bemoan the shortages, suggests the supply may not actually be as bad as we think.

    Ampere was launched in an abnormal time. People's ability to spend their environment dollars has been severely curtailed. A lot of money is funneled into electronics that would in normal times would be spent at bars and restaurants or on oversea trips. $700 really isn't a whole lot of money. I've spent that much at a New York bar in one night.
    Reply
  • jpe1701
    I've logged into steam on most days for the past 6 years and have never been asked to participate in the survey. I've had both Nvidia and AMD gpus and Intel or AMD CPUs over that time too so it would be interesting to know how they choose who is included.
    Reply
  • scottsoapbox
    The author Jared seems pretty bitter.

    I was able to buy an RTX 3070 on Amazon using a simple in-stock email notification service. And apparently, anyone near the vaunted Microcenter has been able to buy them. And that's just at MSRP.

    Scalpers wouldn't be scalping if the cards weren't selling. It sucks that it's happening but it doesn't mean those cards are getting to gamers - it just means gamers have to pay more to get them.
    Reply