how fast can Samsung fulfil Nvidia's second contract orders?

Considering the RTX 3090 launched on the 24th September, and it was sheer luck I was able to order one on the 3rd December (availability still poor)? I would hazard a guess at not fast enough.

Also, I'd assume the new production deal also covers the lower tier chips as well now? So that'll dilute the production of any given chip volumes. Hopefully by this time next year they may actually have some stock to sell.

Seeing 3060Ti, 3070 & 3090 pop up daily right now, but 3080 are still like rocking horse manure.

that's good considering how TSMC dominates the market.
Anyway no matter how much samsung will produce, nvidia is selling to miners first…

iworrall
Seeing 3060Ti, 3070 & 3090 pop up daily right now, but 3080 are still like rocking horse manure.

Where? I cant get my hands on one.

ccl - some of them are pricey though.

DevDrake
that's good considering how TSMC dominates the market.
Anyway no matter how much samsung will produce, nvidia is selling to miners first…

They can sell to who they like though…

Has anyone heard anything recently about Samsung's 3nm GAA node which was meant to due in 2021?
I'm obviously not expecting to see this happening but I was wondering how bad the delay is going to be.
The industry desperately needs Samsung to bring some competition to TSMC's 5nm and with a really decent capacity too, I know Samsung has enough EUV machines ready to do this but they just need to perfect the node first.
Intel could then use TSMC where it wants, Samsung where it wants and still us it's own 10nm SF for some products and 14nm for it's bulk production of low performance (cheaper) products.

If Intel had swallowed it's pride in 2015 after seeing it was having problems with the 10nm node they could have scaled back their ambitions in an attempt to deliver something, then said if it still looks like the issues won't be solved by 2017 they should have started discussion with TSMC. Intel easily has enough money to have two design teams running side by side, one working on their own ill fated 10nm node and another using TSMC. That would have meant in 2019 if the in house FABs hadn't sorted themselves out Intel was ready to go at TSMC. They could have released high end products on TSMC nodes and cheaper volume products on their own FABs.

One other avenue I think should have been explored was Intel licencing TSMCs IP for the 7nm node maybe a deal all the way through to 3nm when everything is changing to GAAFET. TSMC doesn't have anywhere near the capacity to serve their long term customers and also help Intel out in a big way and Intel has the capacity just not the IP. So put these two together and TSMC could make an absolute fortune by licencing their node knowhow to Intel and Intel wouldn't be where it has been for the last 3 years and where it will be for the next 5!

Oberoth
Has anyone heard anything recently about Samsung's 3nm GAA node which was meant to due in 2021?
I'm obviously not expecting to see this happening but I was wondering how bad the delay is going to be.
The industry desperately needs Samsung to bring some competition to TSMC's 5nm and with a really decent capacity too, I know Samsung has enough EUV machines ready to do this but they just need to perfect the node first.
Intel could then use TSMC where it wants, Samsung where it wants and still us it's own 10nm SF for some products and 14nm for it's bulk production of low performance (cheaper) products.

If Intel had swallowed it's pride in 2015 after seeing it was having problems with the 10nm node they could have scaled back their ambitions in an attempt to deliver something, then said if it still looks like the issues won't be solved by 2017 they should have started discussion with TSMC. Intel easily has enough money to have two design teams running side by side, one working on their own ill fated 10nm node and another using TSMC. That would have meant in 2019 if the in house FABs hadn't sorted themselves out Intel was ready to go at TSMC. They could have released high end products on TSMC nodes and cheaper volume products on their own FABs.

One other avenue I think should have been explored was Intel licencing TSMCs IP for the 7nm node maybe a deal all the way through to 3nm when everything is changing to GAAFET. TSMC doesn't have anywhere near the capacity to serve their long term customers and also help Intel out in a big way and Intel has the capacity just not the IP. So put these two together and TSMC could make an absolute fortune by licencing their node knowhow to Intel and Intel wouldn't be where it has been for the last 3 years and where it will be for the next 5!

Why would TSMC want to licence tech out to Intel? Intel would be probably the worst company to partner with in anything like this….

DevDrake
that's good considering how TSMC dominates the market.
Anyway no matter how much samsung will produce, nvidia is selling to miners first…

Is there any actual evidence of this?
https://profit-mine.com/dashboard
Rates a 3060Ti at $2.50 per day but (doesn't seem to list electricity prices so maybe net?).
https://cryptoage.com/en/2285-nvidia-geforce-rtx-3060ti-first-ethereum-mining-test.html
Reports a 3060 TI getting around 62MH/s at 140W.
https://whattomine.com
Calculates that as also being around $2.42 per day less electricity (so $2.08 if paying $0.10 per kw, or around $1.7 at $0.20 which is more realistic rate for Europe).

Don't see the return versus risk in that.
Sure, people who have bought a card for gaming can make some of their cost back but for running a mining farm I think $60pm (at current rates) it is too risky.

3dcandy
Why would TSMC want to licence tech out to Intel? Intel would be probably the worst company to partner with in anything like this….

Yes, totally agree. TSMC will take some Intel money but they would never fool themselves that Intel is a long-term partner and licencing their node tech would just be asking for long-term trouble. And Asian companies generally think (at least somewhat) long-term.

kompukare
Is there any actual evidence of this?

No, one website no one has ever heard off put some story behind a paywall as click bait.

Dribble
No, one website no one has ever heard off put some story behind a paywall as click bait.

Well the rumours about Lovelace (RTX 4000 series) say it will be on Samsungs 5nm node…

So although I'd take it with a pinch of salt, the rumours are beginning to get stronger