News Nvidia RTX 30-Series Ranks First in Latest Steam Survey

AgentBirdnest

Respectable
Jun 8, 2022
271
269
2,370
Woah! I honestly didn't expect this to happen anytime soon. Not only because of the shortage for the first ~18 months of the 30-series' existence, but just the prices alone. I don't even mean inflated/scalper prices - I mean the MSRPs are ridiculous. There is no more low-end, and the prices of mid-range are what high-end used to be a few years ago.

It does make sense, with all the price drops over the last couple of months. And I didn't really think about laptop 30-series, which probably makes up a good chunk of that 21.7% (I assume you counted those?)
Still, I just didn't expect it. Maybe I'm biased with my 20-series. : P In any case, it's interesting to see.
 
  • Like
Reactions: bigdragon
An interesting point to extrapolate from this:
The vocal minority telling everyone to not buy yet, and keep waiting for lower prices, apparently arent stopping the vast majority from buying 30xx series cards now that they are available. Which likely means prices will bottom out higher than theyd like.
 

DaDude1

Honorable
Apr 16, 2016
5
0
10,510

bigdragon

Distinguished
Oct 19, 2011
1,145
620
20,160
I'm honestly surprised so many people are willing to accept the higher prices of Nvidia's 30-series. The value proposition just isn't there for me. Maybe my opinion would be different if we had Crysis-like PC exclusives frequently dropping, but we don't.
 
  • Like
Reactions: AgentBirdnest
Jarred,

I don't know how you got to these numbers, but they are absolutely wrong. When I go to the steam survey and only add AMD RX570 and 580 I get to 4.5 %, but In your table you show for all of AMD 3.9%.
Yeah, sorry. I just fixed them. I had an error in one formula row where it added up the wrong rows. It should have been 5.90% for August, not 0.7%. Still, AMD's RX 580 and RX 570 numbers are much lower than what you used.

This is the main GPUs page. 1.28% + 1.04% would be 2.32%.134

I used the API page, under the DirectX 12 API. There's a 0.01% difference on the RX 570 number. I also included the RX 590, 560, and 550 along with the Vega cards (not mobile Vega, though). But then I sum up the entire column and divide the figures by that total to get the final "percentage of the whole" that I reported.

135

You probably used the Vulkan API, which is fubar numbers. I don't know why, but the Vulkan numbers are all basically double what they should be, across all GPUs. If you sum up the Vulkan API columns, they equal 194% to 196%. Here's my "corrected" Vulkan API data, the last few rows. To the right is the adjusted result (divide the main percentage by the total for the column), which correctly sums to 100%. I don't know what Valve omits, but it's problematic that Vulkan sums to 194% or more while DirectX 12 sums to around 90%.

140

Here's the same section using the DirectX 12 data:

139

And I didn't really think about laptop 30-series, which probably makes up a good chunk of that 21.7% (I assume you counted those?)
Yes, laptop data is generally included, for dedicated GPUs. I omit the AMD Vega Graphics integrated solutions, though, as those would skew the Vega + Polaris numbers. If you're wondering, total Nvidia mobile GPUs (at least those that are explicitly identified as mobile) for RTX 30-series accounts for 6.76% of the total, or about a third of all RTX 30-series. For the RTX 20-series, as well as GTX 16 and GTX 10, there's no clear way of seeing how many GPUs are mobile. AMD's integrated Vega Graphics solutions meanwhile account for 2.38% of all surveyed PCs, and most of those are probably laptops.
 
Sep 8, 2022
1
0
10
This data is very skewed. The reality is that after 2 years since the launch of the RTX 3XXX the percentage is 13.05 without counting the laptops. Very sad. They are not selling GPUs.
 
This data is very skewed. The reality is that after 2 years since the launch of the RTX 3XXX the percentage is 13.05 without counting the laptops. Very sad. They are not selling GPUs.
This is where it would be super useful if Steam/Valve were more transparent. Assuming random sampling of all active Steam users in a month, Valve apparently has 120 million users as the population (source here). If 13.05% of those are RTX 30-series desktop card owners, that's 15.66 million GPUs — and another 8.11 million laptop users. I remember a couple of years ago when Nvidia made a big deal about the total number of RTX users (before Ampere launched) surpassing 15 million. Plus, I would wager at least a third of all RTX 30-series GPUs still remain in the mines and wouldn't show up on Steam, so Nvidia has potentially shipped closer to 35 million RTX 30-series cards. They have sold every Ampere GPU manufactured up until the past three months, basically — just not to gamers. And even among gamers, Ampere has sold very well.
 
  • Like
Reactions: KyaraM

russell_john

Honorable
Mar 25, 2018
123
95
10,660
damn those new AMD GPU mumbers @ 1.8% :ROFLMAO:

even with RTX30 crazy prices , Nvidia sells 10 cards for every single AMD 6000 card.

What it really illustrates is how much harder hit AMD was by the chip shortage ..... They were just as hard to get yet Nvidia was selling 10 times more than they were because the availability was better from Samsung than from TSMC. Everyone griped about Nvidia using Samsung 6nm over TSMC 5 nm but they would have been really screwed if they had to rely on TSMC for chips. Looking back now it was the right decision going Samsung
 
What it really illustrates is how much harder hit AMD was by the chip shortage ..... They were just as hard to get yet Nvidia was selling 10 times more than they were because the availability was better from Samsung than from TSMC. Everyone griped about Nvidia using Samsung 6nm over TSMC 5 nm but they would have been really screwed if they had to rely on TSMC for chips. Looking back now it was the right decision going Samsung
Samsung 8nm vs. TSMC 7nm, but otherwise, yes. Also, AMD had consoles and CPUs they needed on TSMC N7, and that used up a ton of the potential capacity. Zen 3 CPU chiplets are only ~84mm^2 and could sell for about $250 each, while Navi 21 is 520mm^2 and could sell for at most $750 or so each. Six Zen 3 chiplets for every Navi 21. AMD did the math and focused more on increasing CPU supply.
 

AgentBirdnest

Respectable
Jun 8, 2022
271
269
2,370
Also, AMD had consoles and CPUs [...]
I know I'm going a little bit off topic here, but this is something I've been wondering about a lot lately. Console SOCs can't be binned, can they? There aren't different levels of PS5. There is just one (well, two, but the digital version has the same SOC.) So, if a core is defective or the SOC can't meet advertised clock speeds, then is the whole chip wasted? They must have a ton of waste from their consoles. Or can the chips be repurposed for, say, office PCs or something?
I dunno if that's in your area of expertise, but I figured you probably know more than the average person. :p
 
I know I'm going a little bit off topic here, but this is something I've been wondering about a lot lately. Console SOCs can't be binned, can they? There aren't different levels of PS5. There is just one (well, two, but the digital version has the same SOC.) So, if a core is defective or the SOC can't meet advertised clock speeds, then is the whole chip wasted? They must have a ton of waste from their consoles. Or can the chips be repurposed for, say, office PCs or something?
I dunno if that's in your area of expertise, but I figured you probably know more than the average person. :p
AMD actually did "binned" PS5 chips of a sort. The Ryzen 4700S is basically a PS5 SOC with the GPU turned off. I wish AMD could have done that chip for PC with the GPU enabled, or at least partially enabled, but I don't know if there were contracts preventing them from doing so or other considerations.

Anyway, I think the PS5 and Xbox Series X chips both have some extra CUs that can be disabled. So they don't need to be perfect chips, but more like 90-95% functioning CUs or something. Then they set the clocks at a level that will be achievable by almost all the chips as well, even though some of them could probably run the CPUs 500MHz faster than the baseline, and maybe the GPUs could run faster as well. Basically, console SOCs are usually defined to the lowest common denominator so that the majority of chips can be binned to that level or higher.
 
  • Like
Reactions: AgentBirdnest

AgentBirdnest

Respectable
Jun 8, 2022
271
269
2,370
AMD actually did "binned" PS5 chips of a sort. The Ryzen 4700S is basically a PS5 SOC with the GPU turned off. I wish AMD could have done that chip for PC with the GPU enabled, or at least partially enabled, but I don't know if there were contracts preventing them from doing so or other considerations.

Anyway, I think the PS5 and Xbox Series X chips both have some extra CUs that can be disabled. So they don't need to be perfect chips, but more like 90-95% functioning CUs or something. Then they set the clocks at a level that will be achievable by almost all the chips as well, even though some of them could probably run the CPUs 500MHz faster than the baseline, and maybe the GPUs could run faster as well. Basically, console SOCs are usually defined to the lowest common denominator so that the majority of chips can be binned to that level or higher.
Oh wow! Hadn't heard of that. That's an interesting chip... doesn't look useful based on that review, but it's interesting just knowing where it came from.

Thanks a lot for that explanation! That makes perfect sense. That satisfies my curiosity. :)
 
  • Like
Reactions: JarredWaltonGPU