Name                                      Chip                      Memory Shaders   TMUs    ROPs   Base Clock   Boost Clock Memory Clock
NVIDIA GeForce RTX 3080      GA102-200-KD-A1    10 GB    8704       272       96     1440 MHz    1710 MHz    1188 MHz
NVIDIA GeForce RTX 3080 Ti  GA102-300-A1          20 GB    10240     320       112    1410 MHz    1740 MHz    1188 MHz
NVIDIA GeForce RTX 3090      GA102-300-A1         24 GB    10496     328       112    1395 MHz    1695 MHz    1219 MHz

https://www.techpowerup.com/gpu-specs/nvidia-ga102.g930

On techpowerup ga102 page has this table on 3080ti, how much this might be true that is another thing hehe but I somehow believe this table to be spot on. March next year perhaps? I don't think AMD will have anything in this bracket so nvidia will charge whatever they want. I still think 10gb is not enough in 2 years time, it is today and probably next year too, getting a 10gb now and reselling in 2 years is not something i would like to do. The only thing that remains is what will be the price of the 20gb, I predict maximum $849 which if you think about is not that bad, charging $899 or $999 might be just too much for it.

I will make the setence better,  "efficiency, density and price".

I guess if that comes to happen, it will be like the 1080 x 1080ti, performance like or a bit lower than 3900 but cheaper and 384bit + 20gb, 3080ti around $1199, I guess.

It's funny really, dont know why you trolls are so upset about the 3080, there must be something why it triggered the negativity on some trolls, its clearly the best gpu at moment, even the chinese stocked thousands of it 2 weeks prior of the launch, countless threads on baidu about 3080 farming, I myself never thought the chinese would show so much interested on the gpu because they have everything in there, why stock on 3080 while they can buy asics or better gpu performance price efficiency? factories and factories are on their doorstep, do the chinese are stupid for such action? on the contrary, they are very smart. Besides the chinese, the same can't be said for us here in the west, because the chinese dont buy something to resell for something more, they have factories and factories close to where they live, the resell thing do not work there like it works on the west, the buying spree yesterday was because people wanted to buy to resell. The resellers have thousands of bots that dont let people to buy then they buy almost everything and then they resell for 2 to 3x more.

https://www.youtube.com/watch?time_continue=294&v=qHogHMvZscM&feature=emb_logo

yeah, amd will go for pricing. There is no question about that and I do not expect to be close to what we see rtx 3080 produce, at maximum, in my opinion, 2080 ti performance.

i have no AMD 5700, 5600, but i have much nvidia cards like p104, p106, 1070 and much more AMD cards.

arielbit you're not making much sense. These cards are for gaming, it's totally possible that the memory bus is tailored for that, for cost reasons, and that if you could have more memory bandwidth it would help for mining. If you look at the RTX 3080 vs 3090, it's the same chip, so the 3080 could be run on the same PCB as the 3090 with the same memory bus (you'd have to not fuse off parts of the memory controller of course) and you would get a gain for mining. However the card would be more expensive, and games wouldn't benefit much, so nvidia doesn't do it.

In the past it has happened that a high end GPU was remade on a smaller process, the memory bus cut, and that was the new middle end GPU.

How can I calculate the memory bandwidth thats important for ETH mining:

(memory clock in Hz × Memory Interface ÷ 8 ) × memory clock type multiplier = Bandwidth in MB/s

memory clock type multiplier:
HBM1 / HBM2: 2
GDDR3: 2
GDDR5: 4
GDDR5X: 8

i am out, i only discuss with person´s you understand technology.

best sentence from you, that show your knowledge:
anyway my point is about 150mhs "possibility" from gddr5x to 6 to 6x. that's two vram generation leaps. understand the analogy. then you will understand why it is a "possibility"

And for you, i have no AMD 5700, 5600, but i have much nvidia cards like p104, p106, 1070 and much more AMD cards.

Oh man please let it rain brain. Your knowledge is incredible XD it is product. What an answer. You know the difference between 512bit und 256bit? You know that 256bit produce halve memspeed as 512bit can do?

dude i don´t answer you anymore and set you on my ignore list. I can´t read your shit no more

And i will never win? What will i win? A trip? You have no arguments and nothing and with your 3 cards running you think you are a big player XD

XXXXDDDDD ok now i am out, I don´t discuss with persons they have really really really and really no knowledge about what they talking about. So you say, if i can halving the memory interface of an r9 390 to 256, this will change nothing in eth speed? you cut of the halve mem speed and this will do nothing with the speed? Ok dude you must learn much more, i don´t believe this, i know this.

How much cards you have for mining?

For example: R9 390 can reach over 40mh if you rise the core to the hell, an rx 580 with almost the same number of shaders, but halve mem interface cannot achieve these hashrate.
Next Example: R9 280x 384bit mem interface, can achieve 28mh on 1GB DAG, R9 380x same shader count, but interface cuted to 256bit, max it can achieve 25mh with some optimizations.

Can you prove anything or do you have any comparisons that prove your statements?

Hmm i was really close with my expect hashrate, you say it comes with over 100mh or more. 65mh to 80mh? and over 100 to 80mh. who was closer?
And whats the truth? Lol your comment, i see your brain. stupid sorry dude nothing more.

for me sweetspot is at powerlimit 65%.

I know that this are gddr5, but the memory is not the only thing. The R9 390 has 512bit memory interface, that is why i say incredible. Or you know much more cards with such a big memory interface like this?
Don´t compare only gddr5 with gddr6 or hbm, you must see the whole thing. memspeed is combination of memory interface and memspeed.

Yes, the vega7 can do much more in hashrate, because the memspeed was incredible, samething as the old AMD R9 390, it has so much mem speed, but the bottleneck was the core.

Yeah but nvidia did not need hbm2. So all in fairness still impressive as is just gddr6x.

I don't know who is dumping, but I know someone who used to work for them in the past and he was kicked out, he said they are greedy and I believe him since is not hard to figure out this if you look at the fact that one of their team -eggrasyll is charging a 10% fee on miners just because he can, and the rest is ok with it.

They are still limiting the purchase quantity?


This reminds me of how back in the early iPhone days like the iPhone 4 or 4S or 5 they would have a limit of max quantity of 2. And you would see ads all over Craigslist where people would pay you $50 to buy the 2 phones for them. Basically the guy would meet up with you outside the Apple store, he would tell you to buy 2x of a White iPhone 4 or Black iPhone 4 and you would pay either by Credit card or cash. And then he would pay you right outside the store to give it to them.

The workers there pretty much saw what was going on and there is nothing they could do. What happened after to those iPhones? They basically got sold on Craigslist for double the price or they got exported to Russia or China. Crazy how the demand was back then for iPhones.

However I am wondering why they are limiting it for GPUs since there is no "GPU gold mine rush" like there was in 2017.

as is efficiency

the big gain is density

the board below can do 5 amd cards 4 x 5700 and 1 x 5600

https://www.ebay.com/itm/biostar-TB250-BTC-D-Pro-Motherboard-Mining/124217882169?

it will do 230-240 mh

or use same board with 4x 3080 and do  320mh