News Nvidia announces RTX 50-series at up to $1,999

[TeamRed2024]

No future, seems a bit .. extreme.

Back in 2009 1080p was rare, now it's the most common resolution. 6-7 years ago 1440p was also pretty uncommon. These sorts of things change over time and 2160p is growing.

That was pretty much my point from the beginning when I said I was shocked... that such a low resolution by today's standards is the most common.

It is what it is. Nothing I care all that much about one way or another. I prefer eye candy when I game... fps means diddly squat with the gaming that I do.

As for the 5000 series... with Newegg offering $1350 in trade for my 4090 that I paid $1600 for I'll definitely bite on that deal the day I can get a 5090 from them at MSRP.

No hassles dealing with eBay etc... and paying the difference to get the 5090 makes it a no brainer.

 

[Ogotai]

I've had it ask several times, its random, which is the entire point in the first place. Steam has an estimated 120~130 million worldwide monthly active users. A 1% random sampling would be over one million records, even at a 50% response rate that is still over 500,000 data points. That is more then sufficient to get a high level overview of what the average PC gaming platform looks like.

its so random, and yet, i have never been asked.. ever, but you have been asked several times, that kind of proves its not that accurate. so... who knows how much hardware is being left out, so there is that possibility that the steam survey, isnt as accurate as you want it to be... which which teamred also kind of alludes to....

that such a low resolution by today's standards is the most common.

as seeing as you glazed over my post a few back :

teamred, you should also keep in mind... you kind of need a higher end video card to be able to use that 4k screen @4k... which most may not have.. hence the 1080p res, or maybe 2k.... when i went from 1080p to a 2k capable screen, with my 1060, my frame rates were pretty much cut in half, going to a 3060, brought my frame rates back to where where were @ 1080p. to be able to use 4k, you probably need at least a 3090, or 4070. thats probably why 4k isnt used as much as you want it to be.

 

[palladin9479]

That was pretty much my point from the beginning when I said I was shocked... that such a low resolution by today's standards is the most common.

This is known as availability bias, where people prefer the most common data points they have available and can be related to in-group bias where people prefer data from groups they are associated with.

As an example we'll use Jim, a senior account executive at a prestigious financial institution. Jim was raised in an upper class family who sent him to a nice university where he met other friends from the same socioeconomic class. After school they were recruited into the firm and have worked there for two decades now. Jim's idea of the most common car, home and disposable income would be radically different then the average. If we told Jim that the Median Personal Income for 2023 was $42,220 or that the Median Household Income was $80,610 he would say he was shocked it was that low. From Jim's POV everyone he knows has nice things, conceptually he knows there are poorer people but he doesn't have enough personal experiences to understand the scope and scale of the population as a whole.

In the group I primarily game with we have one guy who is a mid level executive at a very large global corporation, he collects expensive muscle cars. He also has a very nice gaming PC and updates it every year whenever new technology comes out. There is also another guy who is frequently unemployed due to medical problems and is using a computer I made for him out of older spare parts I had laying around. He struggles to keep his home fixed with the sheer number of old things that keep breaking. The rest of us are in between those two extremes.

While I enjoy the shiny stuff, I would prefer PC gaming remain accessible to those who represent the majority of people.

 

[palladin9479]

its so random, and yet, i have never been asked.. ever, but you have been asked several times, that kind of proves its not that accurate.

This is illogical. What you are asking for is called selection bias and is something scientific research avoids at all cost. Steam has something like 120~130 million active accounts per month with something like 60~70 million active accounts per day. There are three ways to go about gathering data, the first is the Google / Microsoft / Apple model of mandatory data theft. Valve wants to ask permission so that idea is out. The second method, what you are asking for, is a full voluntary opt-in method where the individual clicks a button in settings to send data every month, this is the very definition of a self-selecting population and introduces a lot of selection bias. Any paper that did this would be rejected by peer review for data quality issues.

The third method, which is what Valve and every social studies researcher does is to randomly select people for participation. This is where someone comes and knocks on your door, calls you on the phone or walks up to you on the street and asks you to participate in a study. Data bias can be removed by making the selection pool big enough and ensuring it's randomized.

Common Types of Data Bias (With Examples) | Pragmatic Institute

Data bias influences how we analyze and understand data. Explore 5 common types of data bias (with examples) how to avoid them.

www.pragmaticinstitute.com

An example of selection bias is a study on the health effects of alcohol on the general population. If researchers recruit participants exclusively from bars and nightclubs, they are selecting participants whose behavior may not represent the full population.

Selection bias can arise due to poor study design if the sample taken was too small, or the sample is simply not randomized.

You startup the Steam client and a pop-up appears asking you to participate in the hardware survey, you click yes and it submits data to Valve then goes away. If you want to see what gets submited click About -> System Information. This makes it very accurate for things like CPU / Memory / GPU / Primary Monitor Display / Operating System, which is what we care about.

 

[Ogotai]

if you say so... the fact that you have, as well as others im sure, have been asking multiple times, over, lets say a course of 5 years, where others have not been asked at all, shows its NOT random, period random would mean of 120-130 million users, would each at some point get asked at least ONCE over a 5 year period.

This is where someone comes and knocks on your door, calls you on the phone or walks up to you on the street and asks you to participate in a study.

cause you have been asked multiple times, shows this method is not what valve uses. if a person gets asked more then twice in a 5 year period.. means either valve is picking people some how, or there is another reason.
in your own analogy, if the same survey group keeps coming do your door, thats NOT random, no more then it is with your study on alcohol, and only going to bars or pubs'

The second method, what you are asking for, is a full voluntary opt-in method where the individual clicks a button in settings to send data every month

i didnt say that, or ask that, at all.. IF the survey is volentary and random, then at some point, every one should be asked at least once in a 5 year period. which is not the case... what you are suggesting would mean if a person could do the survey when they want to, any time they want to, and as much as they wanted to, that that would mess up the results more so then the person being selected 4 times over 5 years.

either way.,.. the steam survey should NOT be used as the be all tell all of what hardware is out there and being used, unless , each and every user, is asked at least once in a 5 year period, as there is alot of hardware out there that isnt been polled and added to this survey.

now, IF valve changed it so that each person can go in, and take the survey manually once a year, and only once a year, that that would be alot more accurate, then how it is done now...

at this point.. im just got to say this :

agree to disagree

 

[emitfudd]

$1999 for the Nvidia card. I'm guessing the MSI, Gigabyte, Asus, etc cards are going to be more like $2499. Fortunately for me I can skip the 5000 series. I have a 4090 and that is good enough for me.

 

[spongiemaster]

if you say so... the fact that you have, as well as others im sure, have been asking multiple times, over, lets say a course of 5 years, where others have not been asked at all, shows its NOT random, period random would mean of 120-130 million users, would each at some point get asked at least ONCE over a 5 year period.

That's not how random works. If you roll a 6 sided die 7 times, though not statistically impossible, you're not going to get 6 different numbers in those first 6 rolls before getting a duplicate the 7th roll. If you keep going you're most likely going to get some number multiple times before the last number makes a first appearance.

 

[Loadedaxe]

I can not believe this Steam Survey is getting beat like a dead horse again.

Its Random...Random...is it 100% accurate? Yes for the people that completed the survey it is.
Random mean you may never get asked, and then again you may get asked every time they do them, which is every quarter I believe.

Is it perfect...no, but it is the best we have to go by. The rock stopped skipping across the water.
Unless someone can come up with a better way. Its all we have to go by and its not perfect!

 

[palladin9479]

if you say so... the fact that you have, as well as others im sure, have been asking multiple times, over, lets say a course of 5 years, where others have not been asked at all, shows its NOT random, period random would mean of 120-130 million users, would each at some point get asked at least ONCE over a 5 year period.

This is only because you do not understand how statistical sampling works. Your entire argument, everything you've said, is because you just do not understand. Understanding population statistics requires people to get out of their feelings and accept that none of you are unique.

First and foremost, you are not special. It doesn't matter if Valve asked you, or anyone you know, because you are not special. There are tens of thousands of people out there just like you, and you do not matter provided the sample set is large enough. Every argument made so far is "buh buh wah about me!", which is entirely irrelevant. It's like saying the BLS MHI data is bad because you didn't personally ask you, it doesn't matter because you don't matter.

When doing population sampling, which is what this is, we need a large amount of randomized data points. The larger the number of samples the more diluted any individual data becomes and instead the law of large numbers kicks in and we get to see overall trends.

https://en.wikipedia.org/wiki/Law_of_large_numbers

Lets take a coin and saw it has a 50% chance of heads, 50% chance of tails. Now it's entirely possible to flip a coin ten times and get six heads four tails which is 60/40. Does that mean our 50/50 prediction is wrong? No we didn't have a big enough sample set, keep flipping that coin and after a thousand flips your going to see a strong 50/50 split. As long as the dataset is large and random then individual data points no longer matter, only the aggregate.

 

[Moonstick2]

Barely half of my games I purchased through Steam...

By having even one Steam game, you're adding weight to my claim that Steam is indicative of people who game on PCs. How many people out there don't use Steam at all for their PC gaming?

... and although I have somewhere like 800 hours played across a dozen or so Steam titles I own and I've never once been asked to answer a survey by Steam regarding my hardware.

This is a bit like the birthday problem. Lots of people don't understand statistics enough to realise just how few samples you need for an accurate assessment of a population:

A good maximum sample size is usually around 10% of the population, as long as this does not exceed 1000... In a population of 200,000, 10% would be 20,000. This exceeds 1000, so in this case the maximum would be 1000.

How to choose a sample size (for the statistically challenged) - tools4dev

One of the most common questions I get asked by people doing surveys in international development is “how big should …

tools4dev.org

see also The RCSI Sample Size handbook

Yep. Out of all their millions of users, they only need to get answers from 1000 for the results to be pretty accurate, so you and a couple of other people never being asked means literally nothing. The biggest problems for sample surveys are usually getting correct answers (not an issue in this case since it reads the PC config) and avoiding bias in the sample.

Spoiler: For those who still can't quite believe huge populations don't need huge samples

Imagine starting with 1 million people and you decide that you need to survey 10% of them to get accurate results. You do that and you get your table and you're happy that it gives you a good result. Now another 1 million get added to the population. If their distribution is basically the same, why would you need to increase your sample size if you were to repeat the survey? If ~ 20% of the first 1 million fall into Box A and the same is true of the second 1 million, you're only going to get about the same answer. So now you've got a good result from a sample that's 5% of the total, not 10%. Continue adding millions. Turns out what matters is not the % of the population (when the population is large) but the absolute number, which turns out to maximise about 1000 for decent accuracy. About 100 is generally the minimum.

Note that this is about getting decent accuracy. 1000 of a large population gives a margin of error of about 3%. So if your survey says 5% of people fall into Box B, you're actually saying it's 4.85% - 5.15%. If your population is 100,000 then you might be out by about 15,000. If it's 10,000,000 you're now out by 1.5 million. So yes, that comparatively tiny sample of your huge population has given you a result that could be out by over a million, which sounds awful and is why so many people believe that samples need to be so big. But it's in absolute terms. In proportional terms it's within a fraction of a % of 5%, and generally that's good enough. You rarely really need to know if the answer is 5.0% or 5.1%.

So 4K gaming has no future because only 4% of Steam gamers surveyed play in 4K?

Where have I said it has no future? The point is that it's unsurprising that it's only used by a small % for PC gaming, and that it's highly unlikely that it ever will be the norm. You said it yourself, 4K is hardly new now. It's been out years, the majority of TVs are 4K now, and yet it still hasn't really made serious inroads into PC gaming. It's the law of diminishing returns, the same way that despite the x2.5 speeds nobody's rushing to upgrade their drives from NVMe 4.0 to 5.0 like they did HDD to SSD. The majority of people look at 4K gaming, take into account the choice between increased cost (in both panel plus the hardware needed to drive it) or reduced quality, note the amount of desk real estate they need to give up for it to make a difference, and decide that a tier or two down is plenty good enough. If 4K was such a must-have, it'd be a far bigger proportion by now. It looks like the only way it'll become the standard resolution is if manufacturers stop producing the lower panels.

I feel inclined to say that 1440p will eventually become the majority resolution, yet given that even now it's still only around 1/2 the number of 1080p makes me think I'm falling into the same trap of it's a better number, so surely...? But elements like cost, perf/$, practicality and perceived benefit all come into play. Same way that 5.1 surround speakers are better than headphones, all else being equal. But all else isn't equal, so headphones are more popular for gaming and probably always will be.

 

[palladin9479]

Yep. Out of all their millions of users, they only need to get answers from 1000 for the results to be pretty accurate, so you and a couple of other people never being asked means literally nothing.

The only caveat with this is that you really want more data points just to protect against sampling bias since populations are not truly homogenous (China's market is different then the USA). With smaller data sets you run the risk of sampling selection error over representing one demographic or another. You either use the LLN to smooth that over or just rerun the analysis with different random datapoints a few more times.

 

[TeamRed2024]

The only caveat with this is that you really want more data points just to protect against sampling bias since populations are not truly homogenous (China's market is different then the USA).

Thanks for the summary... I didn't care to take the time to read his (obviously) biased post. 🤣

At the end of the day I could care less. 4K gaming is where it's at... and anyone can feel free to disagree. Newegg offered $1350 in trade for my 4090... so upgrading to the 5090 for a $700 difference is an easy decision.

 

[valthuer]

Thanks for the summary... I didn't care to take the time to read his (obviously) biased post. 🤣

At the end of the day I could care less. 4K gaming is where it's at... and anyone can feel free to disagree. Newegg offered $1350 in trade for my 4090... so upgrading to the 5090 for a $700 difference is an easy decision.

Just out of sheer curiosity: which version of 5090 will you choose? If I were you, I’d go for Founders.

 

[spongiemaster]

Just out of sheer curiosity: which version of 5090 will you choose? If I were you, I’d go for Founders.

Newegg doesn't carry the Founders Edition. Only Best Buy, and maybe MicroCenter will have it.

 

[valthuer]

Newegg doesn't carry the Founders Edition. Only Best Buy, and maybe MicroCenter will have it.

Sorry. Had no idea. I live in Greece, so I don’t know much about the US market.

 

[TeamRed2024]

Just out of sheer curiosity: which version of 5090 will you choose? If I were you, I’d go for Founders.

Newegg doesn't carry the Founders Edition. Only Best Buy, and maybe MicroCenter will have it.

No Microcenters within 1000 miles of me... and I am definitely doing the Newegg trade so will get whatever I can get my hands on. I bought my Gigabyte Aero 4090 from Newegg and may go that route again... only thing is this time I will get one to match my build rather than whatever is available. All my components are black except the 4090 which is white.

Gigabyte, Asus... no preference honestly. I've had EVGA cards in the past but I know that's not an option.

 

[Moonstick2]

The only caveat with this is that you really want more data points just to protect against sampling bias since populations are not truly homogenous (China's market is different then the USA).

Increasing the number of samples doesn't protect against bias. It's even in one of those links: "An unrepresentative sample will result in biased conclusions, and the bias cannot be eliminated by taking a larger sample." taking the classic example of surveying people at home Mon-Fri 0900-17:00 biases against office workers, simply phoning more people at that time isn't going to reduce the bias.

All a survey on a population will tell you is something about the population. No population is homogenous otherwise you'd only ever need a sample of 1. But imagine a 1000 x 1000 grid of squares, each square one of nine shades and those squares evenly dispersed throughout the grid. Choosing 1000 truly random samples gives you a survey of the distribution of shades through the population.

Now you assign one of seven colours to each shade. Random-ish but with some bias - you make darker shades more likely to be blue, for example. Your shade-survey results don't change.

If you then rearrange those squares so that all the colours are grouped together in seven regions it's a mistake to think that the original sampling method won't work, or that a larger number of samples is needed to ensure that a sufficient number of samples is guaranteed from each colour region (or even worse, to artificially sample a set number from each region according to its size). If the sample selection was truly random to begin with (not biased towards one edge of the grid, for example) then it's not going to be affected by the rearrangement.

Now take Steam users: 14 m in the US, 11 m in China, 9 m in Russia, etc. Take all their User ID numbers, put them in a list, shuffle them until they're completely randomly distributed. You can take your 1000 samples and get a reliable survey. It's the reverse of the above process. You don't have to worry how many you've sampled from each country or whether certain nationalities are more or less likely to have a 4090.

Of course, the Chinese and US markets are different, but the survey just says 4% of Steam users, not that 4% applies individually to each and every country. The original remark that started this off was astonishment at the statement that "2/3rds of gamers game at 1080p or lower" (which is a bit melodramatic - only 7% game below 1080p, it only gets close to 2/3rds because 56% game at 1080p). I just looked at it and thought "well yeah, a lot of countries and people in the world with less money who still play games."

Dragging it back to the original question, "what's a mainstream [PC] gamer today?" I'm more inclined to say it's somebody who falls around the median of the Steam survey results than what I see online when people talk about their gaming PCs. Otherwise I'd also think the majority of gaming PCs are water cooled.

 

[palladin9479]

Increasing the number of samples doesn't protect against bias

It protects against selection bias because the random selection process can sometimes result in over representation of one segment or another. The probability of that happening goes down as the total number of selections goes up, this is why 1000 is generally "good enough" instead of just 10. What you are essentially doing is reducing the error bar on each metric.

 

[Moonstick2]

It protects against selection bias because the random selection process can sometimes result in over representation of one segment or another.

That's honestly not the case, for reasons already explained.

If you're doing a survey on average wealth, and your population divides up into two groups of 20% rich, 80% poor, a truly random selection will already select samples from each group in approximately the correct proportions. It doesn't need to be exact for sufficient accuracy. If your selection process means one of the groups is over-represented, increasing the number of samples can't fix that. It's like saying "If I throw 1000 dice, sometimes I might not get 1/6 of each number, so I better throw more."

Similar to the Monty Hall problem, people's intuition often leads them astray. People think larger populations need larger samples for sufficient accuracy, but it's been known for a long while that this isn't the case and that 1000 is basically the limit for populations of 10,000 or more. Going higher gives no meaningful improvement.

It's also covered in far more detail in the links I provided, which give the same conclusion.

 

[palladin9479]

That's honestly not the case, for reasons already explained.

Yes it does, it's known as the Law of Large Numbers. Otherwise we would only need 10 samples for anything, or even just 1.

The larger the sample size the smaller the difference between measured mean and the true mean, the only time both measured and true is identical is when sample size equals population. The mistake you are making is assuming that random samplings are always perfectly representative, which is untrue.

Coin flip is a perfect example. Flip a coin 10,000 times then randomly select 10 of those results with heads equaling 1 and tails equaling 2. There is a very very low probability that your measured value is going to be 1.5, randomly select 100 of those results and you get closer to 1.5. You have a higher chance of getting a 6/4 split with 10 coin flips then with 100, 1,000 or 10,000 because you dilute the weight of outlying samples.

Anyhow this is way outside the scope of the topic so lets just agree to disagree here.

 

[Moonstick2]

Otherwise we would only need 10 samples for anything, or even just 1....Coin flip is a perfect example. Flip a coin 10,000 times then randomly select 10 of those results ...

Nobody would take a sample of 10 from a population of 10,000 and believe it to be representative. 100 is generally accepted as the minimum meaningful sample for reliable results regardless of population size. Beyond that, to quote it yet again:

"A good maximum sample size is usually around 10% of the population, as long as this does not exceed 1000"

If you wanted to know the proportion of left-handed students in a class of 30 you'd ask all 30. In a year of ten such classes, you'd ask 100. In a school of 5 years, you'd ask 150. And if in a country of 200 schools, you're fine asking 1000, if you correctly randomly sample.

The mistake you are making is assuming that random samplings are always perfectly representative, which is untrue.

“If you don’t believe in random sampling, the next time you have a blood test, tell the doctor to take it all.”

The point is that 1000 from populations of 10,000+ is widely known to have a reasonably small error in % terms, which is all representative means. It can be considered perfect as in 'perfectly fine'. The only literally "perfectly representative" sample would be 100% of any population. The mistake you're making is thinking that in a population of 1 m, sampling 10,000 gives a more meaningful result than 1,000. It'll increase precision, sure, but by fractions of a %. It's like arguing for people's weights to be quoted to the milligram. And contrary to your assertion, it does nothing about bias.

You have a higher chance of getting a 6/4 split with 10 coin flips then with 100, 1,000 or 10,000 because you dilute the weight of outlying samples.

In a population of coin-flips, no single coin-flip can be labelled an outlier. In a more relevant example. like the Steam/GPU stuff, then the handful of people running triple 4090s are outliers. Your viewpoint appears to be that because a 1000 sample scheme is virtually never going to sample any of them, the result is unrepresentative because they're not represented; so this has to be resolved by increasing the number of samples until they get sampled. This might sound right, but it's wrong. If there are say 1000 of them out of 132 m users, they make up less than 0.0008% of the population. And generally speaking, surveys don't claim to be accurate to 4 decimal places. because nobody cares if A is 32.4% or 32.4107%, so 0% having 3x4090 is an entirely acceptable statement and the fact there are actually a few out there is accounted for in the error. 0% means "None, or basically as good as none."

Anyhow this is way outside the scope of the topic so lets just agree to disagree here.

Fair enough, but but just to be absolutely clear here, this isn't you and me disagreeing. This is you disagreeing with pretty much every statistician out there.

 

[HardwiredWireless]

Ok, that's a personal preference.
The 5090s still aren't below 2K, except for the five-finger discount; everyone else is going to pay at least 2K due to taxes/vat.

The tax money is not collected by Nvidia and kept its passed on. If you're worried about the price maybe you need to talk to the people who are taxing you to death.

 

[valthuer]

The tax money is not collected by Nvidia and kept its passed on. If you're worried about the price maybe you need to talk to the people who are taxing you to death.

I don't believe he wrote that with the intention of blaming Nvidia: he just mentioned it as an undisputed fact.

At the end of the day, whoever's responsible for these prices - Nvidia and/or the government - it doesn't really matter.

All that matters is, whoever's willing to buy 5090, will have to pay more than 2K.

Unfortunately, reality is as simple as that.

 

[HardwiredWireless]

I don't believe he wrote that with the intention of blaming Nvidia: he just mentioned it as an undisputed fact.

At the end of the day, whoever's responsible for these prices - Nvidia and/or the government - it doesn't really matter.

All that matters is, whoever's willing to buy 5090, will have to pay more than 2K.

Unfortunately, reality is as simple as that.

And that's not a problem. It's a very reasonable price in my opinion for what you're getting and the money you can make over the year with it.