You’re comparing two completely different things.
JPR tracks market share in terms of recent sales. Steam reports market share in terms of installed base.
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
RTX 3xxx performance speculation
- Thread starter Nebell
- Start date
You’re comparing two completely different things.
JPR tracks market share in terms of recent sales. Steam reports market share in terms of installed base.
Bottom line is we have no other source of data on GPU installed base. It’s fine if you don’t like what the numbers are saying but it’s surely more reliable than people’s feelings.
oldmanbal
2[H]4U
- Joined
- Aug 27, 2010
- Messages
- 2,613
I'm not sure why you are just making things up to argue, this is turning into a wccftech comment section, enjoy.
Viper87227
Fully [H]
- Joined
- Jun 2, 2004
- Messages
- 18,017
This is exactly my take.
We're all smart enough to know it's not an exact snapshot of the install base. It's the best source were going to get for this information though. It's accurate enough to make reasonably sound estimations of where the market is at.
Given the market penetration of steam among gamers, it's likely to be quite decently accurate representation of gamer HW.
It has also been running long enough, that it has had some issues and worked past them.
Also if their sample size wasn't large enough we would see evidence of this in wild fluctuations month to month. Where cards would keep changing position back and forth. We don't really see this.
Nothing is perfect, but there is no reason to expect there is anything seriously tilted in the data. It is a good yardstick to see what HW installed base that gamers are using.
There was a tone in your post that hoped for/expected weak sales, would result in NVidia lowering prices.
WCCFT comment section level posting, is calling the CEO a liar when he reports thing different from your expectations.
Gideon
2[H]4U
- Joined
- Apr 13, 2006
- Messages
- 3,556
Clearly not otherwise steam would show higher numbers of AMD cards installed. Simple fact is JPR numbers are a much more realistic look at the current market and you can track it over the years. Steam survey is about as accurate as a political poll these days.
What data are you basing that claim on?
No it is not possible to translate JPR sales numbers over time into installed base numbers. For any given sale of a new card you don’t know if it’s replacing an installed card. You certainly don’t know which card models are being replaced.
defaultluser
[H]F Junkie
- Joined
- Jan 14, 2006
- Messages
- 14,398
No, but you can look at one trend vs another, and see how they correlate data INDEPENDENT OF ONE ANOTHER.
Anyone with half a brain on his head can see that AMD's lack of sustained sales growth on JPR also correlates with AMD's lack of sustained product installed on Steam (although Steam will take another 6 months to really show,)
We're not in some Mining Wunderland where every second card sold goes into a datacenter - most of these are going to consumers .
Last edited:
Hard to tell if you’re agreeing or disagreeing with anything I said. To summarize....
1) JPR and Steam measure different things.
2) You can’t use short term JPR data (sales) as a proxy for long term Steam data (installed base).
Of course over the long term higher sales directly leads to higher installed base.
Gideon claims that Steam should be showing higher installed base for AMD. Why?
Gideon
2[H]4U
- Joined
- Apr 13, 2006
- Messages
- 3,556
If they are selling at 30% of the market then you should see something near that level in steam as well if it was accurate. Since gross margin is not that spectacular on AMD's side they are not selling a ton to the professional market and in Nvidias case they are no longer making the most from the gaming side as it has slid under 50% of their profit. Plus with Steam you only get a look at people that actually use it and many don't use steam. But if you believe Steam is accurate then someone else should have ended up as President a few years ago. This is also why Nvidia and AMD refer to JPR charts in their own investor meetings and not Steam. Steam is good for watching trends in the gaming industry but hardly means anything when it come to market share.
How many quarters has AMD had >=30% ?? The link you showed had 25% last year, 27% this quarter. Install base will depend on sales over time, not just a couple of peak quarters. So it looks more like AMD was traditionally hovering around 25% on JPR numbers as well, so not that different.
Not sure what margins have to do with it. For Market share both JPR and Steam are counting cards, not revenues.
You haven't presented any kind of sound theory, let alone evidence, of why AMD would be over-represented among people that don't use steam.
JPR would be better for Investor meetings, because it is current sales, not installed base....
Have you looked at the data? Seriously?
All Video Cards
1060 variation from Dec to Jan report a change of 2.87% (16.38% going to 13.51% of all video cards represented by 1060s) -> That usage change is more than all 1080 usage for January, about four times more than all 2080Ti's - what those 1060's exploded at a high rate, turned off? -> This does not indicate a high or accurate month to month survey. The whole report has these types of discrepancies.
The idea that since this is the only report we have to go by thus makes it legit is also laughable. It has to stand on its own merits. Granted it does indicate overall penetration but questionable accuracy is my point and the data itself shows this I think.
The GTX 1060 launched May 27, 2016 which is almost four years ago. While I have as yet to be interested in replacing my 1070 that is only because I play games that are laughably easy to run (Civ V, Beyond Earth, UT 2003, DIablo III, WoW, and the venerable Moo ][ in DOSBox).
EDIT to add: The 1080 is quite a bit faster than the 1060 and is not as likely to be experiencing upgrade pressure.
EDIT to add: The 1080 is quite a bit faster than the 1060 and is not as likely to be experiencing upgrade pressure.
OK:
GTX 1660 was at 1.42% of all video cards used for Steam gaming in Dec, but then 1.18% of all video cards used for Steam gaming in Jan. Basically 20% less 1660's in one month surveyed -> One would expect 1660 numbers to actually to go up since they are still selling. This hints at an inconsistent, high variation data survey.
Not to mention some other bizarre stuff such as 5700XT's only being DX 12 cards in the month of Dec at .47% - I guess they did not exist prior to that as DX 12 cards while they did for All cards.
Last edited:
One would actually expect 1660 sales to start dropping after the 1660 Super was released for not much more, and dropping sales, can look like a regression if other cards are selling better.
I don't see any cause for alarm in the fluctuations, the don't really change the big picture.
You can do a rolling 3 month average if you want to smooth the fluctuations.
Dayaks
[H]F Junkie
- Joined
- Feb 22, 2012
- Messages
- 9,774
Steam has one billion accounts. A few percent (I realize it’s a randomly sampled subset) is still a ton of GPUs. We know nVidia sells millions a month, we just don’t know the exact number.
Some people replace their cards ever few generations. I have my system, and I usually pass my old cards along to family/friends. I think RTX is doing fine.
Also some rando nVidia charts above, not sure if posted somewhere yet but found them interesting.
Some people replace their cards ever few generations. I have my system, and I usually pass my old cards along to family/friends. I think RTX is doing fine.
Also some rando nVidia charts above, not sure if posted somewhere yet but found them interesting.
Sells may start dropping but the card usage should not suddenly drop. There are way more examples one can find even worst. Yes averaging will smooth out the inaccuracies if the data is consistently taken, if not then it would also not increase accuracy. This is not sells but Steam hardware usage which may or may not reflect population of cards. AMD may have many more business orientated builds over Nvidia due to AMD having CPUs/chipsets which can be very significant but not indicated by the Steam Survey. Anyways enough said, I hate getting off topic but this keeps coming up and up. The data if looked at makes it very questionable is my view.
The only reason there is any argument, is that some AMD fans have "feelings" that the Steam Survey is somehow short changing AMD.
There are never any sound reasons involved, just attempts to attack the survey as a whole, because it disagrees with those "feelings".
Once you consider the different things they cover, JPR results, only help confirm the Steam Results. Since Maxwell, NVidia has largely been hovering between 70% and 80% on JPR AIB survey with a couple of dips below 70%. This seems to match well with Steam install base, given the differences in coverage.
There really should be no argument about the data here.
Perfect? No.
Representative? Obviously YES.
Last edited:
I had to laugh at the very first chart - Revenue - We hiked the price outlandishly and look - we made much more money hehehehahaha. Hopefully we will have some decent competition beyond current RNDA.
The second chart shows 2% Turing, 50% Pascal and 48% other generations for installed based. Really??? So Pascal at the time of this chart was 25x more installed than Turing yet Nvidia is making a bundle with Turing?
Third Chart: You see, everyone is happy in spending more money to buy our products, we make them happy by charging way more and giving them the same performance.
Are these actually real?
Note that those slides are almost a year old. They are from March 2019:
https://www.game-debate.com/news/26...says-majority-happy-to-pay-for-price-increase
So the only argument is if some feel AMD have feelings? WTF? No not at all, at least in my case, the data or survey is very limited and folks take it way out of context over and over again. Gideon was right on the money for those looking at actual or more accurate data dealing with Nvidia vs. AMD for sells.
Thanks, I was scratching my head a little here.
I addressed Gideons post, and both you, and he exemplify my point.
You both feel AMD results should be higher based on nothing. You have no grounds. He rambles about margins and people that don't use Steam, without explanation nor evidence why people who don't use Steam, would by any more likely to have AMD cards than those who do.
Some people don't use Steam, but there is no reason why AMD users would in particular avoid Steam.
You tried the argument, that the monthly fluctuations somehow make it invalid.
Do you really see the JPR results as that much better? They still show NVidia hovering between 70-80% market share since Maxwell.
Gideon
2[H]4U
- Joined
- Apr 13, 2006
- Messages
- 3,556
You have not proven that Steam is anymore accurate either, other then your holier then thou belief. I suppose you also believe that Turing is selling as well as Pascal despite obvious evidence to the contrary. They may be making the same profit but that is due to the higher prices on Turing compared to Pascal. As for Steam plenty of people buy direct from Origin, Bethesda, Blizzard and even Rockstar now and some only purchase a game on a rare occasion. Hell quite a few people don't even know what Steam is and we also have no metrics on how Steam polls it's users and at what %. I will take JPR as a more accurate picture of the current market then a Steam hardware survey, I only have seen Steam used in fanboy arguments and never in professional discussions of market share. Yeah Nvidia has owned 60% or better of the market for awhile and thats why you see 1000+ dollar gaming cards, plus you can see on this forum alone that many chose to stick with their current card rather then upgrade due to the increase cost for minimal improvement in performance. Another reason Nvidia and AMD are looking more toward the professional space and less toward the gaming market these days. JPR is a far more accurate look at the market but your never going to get exact numbers no matter what method you choose and that is on purpose by the manufacturers. But you go on using Steam as your base until it no longer suits you, plenty have done that before.
But I am not claiming either is more accurate. Given the differences in what they report, they essentially show the same thing, and reinforce each other (as I said before).
You OTOH, called the Steam HW survey "garbage", so the onus is on you to prove your charge.
No. Next.
Again, completely irrelevant unless there is a reasonable expectation that ratio of GPU types would be significantly different on different services, and there is no such expectation.
Not sure where you get this "far more accurate" from, but these are sampling type analysis as well, based on outlets willing to talk to them, and some likely don't. Back when Apple reported unit volumes, these types of analysis routinely had significant errors.
Just like Steam Survey, JPR is the right ballpark, but I don't see any basis for calling them "far more accurate".
In the overall numbers, after factoring differences in what they report, they show about the same ballpark, so I don't really care which one you want to use for overall market share. That's actually the least interesting thing to me.
Steam Survey is MUCH more useful because it reports on the individual GPUs. So we can see the relative market share of 1660 Ti vs 2060 for instance. Or we can total up all the RTX GPUs and see what kind of RTX market share there is with gamers. JPR doesn't give us any of that capability, it's just a simple market share number.
Last edited:
Where did you get I had a feeling that AMD data is incorrect??? Your feelings? lol. No, not my point, data scatter is too great to take the Steam survey as being accurate for Nvidia or AMD. Plus their are many different markets outside of Steam which are not represented with the Steam results. Yet folks over and over again use Steam results as some kind of reliable data and use it rather stupidly for or against Nvidia or AMD. Good grief, waste of time.
Well if Steam results are useful then RTX is an utter waste of die space/money/time: Very few on Steam percentage wise are wasting their time it seems with RTX, tell me the games below and which one's have RTX and for those that do, do folks actually use it? These are ranked top to bottom the games being played most:
Last edited:
OutOfPhase
Supreme [H]ardness
- Joined
- May 11, 2005
- Messages
- 6,579
I am fascinated by the arguing against Steam survey results. They are what they are: a statistical sampling of steam users. No more, no less.
It is of course not unreasonable to assume it's a decent perspective of gamer machines in general, unless you can provide data that there are huge outcroppings of one vendor who meticulously avoid Steam.
That's what MangoSeed has been asking: if you feel this does not represent installed share, please explain why it would favor one vendor over another in reality. Anecdotes of "it didn't ask me" are not really useful. Well, unless the assertion is that Valve itself is biased for some reason. If so, please just assert that and we can move on.
Also, it has nothing to do with perceived / marketed merits of the hardware. It's a "what is actually in use" metric.
It is of course not unreasonable to assume it's a decent perspective of gamer machines in general, unless you can provide data that there are huge outcroppings of one vendor who meticulously avoid Steam.
That's what MangoSeed has been asking: if you feel this does not represent installed share, please explain why it would favor one vendor over another in reality. Anecdotes of "it didn't ask me" are not really useful. Well, unless the assertion is that Valve itself is biased for some reason. If so, please just assert that and we can move on.
Also, it has nothing to do with perceived / marketed merits of the hardware. It's a "what is actually in use" metric.
Last edited:
Gideon
2[H]4U
- Joined
- Apr 13, 2006
- Messages
- 3,556
It's hardly reliable and you just have to take their word they got it all fixed now. https://techreport.com/news/33604/s...shows-big-changes-after-valve-fixes-counting/
Bingo. At this point I think people’s emotions are simply getting the better of them.
OutOfPhase
Supreme [H]ardness
- Joined
- May 11, 2005
- Messages
- 6,579
A scientist correcting their model to fix an observed inaccuracy does not mean that scientist is unreliable. In fact, it tends to imply the opposite. That’s the scientific process itself.
Do you have solid data showing it is still unreliable for what it purports to be? Please share it widely so the model can be corrected. I strongly suspect they aren’t doing this in the hopes of generating incorrect data.
Gideon
2[H]4U
- Joined
- Apr 13, 2006
- Messages
- 3,556
You wanted to see proof of a flaw and now want to move the goal posts. https://www.techradar.com/news/amd-...-why-does-steam-hardware-survey-not-show-that , https://www.extremetech.com/computing/258913-what-the-hell-happened-to-the-steam-hardware-survey , https://www.guru3d.com/news-story/steam-hardware-survey-shows-impressive-gains-for-amd.html , https://www.gamingonlinux.com/artic...-look-at-daily-and-monthly-active-users.12887 , https://seekingalpha.com/article/978681-graphics-wars-analyzing-the-steam-hardware-survey-numbers
It's all over the damn place, yet you all think it's accurate.
Many of those links are repeats of the same 2017 issue. The rest are mostly nonsense (Scott Herkelman whining about 2017 issue).
The thing is Steams Monthly update, and what it tracks, gives us the means to Sanity Check it's data, that we don't have with JPR.
Which is how the 2017 anomaly was easily seen by everyone.
JPR data has no sanity check. It could be wildly inaccurate, and there would be no way of knowing.
Being more ignorant of JPR inaccuracies, doesn't mean JPR is more accurate.
Last edited:
It’s infinitely more accurate than wishful thinking. You’re railing against Steam but aren't offering any alternative.
limitedaccess
Supreme [H]ardness
- Joined
- May 10, 2010
- Messages
- 7,594
Just going to comment in general sense this discussion thread is all over the place -
Whether or not Steam, or any poll/survey, is accurate or not isn't really important as it isn't really a binary distinction. What is actually important is how accurate or representative it is within a high probability, and if that is suitable enough for the argument/discussion taking place. Even if we feel say Steam has Nvidia at 70% market share isn't "accurate" does it really matter if it differs from the reality which is at 68%? Well it depends on the context of the discussion. The likelihood is also very low that it is off by >10%.
With regards to Turing pricing people should consider what the broader market situation was in 2018 when Turing launched versus 2019 with both Navi and the Super refreshes. Turing's launch was still coming off the heels of the mining run up with excess inventory in the channel. The reality was there was likely a business decision made to not write off excess Pascal inventory which meant you had to price with clearing Pascal with heavy discounting. But more 2018 the market was still in the highs of memory undersupply combined with GDDR6 being new. 2019 saw memory prices collapse across all types. GDDR6 prices were almost certainly lower. If you look at Nvidia's earning report despite the "price cuts" with the Super refresh they're margins actually increased which does suggest there was ample cost structure since 2018.
One thing that I haven't seen brought up as possibility with Turing is that Nvidia actually missed performance targets (yet still being more than competitive shows how much momentum they have, or how much issues the competition has). While the broad marketing literature and therefore attention became focused on RTX feature/hardware there was other changes in Pascal->Turing as well. Turing had extensive redesign on the shaders for example and if you search for it there is literature that Nvidia had with things such as "50% higher shader performance per CUDA core." But if you look at it Turing has nowhere near that level of "IPC" increase in practice that would be suggested by "50% higher core to core shader performance" and 40% if not more (GDDR6 14gbps vs G5X 10Gbps) memory bandwidth looking at comparisons of say the RTX 2070/2080 vs GTX 1080. Even conservative number that does show up with regards to how many INT:FP32 operations there are that still would suggest a 35% speed up which wasn't achieved. "IPC" increase seems closer to the 20% range (if that). It'd be interesting to see if one could truth serum them what they would actually say about this in terms of what they were planning for and the actual end result.
So with the above if market conditions and normal and performance targets were hit my feeling is Turing prices at launch likely would been closer to the Super refreshes while being 10%+ higher performance. This coincidentally would've roughly matched Kepler->Maxwell uplift in performance and price/$. Who knows, fun speculation though.
Whether or not Steam, or any poll/survey, is accurate or not isn't really important as it isn't really a binary distinction. What is actually important is how accurate or representative it is within a high probability, and if that is suitable enough for the argument/discussion taking place. Even if we feel say Steam has Nvidia at 70% market share isn't "accurate" does it really matter if it differs from the reality which is at 68%? Well it depends on the context of the discussion. The likelihood is also very low that it is off by >10%.
With regards to Turing pricing people should consider what the broader market situation was in 2018 when Turing launched versus 2019 with both Navi and the Super refreshes. Turing's launch was still coming off the heels of the mining run up with excess inventory in the channel. The reality was there was likely a business decision made to not write off excess Pascal inventory which meant you had to price with clearing Pascal with heavy discounting. But more 2018 the market was still in the highs of memory undersupply combined with GDDR6 being new. 2019 saw memory prices collapse across all types. GDDR6 prices were almost certainly lower. If you look at Nvidia's earning report despite the "price cuts" with the Super refresh they're margins actually increased which does suggest there was ample cost structure since 2018.
One thing that I haven't seen brought up as possibility with Turing is that Nvidia actually missed performance targets (yet still being more than competitive shows how much momentum they have, or how much issues the competition has). While the broad marketing literature and therefore attention became focused on RTX feature/hardware there was other changes in Pascal->Turing as well. Turing had extensive redesign on the shaders for example and if you search for it there is literature that Nvidia had with things such as "50% higher shader performance per CUDA core." But if you look at it Turing has nowhere near that level of "IPC" increase in practice that would be suggested by "50% higher core to core shader performance" and 40% if not more (GDDR6 14gbps vs G5X 10Gbps) memory bandwidth looking at comparisons of say the RTX 2070/2080 vs GTX 1080. Even conservative number that does show up with regards to how many INT:FP32 operations there are that still would suggest a 35% speed up which wasn't achieved. "IPC" increase seems closer to the 20% range (if that). It'd be interesting to see if one could truth serum them what they would actually say about this in terms of what they were planning for and the actual end result.
So with the above if market conditions and normal and performance targets were hit my feeling is Turing prices at launch likely would been closer to the Super refreshes while being 10%+ higher performance. This coincidentally would've roughly matched Kepler->Maxwell uplift in performance and price/$. Who knows, fun speculation though.
Gideon
2[H]4U
- Joined
- Apr 13, 2006
- Messages
- 3,556
I provided proof and you provide nothing to prove JPR is inaccurate, yet boast it is flawed. Your talking with feelings and no poof. So either provide it or give up on saying JPR is flawed. Reality is you and couple others have nothing but demand others provide data that was not hard to get and then try to dismiss it as irrelevant when proven otherwise. AMD is not 15.26% share of the video card market as per Steam.
Last edited:
Ok, if it’s not 15.26% please educate us on AMD’s actual share of installed GPUs.