Nvidia's 3000 Series RTX GPU [3050, 3060, 3070, 3080, 3090 now with TIs]

TechPowerUp poll exists to serve relatively enthusiast gamer crowd, it doesn't represent the overall market in any imaginable way.

 

An uncontrolled public poll has the logical biases of both self-reporting (only the people strongly opinionted enough to seek out the poll will report) and the anchoring effect (who is to say all respondents are actual participants versus just trolls on the internet with an axe to grind.)

It also has the side logical bias of everyone reviewing the poll falling for false consensus, where you assume everyone agrees with your view on the subject whether true or not.

 

The problem with your reasoning is that Nvidia doesn't sell anywhere near 10s of millions of RTX gpus per quarter. At the beginning of the year, Nvidia claimed to that 10% of all gaming PCs were RTX based and many extrapolated that the total RTX userbase to be around 20 million after 2+ years on the market.

High-end-based desktop gpus are going to be just a fraction of Nvidia's total GPU shipments on any given year. 9.7 million high-end gpus being swept up by miners since the release 0f 3000 series cards would actually crater the gpu gaming market for gamers.

Maybe, the miners by themselves aren't solely responsible for creating the current reality. In past the Nvidia may have been able to up production to deal with demand. But the current shortages experienced by the semi-conductor market, in general, may have allowed a portion of buyers who have deep pockets and buy in bulk to have an outweighed influence on the market particularly in regards to pricing.

 

But what is high end these days though? Anything 3060Ti/3070 (RX6700xt and up) isnt really high end perhaps, but capable enough and then some for what games are being targetted for during this generation. Even better said, any RTX or RX gpu is capable enough for whats to come for a long time.

 

We can only use the numbers published, and the numbers published say NVIDIA shipped ~18MM GPU's in Q1 of this year and ~17.4MM GPUs in Q4 of last year, along with a reasonable estimation of about ~15MM more up to this point in this quarter. If you read my entire post, you will see I called out exactly the point you're belaboring; we have no specific data on the SKU breakout of those GPUs.

My math still stands: ten million 3060's worth of compute were added to ETH, which is a worst-case scenario for most cards having been sold to miners ostensibly buying them. That's less than 20% of the total GPU's sold by NVIDIA only, assumes zero tuning at all, and assumes literally zero AMD cards were being used. These are already horribly unfair (read, non-realistic) assertions in favor of your argument and not mine, becauase we know A: tuning picks up easily aother 20% hashrate and B: AMD cards are absolutely playing in this space too. Did you not consider how gracious I was beng?

Maybe in your next post, instead of doing literally no work at all except spouting uninformed opinion, you can do some hard math for yourself to see how many GPU's AMD has sold and how financially their cards compare on hashrate at each SKU level. Pick the slowest one, and figure out what percentage of cards would be needed to make up that ETH global hash rise. The consider how much basic RAM underclocking / GPU overclocking helps with power consumption and therefore hash rate on a per-card bsais, and how that hashrate increase in tuning would effect the total number of cards needing to be sold. Maybe you could consider how higher-placed SKU items like the xx70 and xx80 and Ti series cards affects hashrates-per-card-sold too, as a function of how many GPUs have shipped That 20% figure I came up with is completely off-the-ceiling too high as it is, probably by double, if we account for everything else in the SKU stack plus easy and well documented tuning plus AMD's sizable stake in the game.

We also assume nobody was buying cards on the used market for mining. We also assume nobody started mining with cards they already had but just hadn't realized it was profitable; lots of people gave up on GPU mining Bitcoin a long time ago, ETH is a relatively new player. i took every avenue to inflate the numbers in favor of the counter argument (only NVIDIA chips, only the slowest ones, only brand new cards, using absolute peak numbers of hashrate) and the numbers are still not there to support this absurd claim that miners are ruining the GPU economy.

Miners are not ruining the GPU economy, stop the bullshit.

 

I agree with your arguments. If NV could deliver whats on demand, they would just sell a whole lot more. The GPU/PC market is just very much growing, heck it shows the demand when people are willing (crazy enough) to pay thousands of dollars for a new gpu), which even i as a hardcore pc gamer wont. I build my current system (2080Ti, 3900x, 32gb ram, samsung 970 nvme) in 2018, upgraded the ssd in early 2019 somewhere and a cpu/mobo upgrade. More capable then the consoles and i can upgrade the SSD to 7gb/s standards when prices sink somewhat. Probably the best build ive ever done, along with the Ti4200 system almost 20 years ago

 

Just because the access to the numbers is limited, doesn't mean the numbers that are available are good to use or can provide any insight to the situation. You can't comfortably break out numbers for RTX sales than whats the point? Whats worse is you don't even try. I would rather you make a bad assumption with good intention than taking overall Nvidia quarterly gpu sales as its just an invitation to a poor conclusion. Personally, I wouldn't bother generating an opinion based off such limited data. What little research I've done shows that your use of 18 million gpus a quarter is excessive when talking about supply and demand within the RTX market and 9.7 million of 3060 level GPUs going to miners would be rather significant.

You go ahead and perform your thesis-level work about crypto mining effects on the gaming gpu market using a bunch of mindless calculations or reasoning while Nvidia simply states stuff like this...

https://www.tomshardware.com/news/n...u-sales-set-records-as-mining-craze-continues

Now I have no clue whether gpu miners would have an outweighed influence on gpu prices outside the current conditions. But its rather obvious that Nvidia feels the need to make their gpus less attractive to miners and promised to take additional actions to lessen their effect on supplies.

Thats what your argument looks like to me in laymen terms.

 

That is rather the point? The issue was if gamers are driving up prices. Apparently 91% are not. Or is the contention that casual gamers are willing to pay *much* more than enthusiasts?

As opposed to selectively trusting only evidence which supports one's claims. What is that bias again?

 

20 million sounds like a very low number. VERY LOW. There close is to a billion PC out in the world.

Then what are they selling? Pascal was completely phased out 2 years ago, NVIDIA is only manufacturing Turing since at least the beginning of 2020, then came Ampere later.

 

How representative is that poll? I, for one, never heard of that poll until you mentioned it.

 

Last I looked it had over 28k responses, so pretty good sample size.

 

Are they? Only GeForces available I saw in recent quarters were 1050s.

 

The problem is not sample size, but quality of the sample.
Generally I won't put too much trust into internet polls, because for most internet polls people have to actively take the survey, instead of passively being surveyed, which is already a problem. They also tend to break some basic rules (e.g. generally you don't want to have partial results available before the conclusion of the poll), and they are generally not very good at keeping people from voting repeatedly.

On the other hand, if you look at, say, Steam hardware survey, it's actually more sound, statistically. The survey is done actively (people are passively surveyed, though they can refuse to participate, just like in normal polls). It's quite hard to vote multiple times from the same person, as the survey is based on Steam account, and while it's possible to create multiple Steam account, it's hard to imagine why someone would want to do that just to pollute the survey results.

 

Yeah, I took prob & stats when doing my math degree....

 

Perhaps this will clarify the situation. 3070s are going for about 1650 on Amazon. Why? Because it is profitable to buy them at that price. Each 3070 can make 3000+ a year mining. Let that sink in. Hell, I have some land, 15kw of solar, quite a bit of free time, and a chunk of disposable cash. I'm tempted to start buying these bargains up. Just need to get over the moral hangup of it being an ecological disaster.

 

Is there a source for this? I'm genuinely interested, especially Pascal-based 1050 Ti and below still seem to be sold at almost competitive prices (i.e. not individual samples, that some mom-and-dad-shop had lying around the shelf for ages). Heck, there even are a shit ton of GT730s and their like.

There's also a whole lot of Turing-based GTX cards - not in models, but in numbers: 1660, 1660S and 1660Ti, which can be of interest to miners. They are build on 12nm, so there's less contention about 7- or 8-nm-production ressources, and because of their lower price points, traditionally, those GPUs made up the majority of numbers. Steam Survey ranks the cards like this: 1060, 1050 Ti, 1650, RTX 2060, 1050, 1070, 1660 Ti, 1660S, RTX 2070S, Radeon RX 580, GTX 1080, GTX 1660, RTX 2070... of course, that's installed base, not volume sold in the last two quarters.

There's also a whole lot of notebook models based on MX150-450, which are much cheaper than RTX-based notebooks. I take the liberty to assume, the are higher volume than the latter as well. Of course, this is not counting toward Ampere GPU numbers, but GPUs sold in total.

 

-nevermind-

 

People will use this...but reject Steam numbers?

 

The techpowerup poll isn’t useful because it likely includes a significant number of bystanders who aren’t actually in the market for a new video card. You would need to poll people who have either bought or actually tried to buy a card. In general asking people about their “willingness” to do anything is kinda pointless.

That poll certainly isn’t a rebuttal to the hashrate analysis posted earlier. That analysis at least attempts to use hard data.

 

JPR, who has been quoted for numbers before, seems to have taken an interest in crypto as well.
https://www.jonpeddie.com/blog/crypto-minings-half-a-billion-dollar-impact-on-aib-sales
Coming from a different model, based on GPU attach rates, they wrote:
"The model predicts that about 25% of the AIBs shipped in Q1’21 went to miners and speculators. That’s approximately 700,000 high-end and midrange AIBS in Q1’21. And the market value is about $500 million—a half a billion dollars."