Nvidia's 3000 Series RTX GPU [3090s with different memory capacity]

Here is Joel Hruska's contact so you can complain to him instead:
https://www.extremetech.com/author/jhruska
I participate in these forums to have interesting discourses with people like yourself (i.e., fellow forum members). I don't have any desire to correct the author of every clickbaity article written on the web. Have I convinced *you* that that article's claim was flawed?

Taking a few steps back -- to me the SHS seems to show that a good number of 3000-series cards have reached gamers, and in the case of the 3070 it's roughly in line with what we've seen in past generations. We have no idea what fraction of the total cards manufactured this represents, and as a corollary we have no idea how many cards have reached miners. It could be 10% of all cards manufactured, or 50%, or 90%. We have no data.

Obviously there is still a huge pent-up demand for all GPUs. It's hard to disentangle supply from demand. There may be supply, but obviously there isn't *enough* supply to meet demand.
 
If you start with a pre-conceived narrative you can always cherry-pick specific data points that seemingly support that narrative.
Here are some counter-examples to that narrative from the same data source (SHS):
  1. The GTX 1070 (an extremely popular card, $379-$449) launched on 05/27/2016. Five months later, the SHS for 10/2016 shows its adoption at 1.30%: http://web.archive.org/web/20161106071852/https://store.steampowered.com/hwsurvey/videocard/
  2. The RTX 2070 Super (a popular Turing/RTX card, $499) launched on 07/09/2019. Five months later, the SHS for 12/2019 shows its adoption at 0.63%. http://web.archive.org/web/20200103034839/https://store.steampowered.com/hwsurvey/videocard/
The 3070 ($499) reached 1.29% (~identical to the 1070 and nearly 2x the 2070S) of a presumably larger install base than either of those cards within a similar time frame (launch 10/29/2020, SHS 03/2021).

Yeah, Steam numbers nearly doubled since the release of Pascal in 2016. Alone the current users went up from 14 millions to 25 millions: https://www.statista.com/statistics/308330/number-stream-users/
 
Linus is claiming gpu shortage is not due to miners based on discussions he has had. Miners don't help, but they are not reason for shortage. Someone could do the math on eth hashing rate increases to figure out how many gpu's are being added to mining capacity. Just take the increase in hash rate and divide with average hashing rate of modern gpu. Ignore asics etc. to make a worst case estimation.

 
Thanks @manux. Linus's quote from the video: "it's just that way more people want GPUs than usual". Demand > Supply, and it's just tough to ramp up supply quickly to match unforeseen demand. So yeah it's not a very sexy story, just a little sad and frustrating for a lot of people. Miners and scalpers just add a cherry on top.
 
Hopefully someone who has time does the math. I can't be bothered to do that.
It's pretty clear without any math that miners do contribute to shortage. And if you think about it for a minute you'd notice that the GPU prices went through the roof at the same time as BTC and ETH went there - which is as clear of an indicator of whose demand is driving the market right now as there can be.
 
The chart on eth hashrates is interesting. I would say a lot of the gain has been from old GPUs being redirected to ethereum. So it starts at 300 TH/s and goes to 500 TH/s. That is a gain of 200,000,000 MH/s. A single 3080 gets an easy 80 MH/s. So it would be almost 2.5 million 3080s.

I think at least half the gain is probably from 3000 series and the rest from asics and rdna GPUs. So I wouldn't be surprised if 1-2 million ampere GPUs went to miners.

What does 1.3% on SHS represent as absolute numbers? One percent of 25 million is 250,000. That would mean the bulk of 3000 series are going to mining. Not sure how likely that is.

Also I think demand from deep learning is being discounted. The company I work for buys low thousands of consumer GPUs and we wouldn't hesitate to pay $1500 for a 2080 blower. I don't think we are that unique. Though we have moved to Quadros and A6000s just to solve supply issues.

Good time to sell though. Cleared $800 for an old gaming 1080 Ti. That is more than I paid for it nearly four years ago!
 
The chart on eth hashrates is interesting. I would say a lot of the gain has been from old GPUs being redirected to ethereum. So it starts at 300 TH/s and goes to 500 TH/s. That is a gain of 200,000,000 MH/s. A single 3080 gets an easy 80 MH/s. So it would be almost 2.5 million 3080s.

I think at least half the gain is probably from 3000 series and the rest from asics and rdna GPUs. So I wouldn't be surprised if 1-2 million ampere GPUs went to miners.

What does 1.3% on SHS represent as absolute numbers? One percent of 25 million is 250,000. That would mean the bulk of 3000 series are going to mining. Not sure how likely that is.

Also I think demand from deep learning is being discounted. The company I work for buys low thousands of consumer GPUs and we wouldn't hesitate to pay $1500 for a 2080 blower. I don't think we are that unique. Though we have moved to Quadros and A6000s just to solve supply issues.

Good time to sell though. Cleared $800 for an old gaming 1080 Ti. That is more than I paid for it nearly four years ago!

I wonder how many ampere gpu's nvidia sells quarterly? Nvidia did record Q4 where gaming revenue was up 67% compared to previous year's Q4. Q1 was predicted to be slightly better than Q4. Q3 was also huge.
 
It's pretty clear without any math that miners do contribute to shortage. And if you think about it for a minute you'd notice that the GPU prices went through the roof at the same time as BTC and ETH went there - which is as clear of an indicator of whose demand is driving the market right now as there can be.

There is difference between contribute and cause. What linus was claiming is that majority of gpu's are still going for gamers. I'm not claiming this or that. I don't have the sources to make such determination.
 
If you start with a pre-conceived narrative you can always cherry-pick specific data points that seemingly support that narrative.
Here are some counter-examples to that narrative from the same data source (SHS):
  1. The GTX 1070 (an extremely popular card, $379-$449) launched on 05/27/2016. Five months later, the SHS for 10/2016 shows its adoption at 1.30%: http://web.archive.org/web/20161106071852/https://store.steampowered.com/hwsurvey/videocard/
  2. The RTX 2070 Super (a popular Turing/RTX card, $499) launched on 07/09/2019. Five months later, the SHS for 12/2019 shows its adoption at 0.63%. http://web.archive.org/web/20200103034839/https://store.steampowered.com/hwsurvey/videocard/
The 3070 ($499) reached 1.29% (~identical to the 1070 and nearly 2x the 2070S) of a presumably larger install base than either of those cards within a similar time frame (launch 10/29/2020, SHS 03/2021).

Both of those are also unremarkable. Lets compare with something that actually was a huge seller, the GTX 970. In a similar time span the GTX 970 a 2.46% share of the Steam gaming market. The numbers for the 1070 likely would be higher if it's initial timeframe also included the Holiday shopping months like the 2070 S (impacted by the mining boom in the form of a flood of Pascal cards onto the used market) and the 3070.

Steam Hardware & Software Survey (archive.org)

In terms of absolute volume an argument can certainly be made that the volume of cards might be larger or similar. But then if the GTX 970 was selling to the same number of PC gamers as the 3070, you can also make an argument that in absolute sales volume it'd also be significantly higher than then numbers that were sold then. But in terms of the percentage of Steam gamers as a whole adopting new graphics technology, the 3070 is completely unremarkable. Hell, even the Radeon 4870/4850 did significantly better. And those were cards that should be compared against the adoption of the 3090/3080 as that's the market segment they targetted which makes it all the more remarkable.

Now, you can also make the argument that those cards didn't have a mining boom siphoning off significant sales to coin miners ... but that's the point. Mining is greatly impacting the adoption of new cards into the marketplace. In that sense, considering the impact of coin mining, the otherwise unremarkable adoption numbers for the 3070 on Steam become relatively decent purely because it's just some fraction of what it could be if it weren't for the current mining boom. As a needle that moves gaming forwards into a newer generation of GPUs, however? It's pretty average.

Regards,
SB
 
If there's a lot more Steam users than 5 years ago, doesn't that mean a 2% gain then might be equal to a 1.5% gain now for actual GPU counts?
 
If there's a lot more Steam users than 5 years ago, doesn't that mean a 2% gain then might be equal to a 1.5% gain now for actual GPU counts?

Sure, but if the GPU from 5 years ago was sold into the same number of users as exists today would it suddenly sell to less than 2% of users? Probably not, the relative magnitude of sales would likely still be very similar unless the overall demographics of the market has changed. IE - if say enthusiasts make for a far larger percentage of ALL gamers than they did 5 years ago.

That's an assumption we can't make without some form of evidence, however. So it's better to assume that the overall gaming demographic of users on Steam is relatively stable and only the magnitude changes. So 2% adoption from 1 million users would still be 2% adoption with 2 million users.

Regards,
SB
 
Sure, but if the GPU from 5 years ago was sold into the same number of users as exists today would it suddenly sell to less than 2% of users? Probably not, the relative magnitude of sales would likely still be very similar unless the overall demographics of the market has changed. IE - if say enthusiasts make for a far larger percentage of ALL gamers than they did 5 years ago.

That's an assumption we can't make without some form of evidence, however. So it's better to assume that the overall gaming demographic of users on Steam is relatively stable and only the magnitude changes. So 2% adoption from 1 million users would still be 2% adoption with 2 million users.

Regards,
SB

Didn't steam get a ton of new users from china at some point. Or am I remembering wrong?
 
Back
Top