Samsung to manufacture NVIDIA GeForce RTX 3000 SUPER Series GPUs next year (Updated)

[BiG StroOnZ]

NVIDIA looked to Samsung rather than TSMC for the production of its RTX 30 series GPUs. The decision to go with Samsung has caused issues though, not least poor yields on its 8 nm process. Nonetheless, NVIDIA will continue to use Samsung for its RTX 3000-series SUPER GPUs, according to a new report. The RTX 30 SUPER series is due next year.

 

Quote

NVIDIA will not be taping up TSMC's 7 nm nodes anytime soon, according to a new report by The Korea Economic Daily. Instead, Samsung and NVIDIA have reputedly signed a second deal for the manufacture of GPUs. The Korea Economic Daily has not divulged the details of the new deal between Samsung and NVIDIA, but it is expected to run to hundreds of billions of KRW (South Korean won). NVIDIA has already announced a new GeForce RTX hardware event for January 12 where it is expected to present new graphics cards. Based on recent rumors, NVIDIA may launch multiple cards next month, including the RTX 3050 and RTX 3060.

 

The additional deal is speculated to extend to the supply of GPUs for a SUPER refresh of the RTX 30 series too, though. If that is the case, then we would expect NVIDIA to begin selling these towards the end of 2021.

 

Source 1: https://www.notebookcheck.net/Samsung-tipped-to-manufacture-NVIDIA-GeForce-RTX-30-SUPER-GPUs-next-year.510917.0.html

Source 2: https://www.kedglobal.com/newsView/ked202012170015

 

I find this particular story quite interesting, because it makes the lack of supply issue "suspicious" (because of the timing). In that, what if there really isn't a supply issue per se, but they are in actuality preserving the best dies from current yields for this upcoming SUPER series release. Despite issues with production, they (NVIDIA) seem to have still renewed a second deal here with Samsung. If they are stockpiling the best binned dies for the SUPER series release of the 3000-series, that would make sense. Nobody suspected AMD to be as competitive with big Navi as they were (RDNA 2 / RX 6000-series). NVIDIA probably figured it would be best to rush the SUPER refresh parts to gain that slight edge over AMD's competing offerings (thus EOLing non-SUPER parts, and going straight to SUPER parts). With Turing though, once the SUPER series launched, quite quickly, non-SUPER parts went EOL (perhaps instead of refreshing stock of non-SUPER stuff, they forgo it altogether and pivot to the SUPER refresh parts immediately). Here though, if the Ampere SUPER parts prove to be slightly faster (5-12%) than their non-SUPER counterparts (as did with Turing), this is "justification" for charging even more or slightly more (rather than the same, especially if demand allows it). Which begs the question, is the scalping going to in actuality be even worse then, than it is now...? Or will supply meet the demand this time around because Samsung has been prepping for this refresh all along?

 

Update to this story (More info, specifically on the 3080 Super and 3070 Super) ~

 

NVIDIA rumored to introduce GeForce RTX 3080 SUPER and RTX 3070 SUPER series, NVIDIA to give a SUPER Overhaul to its GeForce RTX 3070 and RTX 3080 Graphics Cards:

 

Quote

 

It is rumored that despite preparing as many as four new SKUs, NVIDIA is already looking forward to launching another two high-end SKUs. As mentioned before, NVIDIA is currently focusing on two SKUs: GeForce RTX 3080 Ti and RTX 3060. The latter is expected to make a debut in late January and we actually expect NVIDIA to talk about this SKU in the upcoming news conference, currently planned for January 12th. The high-end GeForce RTX 3080 Ti SKU is now expected to debut in February, possibly after the Chinese New Year. 

 

But joining is the new rumor of the GeForce RTX 3080 SUPER is the GeForce RTX 3070 SUPER... but why? Well, the RTX 3080 has 10GB of VRAM and the RTX 3070 has 8GB of VRAM. Both of their competitors in the Radeon RX 6800 XT and Radeon RX 6800 have 16GB of VRAM and I'm sure NVIDIA wants to something about that. The new rumors are coming from kopite7kimi on Twitter, who says that there are "some new SKUs" coming and they might be the GeForce RTX 3080 SUPER and GeForce RTX 3070 SUPER graphics cards. We could expect the RTX 3080 SUPER to be a tweaked RTX 3080 with 16GB of VRAM -- while the higher-end RTX 3080 Ti would pack 20GB of VRAM. As for the GeForce RTX 3070 SUPER, Kopite ponders if it'll feature a new GA103 GPU that hasn't been used yet -- both the GeForce RTX 3070 Ti and GeForce RTX 3070 SUPER could use this new GA103 -- slotting right between GA104 and GA102.

 

Source 3: https://videocardz.com/newz/nvidia-rumored-to-introduce-geforce-rtx-3080-super-and-rtx-3070-super-series

Source 4: https://www.tweaktown.com/news/77041/nvidia-geforce-rtx-3080-super-3070-teased-to-come-soon/index.html

Source 5: https://www.techpowerup.com/276690/nvidia-could-give-a-super-overhaul-to-its-geforce-rtx-3070-and-rtx-3080-graphics-cards

 

[Spotty]

It's important to note that neither Samsung or Nvidia have disclosed what the new deal is for. There is speculation that it is for Super or Ti cards, however there is also speculation that this new contract is actually for the new generation mobile GPUs which seems more plausible and less interesting. Maybe the new contract is just Nvidia securing more of Samsung's production capacity to meet the rising demand from customers?

 

From TechRadar:

Quote

The deal is "worth hundreds of billions of won," and with one dollar equaling about 1,098 South Korean won, this is very likely a multibillion dollar contract. That can buy a whole lot of processors, but if you're hoping to see a flood of RTX 3080s on the digital store shelves soon, the report doesn't say what those processors are for.

 

While they could be used to produce enough RTX 3080s and RTX 3090s to meet the existing, unprecedented demand for the new cards, these processors could also be used in new RTX cards as well, including the RTX 3080 Ti or mobile versions of the RTX 30-series GPUs for laptops.

 

Hopefully Nvidia continuing/extending their deal with Samsung to produce GPUs is a sign that Nvidia is confident in Samsung's ability to deliver and meet demand and a sign of improved availability to come. (we can hope!)

[GoodBytes]

55 minutes ago, BiG StroOnZ said:

I find this particular story quite interesting, because it makes the lack of supply issue "suspicious" (because of the timing). In that, what if there really isn't a supply issue per se, but they are in actuality preserving the best dies from current yields for this upcoming SUPER series release.

Assuming the news is true, I think Nvidia saw that Samsung is actually fixing its issues at a rapid rate, and yields are rapidly improving. That said, if what you are saying is true, then they won't keep the chips for the Super model, but rather the coming up Ti model (rumor). Nvidia has an annoncement in Jan, about something "exciting for gamers" (or so they claim).

 

But I think that it is either for the mobile GPUs, or it is filled with conditions for Samsung to meet (else the contract may be dropped), or it is just an intent that was discussed only. In other words, talks, but nothing on paper yet.

 

Quote

Despite issues with production, they (NVIDIA) seem to have still renewed a second deal here with Samsung. If they are stockpiling the best binned dies for the SUPER series release of the 3000-series, that would make sense. Nobody suspected AMD to be as competitive with big Navi as they were (RDNA 2 / RX 6000-series).

So currently, AMD GPU are crazy powerful. Not the king, but close. They are excellent graphics card. In my opinion, if they tune their pricing a bit better (which I expect them to do in Feb-March, once Nvidia stock issue should start to be more notably improved, and theirs), then they would be a real good deal for what you get, but anyways, back to topic. AMD GPUs are floating in the 3000 series performance range. This means on the next gen GPU (which could be next year for AMD), they'll go to the TSMC new process (5nm, apparently), and that will automatically put their GPUs over Nvidia 3000 series... mix that additional improvements to the architecture, they would be on top (especially if they manage to improve further the Ray Tracing performance)

 

All to say, I doubt there will be a SUPER... I think Nvidia will release a 4000 series, and that would be the same architecture of the 3000 series with improvements... kinda like what the 500 series was from the 400 series. The 500 series was still the same architecture as the 400 series, but with various improvements and issues fixed, allowing the GPU to operate much more efficiently providing not only cooler operation and lower power draw, but also added performance.

 

Quote

NVIDIA probably figured it would be best to rush the SUPER refresh parts to gain that slight edge over AMD's competing offerings. With Turing though, once the SUPER series launched, quite quickly, non-Super parts went EOL. Here though, if the Ampere SUPER parts prove to be slightly faster (5-12%) than their non-SUPER counterparts (as did with Turing), this is "justification" for charging even more or slightly more (rather than the same, especially if demand allows it). Which begs the question, is the scalping going to in actuality be even worse then, than it is now...? Or will supply meet the demand this time around because Samsung has been prepping for this refresh all along?

Assuming that is true, at a marketing level, having "SUPER" branded 3000 series will appear as if it is old stuff to AMD new 7000 series GPU. I don't see them doing this. I think it will be the 4000 series. I guess we will see. I think it will depend on the info Nvidia has on AMD. They are best to know if AMD will/can release their next gen GPU next year or not.

 

[dizmo]

I mean....their yields aren't that bad. Look at AMD. They went with the known player and their availability is abysmal, even worse than Nvidia in a lot of places.

It wouldn't surprise me if they did go with Samsung again.

[Gaires]

Wait wait wait, so we're gonna have the normal versions, TI and Super cards?

 

And I thought the RTX 2000 series was confusing with too many models.

[RejZoR]

So, all the idiots paying ridiculous prices to scalpers right now for RTX 3080 burned themselves majorly.  Not only they paid stupid price, they don't even have the best GPU and we're talking a span of 2-3 months. It doesn't even matter that they were launched in September 2020 since availability was a turd since day 1. New better cards will be launching when old ones will get better availability. Or they'll replace them with Super models and we're back to September 2020 with availability. Great, this will just keep on sucking through entire 2021 then...

[D13H4RD]

3 hours ago, dizmo said:

I mean....their yields aren't that bad. Look at AMD. They went with the known player and their availability is abysmal, even worse than Nvidia in a lot of places.

It wouldn't surprise me if they did go with Samsung again.

It doesn't help that a heap of AMD's products use TSMC's 7nm node. Zen 3, RDNA2, console APUs, Zen 2, etc. etc. 

[Spotty]

59 minutes ago, Gaires said:

Wait wait wait, so we're gonna have the normal versions, TI and Super cards?

 

And I thought the RTX 2000 series was confusing with too many models.

The 1600/2000 series cards had normal, super, and Ti versions as well. 

 

It'll only get confusing if there's a "3080 Super Ti"

[Gaires]

8 minutes ago, Spotty said:

The 1600/2000 series cards had normal, super, and Ti versions as well. 

 

It'll only get confusing if there's a "3080 Super Ti"

1650

1650 super

1660

1660 TI

1660 Super

2080

2080 Super

2080 TI

 

It does get really confusing fast for persons who don't follow tech regulary. There are just too many cards now, I don't understand why we need a card between every 50 bucks or so, but guess it makes Nvidia and AMD more money.

 

And if the rumours are true and we will get the 3060 8GB, 3060 16gb and 3060 TI, plus other TI models and super cards on top it...

Edited December 20, 2020 by Gaires
Fixed Vram amounts.

[Grant Chen]

They’ll probably refresh the new cards with 7nm EUV, just like GA100.

[Granular]

I don't get the criticism of Nvidia over going with Samsung. Going with TSMC would have also meant supply issues since TSMC doesn't have enough capacity for the clients it already had. If anything, it's good that there's market for Samsung's advanced nodes, otherwise they might abandon them, like Global Foundries did.

Sure, not being able to get a GPU sucks, but that was going to be the case either way.

[AndreiArgeanu]

5 hours ago, dizmo said:

Look at AMD. They went with the known player and their availability is abysmal

But at the same time AMD shares their fab capacity between APU's, CPU's and GPU's, and they're all in high demand at the moment, it would have been shameful for NVIDIA to release earlier than AMD and still have less stock, since their main (and high demand) products are consumer and enterprise GPU's, (and APU's for the nintendo switch, but I don't think Samsung makes those for NVIDIA)

[D13H4RD]

1 hour ago, Gaires said:

1650

1650 super

1660

1660 TI

1660 Super

2080

2080 Super

2080 TI

 

It does get really confusing fast for persons who don't follow tech regulary. There are just too many cards now, I don't understand why we need a card between every 50 bucks or so, but guess it makes Nvidia and AMD more money.

 

And if the rumours are true and we will get the 3060 8GB, 3060 16gb and 3060 TI, plus other TI models and super cards on top it...

Typically, a Ti supplements a non-Ti variant whilst the Super is a refreshed variant which replaces the non-Super one entirely. 

 

Except there were still RTX 2060s when the 2060 Super was released... 

[Gaires]

13 minutes ago, D13H4RD said:

Except there were still RTX 2060s when the 2060 Super was released... 

Reason for 2060 was that 5600xt and 5700 were such a good value around that  price range. So the discounted 2060 was Nvidias answer to that problem.

Assuming rest ofthe rx 6000 series is good, it's not out of  this world to assume Nvidia would keep their initial cards at discounted price, alongside the supers.

[porina]

6 hours ago, Spotty said:

It's important to note that neither Samsung or Nvidia have disclosed what the new deal is for.

I'll throw in one more interpretation: this is simply a continuation of their existing relationship. I'm not sure it matters what they make, if the contract is to secure ongoing service from Samsung. Contracts would typically be fixed in some way, so both sides know what to plan for. If they're happy (or otherwise don't have a better choice) then they can renew.

 

6 hours ago, GoodBytes said:

This means on the next gen GPU (which could be next year for AMD), they'll go to the TSMC new process (5nm, apparently), and that will automatically put their GPUs over Nvidia 3000 series... mix that additional improvements to the architecture, they would be on top (especially if they manage to improve further the Ray Tracing performance)

TSMC 5nm is already in production, used by Apple at the least. I guess it'll come down to how well AMD secured access to capacity when they might move to it, and with what products. Not everything needs to move at once.

 

I also wouldn't assume a node advance to automatically mean a big performance jump. Look at original navi launch. Maybe they've learnt their lesson and can continue their momentum, but I wouldn't bet on it.

 

6 hours ago, GoodBytes said:

Assuming that is true, at a marketing level, having "SUPER" branded 3000 series will appear as if it is old stuff to AMD new 7000 series GPU. I don't see them doing this. I think it will be the 4000 series.

Given the long gaps we have recently seen for GPUs from both sides, I'm not so sure AMD would increase their major refresh cycle to yearly. I had written elsewhere it would be nice if both sides gave smaller improvements every year, not a bigger jump every two years say. Will it happen?

 

I don't think the naming itself is that important here. If people know about this stuff, they'll understand anyway. If people don't, would they know how old a generation is relative to another?

 

2 hours ago, D13H4RD said:

It doesn't help that a heap of AMD's products use TSMC's 7nm node. Zen 3, RDNA2, console APUs, Zen 2, etc. etc. 

So they failed to predict the future, or they did and simply couldn't do anything about it. As great as TSMC might be with their 7nm and newer nodes, it only helps if you have sufficient access to it.

 

1 hour ago, Gaires said:

It does get really confusing fast for persons who don't follow tech regulary. There are just too many cards now, I don't understand why we need a card between every 50 bucks or so, but guess it makes Nvidia and AMD more money.

Not all the cards will be available at the same time. When the Super refresh generation comes out, it largely replaces the non-Super versions.

[CTR640]

14 hours ago, dizmo said:

I mean....their yields aren't that bad. Look at AMD. They went with the known player and their availability is abysmal, even worse than Nvidia in a lot of places.

It wouldn't surprise me if they did go with Samsung again.

Don't forget the consoles as the cpu's and gpu's are from AMD and nVidia only has graphics department. It's really a fair comparisons.

[dizmo]

13 hours ago, RejZoR said:

So, all the idiots paying ridiculous prices to scalpers right now for RTX 3080 burned themselves majorly.  Not only they paid stupid price, they don't even have the best GPU and we're talking a span of 2-3 months. It doesn't even matter that they were launched in September 2020 since availability was a turd since day 1. New better cards will be launching when old ones will get better availability. Or they'll replace them with Super models and we're back to September 2020 with availability. Great, this will just keep on sucking through entire 2021 then...

Not really. We get a revision every year. It's just business as norm. They're likely not releasing the Super/Ti cards for many months.

12 hours ago, D13H4RD said:

It doesn't help that a heap of AMD's products use TSMC's 7nm node. Zen 3, RDNA2, console APUs, Zen 2, etc. etc. 

Doesn't change the fact they had shit availability though.

12 hours ago, Gaires said:

1650

1650 super

1660

1660 TI

1660 Super

2080

2080 Super

2080 TI

 

It does get really confusing fast for persons who don't follow tech regulary. There are just too many cards now, I don't understand why we need a card between every 50 bucks or so, but guess it makes Nvidia and AMD more money.

 

And if the rumours are true and we will get the 3060 8GB, 3060 16gb and 3060 TI, plus other TI models and super cards on top it...

It's not that confusing. If you don't take the time to research what you're getting for your PC (which honestly isn't that hard) then it's kind of your own fault for not knowing what you're getting.

10 hours ago, AndreiArgeanu said:

But at the same time AMD shares their fab capacity between APU's, CPU's and GPU's, and they're all in high demand at the moment, it would have been shameful for NVIDIA to release earlier than AMD and still have less stock, since their main (and high demand) products are consumer and enterprise GPU's, (and APU's for the nintendo switch, but I don't think Samsung makes those for NVIDIA)

That's a poor excuse. It's not like Samsung has an entire fab just for Nvidia.

1 hour ago, CTR640 said:

Don't forget the consoles as the cpu's and gpu's are from AMD and nVidia only has graphics department. It's really a fair comparisons.

What's your point?

[AndreiArgeanu]

8 hours ago, dizmo said:

That's a poor excuse. It's not like Samsung has an entire fab just for Nvidia.

Neither does TSMC for AMD. And that wasn't even the point I was trying to make.

[Mark Kaine]

On 12/20/2020 at 4:12 AM, BiG StroOnZ said:

I find this particular story quite interesting, because it makes the lack of supply issue "suspicious"

yeap, I said this before... I thought they'd be moving to TSMC later though (as has been the plan afaik) but staying with Samsung makes also sense I guess... they've probably improved production and continue doing so with more experience. TSMC might have just told them off though with them making AMD console chips and whatnot... 

 

My take away however is that I should definitely really cancel my evga step up for a 3070, since I'd definitely would rather want a Super... ughhh! 

[BiG StroOnZ]

14 hours ago, Mark Kaine said:

My take away however is that I should definitely really cancel my evga step up for a 3070, since I'd definitely would rather want a Super... ughhh! 

 

One thing I've learned during my many, many years as a PC Enthusiast is: the longer you wait to purchase a specific part (if you can wait), the better off you are; as there is always something better performing per dollar in the pipeline (often for equal or even less money than you originally planned on spending). 

 

Perfect example of this phenomenon was when I was looking to upgrade my GPU to a GTX 980 Ti (Maxwell) nearly six years ago (2015) while I was still using a GTX 780 (Kepler). I was dead set on this purchasing decision, with the 980 Ti usually costing around $650-750. Somehow though, I managed to bite the bullet and tread on without upgrading because I figured based on the leaks coming out at the time that something better was right around the corner. Then, miraculously, Pascal launched and with the release of the GTX 1070. I said "sweet", now I'm getting 980 Ti performance or practically better for around $380-450 (this was in 2016). At this point my Triple A gaming habits slowed down tremendously and I started getting into Free-to-play titles and/or games that generally take a long time to play (RPGs, ARPGs, MMOs, etc.) or playing DLCs of games I already owned. Therefore my GPU horsepower requirements lessened and there wasn't much haste in upgrading. Of course, also, it didn't help that there were rumors of a 1070 Ti coming soon (at that time) for around $400-450, thus allowing me to continue the waiting game (this was 2017). Fast forward two years (2019) after another mining wave, the same disinterest in new games, and enjoying the world of RPGs / DLCs. At last, a worthy contender arose, the GTX 1660 Ti (Turing) for $280. Finally, I figured it was worth it to pull the trigger. By this time I should also mention/note, I had saved Newegg Gift Cards for nearly five years . Basically translating to paying around $120 + shipping/tax out of pocket total for the GTX 1660 Ti. And in the end, I got not only GTX 980 Ti power, not only GTX 1070 power, but practically GTX 1070 Ti power (after overclocked) for practically 80% less money than I originally planned on spending (with the 980 Ti) or around 70% less (with the GTX 1070/1070 Ti). 

[exboxchansan]

I don’t know much about semiconductor manufacturing, but how are exactly going to preserve the best dies from this generation for the super series? Aren’t those dies hardware updates

[leadeater]

19 minutes ago, exboxchansan said:

I don’t know much about semiconductor manufacturing, but how are exactly going to preserve the best dies from this generation for the super series? Aren’t those dies hardware updates

Generally not, the die revisions of say for example GA104 can realistically just be microcode and difference in laser disabled sections of the die (if done at all). So it's not a problem to bank a whole bunch of dies that validate very well, as the manufacturing of the dies aren't any different between products that use the same die. This is not true when there is actually a change to the die design but that's not that common.

[pas008]

New cards might have 21 gbps instead of 19?

[Amias]

Can you just release a card I can actually buy please?

 

 

[Mark Kaine]

On 12/21/2020 at 11:44 AM, Mark Kaine said:

TSMC might have just told them off though with them making AMD console chips and whatnot... 

 

Called it. whatnot = Apple