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

Dec 12, 2020
33
2
35
0
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?
 
Reactions: AtrociKitty

JarredWaltonGPU

Senior GPU Editor
Editor
Feb 21, 2020
917
794
1,770
1
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 [EDIT: 120 million? -- see below] Steam users every month, 1.16% of that could potentially mean over 3 million [EDIT: 1.4 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.
 
Last edited:

spongiemaster

Respectable
Dec 12, 2019
1,352
629
2,060
0
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.


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
 

InvalidError

Titan
Moderator
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.
 

JarredWaltonGPU

Senior GPU Editor
Editor
Feb 21, 2020
917
794
1,770
1
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.
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.
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.
 
Feb 2, 2021
1
1
15
0
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.
 
Reactions: AtrociKitty

Chung Leong

Notable
Dec 6, 2019
387
131
860
0
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.
 
Reactions: bigdragon
Feb 2, 2021
28
16
35
0
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.
 

JarredWaltonGPU

Senior GPU Editor
Editor
Feb 21, 2020
917
794
1,770
1
"Bogus?" Based on what evidence? Gut feel? For a "numbers guy", that's both surprising and disconcerting.

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?
Potentially Bogus was too long, apparently. The HL came from a group discussion of possible ideas. Was it the best? Perhaps not. We went with it, as oposed to "Full Steam Ahead: Nvidia Ampere GPUs Reach 1% of Surveyed PCs" (which was one of numerous other potential headlines). I've removed "Bogus" and put in "Theoretical" now -- does that help you to feel less disgruntled?

The rest, well... why should I feel scorned? At worst, I'm irritated at Valve for not giving better information on the statistics, but I fully understand why it wouldn't do so. There have been multiple cases of the Steam Hardware Survey stats being wrong and/or highly quesitonable over the years -- at one point, internet cafe PCs were getting surveyed far more frequently, which caused the numbers on GTX 1060 to skyrocket (and undercounted AMD CPUs and GPUs) because hundreds of users might use the same PC each month. That's one instance. Valve supposedly 'fixed' the data gathering, but AMD still questioned the results back in 2018 and AFAICT nothing has changed since then. Valve still doesn't explain the methodology behind the survey.

So yeah, until Valve actually says, "We use a purely random sampling of all PCs that log in each month, and we collected XYZ samples out of ZYX active users," the data is at best suspicious and needs to be caveated as I've done. Without information about confidence intervals, margin of error, etc. we're basically given the survey results with zero context and expected to trust Valve. That makes the data, in a word, bogus. (I've followed this stuff for over a decade, so I remember reading about various concerns and issues, and the most common complaint is that Valve doesn't fully disclose what it's doing.)

Let me ask this: What are you trying to attack and/or defend here? Is it the second paragraph, where I call the data into question because Valve doesn't provide context? You don't like my "guessing" -- based off over 15 years of experience -- but here in the forums I've given a specific example of why the RTX 3080 numbers are questionable. Sure, I admit I have no conclusive evidence -- but that's me providing full disclosure, unlike the survey. Do you know who actually has conclusive evidence? Only Valve, full stop. (Probably GOG, EA, Ubisoft, or Epic could present conclusive evidence if their results were wildly different, but none of them even provide limited insight into their user bases.)

In short, I've said that Valve's data is questionable and always has been. Anecdotally I've given examples of why it's questionable. If that's all you've got against this news piece, I'm okay with that. There's no accusation of malfeasance or wilful misrepresentation, because Valve itself has never said people should fully trust the data. All Valve says is, "Steam conducts a monthly survey to collect data about what kinds of computer hardware and software our customers are using. Participation in the survey is optional, and anonymous. The information gathered is incredibly helpful to us as we make decisions about what kinds of technology investments to make and products to offer." That's pretty vague, and intentionally so I'd say. In the early days, the survey also said, "Making these survey results public also allows people to compare their own current hardware setup to that of the community as a whole," but that was removed, which suggests Valve doesn't want people using the data to compare their hardware setup to that of the community as a whole. My take: It's because the data isn't complete / accurate / whatever and Valve doesn't want to be held accountable for it.
 

JarredWaltonGPU

Senior GPU Editor
Editor
Feb 21, 2020
917
794
1,770
1
The author Jared seems pretty bitter.
Let me stop you right there, because you're absolutely wrong. I review GPUs here at Tom's Hardware. I have more than enough GPUs, so I have nothing to be bitter about. The article:
  1. Explains that RTX 30-series GPUs show up on the Steam Hardware Survey now
  2. Looks at the data and notes that the numbers suggest Nvidia's Ampere GPUs may actually be in reasonable supply (potentially 800K RTX 3080 cards on Steam, and another ~300K each for the 3060 Ti and 3090).
  3. Notes AMD's RDNA2 GPUs are nowhere to be found (less than 180K of any GPU, possibly far less)
  4. Points out AMD's CPU numbers look very good
  5. Puts in a clear disclaimer that all of the above is based off the survey results, and we don't know if the results are actually valid. There are many, MANY ways to gather data incorrectly, and Valve has had problems with this in the past, internet cafes being a prime example.
Ampere sales could potentially be far higher than the actually bitter people that can't buy them at MSRP would suggest. (I'm not one of those people as I have at least one sample of every new GPU that has been released, for testing purposes.) I hear from these people all the time, and I'm not discounting their experience. I don't have a Microcenter nearby (not even within about 1000 miles), so I don't know what the in-store experience there has been like. I'm certain some people have been able to buy one of the new GPUs there. I'm equally certain that the number of GPUs sold through Microcenter is less than the number sold through Best Buy (official Nvidia partner for Founders Edition), Amazon, and Newegg. Guess which places have been continually out of stock? Best Buy, Amazon, Newegg, and every other online store. Which suggests perhaps the numbers aren't as good as the survey implies, which again gets back to point 5 where I question whether the data was collected via pure random sampling.

Let me give one example. Suppose the data is nominally randomly sampled, but for "unknown" hardware -- or "new" hardware -- Valve decides to sample at a higher rate. I did programming as a job for years, and this would be trivially easy. Instead of sampling 1% of users (or 0.1% of users), you have a check that says, "Hey, this is new -- let's sample this at a rate of 50% of users." Statistically, you can actually compensate for the higher sampling rate if you want to, while at the same time verifying user uptake of the new hardware. The question isn't whether Valve could do this, but whether it did do this. Only Valve could say, and it tends to maintain radio silence on such matters.

For the record, I write about such things because I'm a passionate computer hardware enthusiast, and it's also my job. I've been accused of "misinterpreting" the Steam Hardware Survey results in the past, arguably rightly so, and I do my best to avoid that these days. There's no bitterness involved in pointing out facts, eg: Newegg has zero stock of the new GPUs, and that has been the case for months now. Newegg has sold GPUs, but the stock isn't remotely close to keeping up with demand. The same goes for Amazon, Best Buy, B&H, Walmart, and every other US online store I've checked. The only places with inventory regularly available are eBay or other sites that allow third party resellers, and their prices are more than double the official MSRPs.
 

JarredWaltonGPU

Senior GPU Editor
Editor
Feb 21, 2020
917
794
1,770
1
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.
I've probably been sampled every year, sometimes more than a dozen times in a year, but then I'm actively logging in from at least half a dozen PCs multiple times per month, every month (often with brand-new hardware). Considering some PCs never seem to get surveyed, I strongly doubt it's anywhere close to 100% of users being surveyed over a year. Probably 10% at best, and even that is unlikely. From a statistics perspective, Valve could sample about 20,000 PCs per month and have more than enough data. That would be 240,000 PCs sampled per year, which would be nothing compared to the installed user base (about 0.2% of active users).

Valve might be sampling at higher percentages than that (perhaps 1% of logins?), but once you get to about 10% of people sampled from any large population, you might as well just sample everyone. For a traditional survey, high sample rates aren't practical, but in this case the population is "all Steam users" and they all have software installed that would make sampling trivial. Since Valve definitely doesn't sample every PC on a regular basis (many people claim to have never been surveyed), that makes me think Valve would be doing more of a reasonable sample size -- 10,000 to 100,000 per month would be ample. (Most election polls only do tens of thousands of surveys, for example.)
 

GenericUser

Distinguished
Nov 20, 2010
223
75
18,790
12
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.
Similar story here. Almost daily usage and I think it's been close to the same amount of time you described (6 or 7 years) since I've last been asked to participate. For comparison purposes, I've stuck with the same CPU and GPU team the entire time. Before the last survey request I'd received though, it felt like I was getting hit all the time for them, going all the way back to when they first started doing them. Then after that last one several years ago, nothing.

Who knows what the story is.
 
Reactions: JarredWaltonGPU

JarredWaltonGPU

Senior GPU Editor
Editor
Feb 21, 2020
917
794
1,770
1
Similar story here. Almost daily usage and I think it's been close to the same amount of time you described (6 or 7 years) since I've last been asked to participate. For comparison purposes, I've stuck with the same CPU and GPU team the entire time. Before the last survey request I'd received though, it felt like I was getting hit all the time for them, going all the way back to when they first started doing them. Then after that last one several years ago, nothing.

Who knows what the story is.
It's precisely these types of stories that make me think Valve isn't doing proper random sampling. The software on your PC could easily detect "nothing has changed" and opt not to sample you again. Meanwhile someone like me who swaps hardware almost daily (because I test hardware) gets sampled on a fairly regular basis. I should keep track of the number of times I get sampled this year. In fact, I think I will! :)
 

InvalidError

Titan
Moderator
I've probably been sampled every year, sometimes more than a dozen times in a year, but then I'm actively logging in from at least half a dozen PCs multiple times per month, every month
I'm just a single user with a single Steam account on a single PC that fundamentally hasn't changed in the last four years and I still get the survey 3-4 times a year every year for about as long as I remember the survey being a thing. I don't see anything special about getting 12+ surveys in a year if you have a multiple PCs all with actively used Steam accounts on 'em.
 

Conahl

Prominent
Apr 24, 2020
219
72
660
0
and this article is part of the reason why the steam hardware survey should be taken with some salt, and not treated as gospel, or a definitive way to see which of the 3 have the most market share with their gpu's and cpu's.
 
Reactions: kylothow

kylothow

BANNED
Nov 13, 2020
5
6
15
0
In the meantime I only see available one of the crappy MSI versions of the 3070 selling for only 1200€; no wonder there are so many around, they are basically gifting them!
Except for a single salty guy defending Nvidia on reddit no one was able to get the card for its retail price in Europe. Nvidia's website hasn't shown any partner retailer on their website since October. Big <Mod Edit> yikes.
 
Last edited by a moderator:
Reactions: Krotow and VforV

VforV

Notable
Oct 9, 2019
262
114
870
1
The GeForce GTX 1060 remains the most popular card, but its total share dropped 1.61%, falling below 10%
It could be all the users with Pascal are finally upgrading
Yes, that was me, I "upgraded" from a GTX 1060 to a GTX 1080 (for a great price) in December 2020. I refused to be the fool who pays these scalper prices to the stores for a new GPU (stores and AIBs are scalping as much as individual scalpers), although I had saved about $500 for a GPU upgrade, at that time.

Sadly there are enough fools who did pay these insane prices and some still do that now... "More money than sense" was never more true than it is today.
 

bigdragon

Distinguished
Oct 19, 2011
767
169
19,160
0
I think that stimulus and trip/vacation money are definitely getting pumped into the GPU market. Finding a new GPU at MSRP -- even the new, elevated MSRP -- is excessively hard. I've had a few 30-series cards in my cart at various stores, but then the checkout process never completes. It's frustrating. Many gamers are choosing to pay more for GPUs by using money from stimulus or cancelled trips, social gatherings, and other events. The scalpers are definitely making bank, and scalped cards are definitely reaching consumers in much greater numbers than anyone would like to admit. The survey results make me extra irritated at Nvidia, AMD, and retailers for not stopping the scalping.

I tend to get prompted for the Steam hardware survey once or twice a year. Sometimes it happens on my desktop, and sometimes when I'm on the laptop. The two results wildly differ, so I'm also curious as to how Valve does the sampling. I do tend to notice the prompt shortly after making a hardware change or firmware update.

Personally, I'll wait out the chaos and upgrade my desktop GPU eventually.
 
Reactions: JarredWaltonGPU

Krotow

Commendable
Oct 2, 2019
1,166
290
1,790
72
... with the second most popular resolution being 1366x768 (yuck)

There goes a ton of cheap laptops with usually i3 and i5 CPU and iGPU. And some gaming laptops with not so good screens. By the way sometimes we are playing ages old UT3 in local network here - one of machines is a laptop with exactly 1366x768 screen.

Personally, I'll wait out the chaos and upgrade my desktop GPU eventually.
Exactly, same there. Scalpers - both stores (they are legal scalpers now) and eBay bastards may go to Hell.
 
Reactions: bigdragon

Gurg

Distinguished
Mar 13, 2013
515
61
19,070
1
Just got sampled again yesterday on my desktop. I usually only play games off line so I can go long periods of time without actually being online with Steam and only use my laptop occasionally when traveling or if I want a second screen to watch two football or basketball games simultaneously. I probably get sampled a handful of times on my desktop annually. I generally update the software including Steam whenever I use my laptop so it only gets surveyed infrequently.

From my experience I would think that laptops may be underrepresented in Steam. Also there are lots of PCs that aren't used for gaming on Steam at all. Obviously Steam users would tend to be gamers, the most serious of, probably buy higher end GPUs.

All of that said: I don't now how you could get more accurate stats on gaming PC configurations than on Steam. Also recognize that Steam asks if you want to participate so many may be leery of participating in any online survey that could indicate personal wealth or possessions.
 
Last edited:

Conahl

Prominent
Apr 24, 2020
219
72
660
0
Also there are lots of PCs that aren't used for gaming on Steam at all
and there are probably the same, if not more notebooks that arent used for gaming on steam, as well, point is?
the notebook i have doesnt even have steam installed on it, but is capable of playing some older games, and, is barely able to play WoW, at just above minimal graphics settings :)

All of that said: I don't now how you could get more accurate stats on gaming PC configurations than on Steam.
are you serious ? did you even read this article ? im guessing, you did not.
how could you get a more accurate stats ? oh i dont know, a separate program that only gets which cpu and gpu is in some ones comp ? ( although unlikely cause of malicious software could be in it, unless from a trusted source ) and if other gaming platforms like Epic, GoG, Ubisoft, and Origin, also released such stats. come on gurg, i know you LOVE to tout steam as the divine gospel and definitive way to claim intel and nvidia are ahead of amd, but come on, at least read the article and Jarred's replies.
 

InvalidError

Titan
Moderator
how could you get a more accurate stats ? oh i dont know, a separate program that only gets which cpu and gpu is in some ones comp ?
A separate program is kind of problematic since such a program will only be run by people who can be bothered to run it. Steam has one of the largest install bases among at least somewhat gaming-oriented stuff.
 

ASK THE COMMUNITY

TRENDING THREADS