Nvidia to shift RTX 30 GPUs to TSMC 7nm in 2021, says report

Quote:

---

Currently all Nvidia GeForce RTX 30 GPUs are fabricated by Samsung on 8nm.

---

Read more.

IF the TSMC cards perform closer to what was originally expected, another kick to Samsungs capabilities on this front...

First thoughts - If Samsung gave a considerable discount does moving to 7nm TSMC mean a higher price for the graphics card....

Quote:

---

It is possible that mid-to-lower-end GeForce Ampere GPUs will still be fabricated by Samsung.

---

It could go the other way imo, TSMC does the higher volume 'low end' cards, even more so if Samsung is having yield issues like has been reported.

Got to be honest, this cluster **** of a rtx3xxx release is completely screwing up any potential plans to upgrade at the start of the new year...

Nvidia have stopped selling founders edition (I want a founders card) on their website and are now only selling via best buy in the US, nowhere in the EU let alone UK. I really wish I could use an AMD card because honestly, if the 'leaks' are true, I'd grab an AMD one without any hesitation if it wasn't for my software only supporting cuda.

Makes me want to wait. Oh well, virtual flying will have to wait as well :p

Quote:

---

Originally Posted by LSG501

First thoughts - If Samsung gave a considerable discount does moving to 7nm TSMC mean a higher price for the graphics card....

It could go the other way imo, TSMC does the higher volume 'low end' cards, even more so if Samsung is having yield issues like has been reported.

Got to be honest, this cluster **** of a rtx3xxx release is completely screwing up any potential plans to upgrade at the start of the new year...

Nvidia have stopped selling founders edition (I want a founders card) on their website and are now only selling via best buy in the US, nowhere in the EU let alone UK. I really wish I could use an AMD card because honestly, if the 'leaks' are true, I'd grab an AMD one without any hesitation if it wasn't for my software only supporting cuda.

---

Undoubtedly it means higher prices, they would have to manage two different sets of yields (memory is Samsung) from two different foundries for whichever products they decide to produce on different processes. I can't see that being cheaper, even with good yields when you're talking volumes of any product the costs can increase massively (also the other way). Then there will also be the question of which process uses less power etc.

Nvidia have fluffed the launch of the 3xxx series, including the issues people are having with unstable drivers etc. Adding into that their decision to halt sales of founders edition cards on their main site (because Digital River are incapable of countering bots), it makes you wonder who they "may" decide to partner with in the EU countries and the UK to sell the founders edition card. Personally I think it was a decision so they could concentrate on getting US customers cards while they also fix supply issues to AIBs / System builders etc. Either way, with this news, now is not the time to purchase one.

All this screams to me is that faith in Samsungs 8nm process is limited from Nvidias front. I suspect that when 7nm failed to deliver in Samsung, the shift to 8nm was open but Nvidia tried to claw back into TSMC but they said not on your nelly on that timeline. So Nvidia was forced to suck up a bad node into their portfolio while trying their hardest to get back onto the superior TSMC 7nm node for their enterprise grade GPUs.

On top of that, it looks like the 8nm process is not conducive to an architecture sat in DL pod, it's too warm and needs too much power.

I really hope in terms of performance/transistor RDNA2 is a good uarch. If not Ampere on 7NM might be a monster.

this is why I wait with buying new hardware, the fanboy/girl thing, is not a factor, I love technology and such, but not enough to just buy buy and buy, it gotta offer numbers, so I am still looking at what AMD is serving us next.

So 3080-70 Super's or Ti's in 2021 then.

I am surprised the next node isn't 5nm, because reports world-wide are all saying 5nm has consistently better yields over 7.

Quote:

---

Originally Posted by six_tymes

I am surprised the next node isn't 5nm, because reports world-wide are all saying 5nm has consistently better yields over 7.

---

They will go to 5nm after 7 I'm sure, but there isn't the capacity at the moment.

Quote:

---

Originally Posted by Iota

Undoubtedly it means higher prices, they would have to manage two different sets of yields (memory is Samsung) from two different foundries for whichever products they decide to produce on different processes. I can't see that being cheaper, even with good yields when you're talking volumes of any product the costs can increase massively (also the other way). Then there will also be the question of which process uses less power etc.

Nvidia have fluffed the launch of the 3xxx series, including the issues people are having with unstable drivers etc. Adding into that their decision to halt sales of founders edition cards on their main site (because Digital River are incapable of countering bots), it makes you wonder who they "may" decide to partner with in the EU countries and the UK to sell the founders edition card. Personally I think it was a decision so they could concentrate on getting US customers cards while they also fix supply issues to AIBs / System builders etc. Either way, with this news, now is not the time to purchase one.

---

Memory IS NOT Samsung it's Micron and developed for NV. It's also the reason for high heat and cost

Quote:

---

Originally Posted by Tabbykatze

All this screams to me is that faith in Samsungs 8nm process is limited from Nvidias front. I suspect that when 7nm failed to deliver in Samsung, the shift to 8nm was open but Nvidia tried to claw back into TSMC but they said not on your nelly on that timeline. So Nvidia was forced to suck up a bad node into their portfolio while trying their hardest to get back onto the superior TSMC 7nm node for their enterprise grade GPUs.

On top of that, it looks like the 8nm process is not conducive to an architecture sat in DL pod, it's too warm and needs too much power.

---

Basically yes... I hear from people in the know that the prices quoted by Samsung are so good they couldn't wait to snap them up but yields are truly awful 25%-40% worse than TSMC 7nm. TSMC politely told them to go forth and multiply so they were stuck with Samsung

Quote:

---

Originally Posted by 3dcandy

Memory IS NOT Samsung it's Micron and developed for NV. It's also the reason for high heat and cost

---

You are correct, I'm not sure why I thought that. :eek:

Quote:

---

Originally Posted by Iota

You are correct, I'm not sure why I thought that. :eek:

---

Hahaha I believe you got your memory manufacturers and fabs messed up

Quote:

---

Originally Posted by 3dcandy

Hahaha I believe you got your memory manufacturers and fabs messed up

---

I honestly have no idea, put it down to a senior moment! :p