Nvidia Post-Volta (Ampere?) Rumor and Speculation Thread

Status
Not open for further replies.
https://www.pcgamesn.com/nvidia/nvidia-7nm-euv-graphics-card-samsung

next gen will use 7nm EUV from Samsung Foundries. Why they did not choose TSMC, despite having free capacity ?

How do you know that they don't choose both? Nvidia is Dual-Sourcing since Pascal and there is no reason to believe that they don't use TSMC again. Could be just next gen small Chips like with Pascal. Digitimes already mentioned, that Nvidia also has TSMC 7nm contracts for this year.
 
How do you know that they don't choose both? Nvidia is Dual-Sourcing since Pascal and there is no reason to believe that they don't use TSMC again. Could be just next gen small Chips like with Pascal. Digitimes already mentioned, that Nvidia also has TSMC 7nm contracts for this year.

I didn´t say they never will ( in the future ) , but they choose Samsung now for a reason , don´t you think so ?
 
I do remember reading some similar info and references to a TU116 chip a well. No idea if it's salvage with dead RT components or purposeful design of lower end chips without tensor/RT cores to produce small dies but with Turing/Volta design. I think I'd prefer the latter.
 
Is anyone here entertaining the idea that nvidia may be launching a GTX-11 series that seems to be Turing chips with the tensor and RT units disabled?

I've ignored those rumors for a while, but references for the 11-series started popping up on benchmark databases, not to mention official listings from laptop makers.

Do you mean "removed"? What would be the purpose of continuing the current Turing series at lower price points with large dies due to Tensor and RT cores? Would it be cheaper in the long run than having independent designs without those? It just sounds odd...

Edit - I really did not understand what they were saying at first, but the news is that the GTX2080 is read as GTX1180 when running on Linux? I would say that's more likely a mistake on the Software from when no one was sure of what the nomenclature was going to be...

Regarding laptop manufacturers, they were always different from Desktop with rebrands galore and different series (e.g. There was a mobile 8xx series while desktop jumped to 9xx). I bet he mobile 11xx series are Pascal rebrands.
 
Last edited:
I do remember reading some similar info and references to a TU116 chip a well. No idea if it's salvage with dead RT components or purposeful design of lower end chips without tensor/RT cores to produce small dies but with Turing/Volta design. I think I'd prefer the latter.

If they exist, I'd bet at salvaged Turing dies with non-functioning RT hardware and/or tensor units. Basically a Volta family of consumer GPUs.

We could imagine if launching these cards would be beneficial to nvidia at all, since they should be trying to push the RTX features as much as possible, and this segmentation would hurt the RTX branding.
Maybe they could launch it to select markets with different preferences, like Russia and China?
 
In what seems to be a bit of a shock, NVIDIA seems to be not interested in 7nm at this time, says it costs too much, and they already achieve good results with 12nm.

Nvidia does not seem to have any acute plans to switch to 7-nanometers at the moment. According to CEO Jen Hsun Huan, the process is too expensive and would bring too little benefit: " We do not want to pay the additional costs, but to gain benefits

He emphasizes that Nvidia could order appropriate wafers from TSMC at any time, but at the moment the Turing architecture is optimally positioned. The advantage of a new process is increased performance, combined with lower energy consumption, and the old 12-nanometer production is still at an advantage over its competitor AMD .

http://www.pcgameshardware.de/Nvidi...he-Kosten-Bezahlen-Vorteile-Erzielen-1278177/
 
Surely that's just something they're saying to avoid Osborning Turing, and they must have 7nm chips in the pipeline.
Surely they do have 7nm in the pipeline, but it wouldn't be surprising if they're waiting for EUV wafer prices which would put those chips with a 2020 release date.

It's not like they have lots of pressure from the competition.
 
I suspect EUV will continue to be a complete shitshow for ages ... those light sources will just eat masks for dinner, even with pellicles. They should have gone with FEL over a decade ago.
 
In what seems to be a bit of a shock, NVIDIA seems to be not interested in 7nm at this time, says it costs too much, and they already achieve good results with 12nm.
Is it really a shock though? They still don't have solid competition. Imagine if they went to 7nm soon with the next architecture, a 3080ti would have to cost consumers $2,000! :rolleyes:
 
Well, it's more of a shock though if the non-gaming architectures won't also use 7 nm. It'd be far easier to recover the costs for those.

Then again and as always, we don't know the full picture. Maybe 12 nm is discounted in a meaningful way to them too.
 
Why would they talk about a 7nm transition since they have just finished(?) rolling out their 12nm line with the GTX 1660?

Competition is non-existent and nV is probably well-informed about the 2019-2020 plans/landscape.
 
Why would they talk about a 7nm transition since they have just finished(?) rolling out their 12nm line with the GTX 1660?

Competition is non-existent and nV is probably well-informed about the 2019-2020 plans/landscape.

Which is probably why they launched the lower cost chips in part, if the leaks of Navi are true it seems to offer an incredible amount of value per dollar. But that doesn't mean they need, or should, start talking about 7nm.

Besides, first 7nm will, in tradition for both Nvidia and AMD as of late, probably be some uber expensive corporate chip for sims and AI and stuff. What, other than being built by Samsung apparently, the consumer or even pro 7nm Nvidia chips look like we don't know. Though with RTX underselling I think we can assume lower pricepoints and saving on silicon cost will be a priority from the outset, rather than trying to put out salvage chips after a stock price hit.
 
Surely they do have 7nm in the pipeline, but it wouldn't be surprising if they're waiting for EUV wafer prices which would put those chips with a 2020 release date.

It's not like they have lots of pressure from the competition.

Yes, i wouldn't expect any 7 nm DUV chips by Nvidia at this stage. 7nm DUV looks like a terrible node at the moment. It's not sure they would get more than 1,8x the density in 7nm vs 12nm, but price per Die area is 2 times that of 16nm according to AMD. Add the design cost for 7nm and it's probably more wise to jump directly to 7nm EUV, than changing your chips fast to the better EUV node. AMDs problem with Navi is, they just can't wait one more year.
 
Why would they talk about a 7nm transition since they have just finished(?) rolling out their 12nm line with the GTX 1660?
With the way the CEO talk about this, they didn't even order 7nm wafers from TSMC. If they are hiding their 7nm parts, they wouldn't use that kind of aggressive language.

Though with RTX underselling
Underselling compared to the mining boom, last week NVIDIA announced that Turing sold 45% better than Pascal in the first 8 weeks of launch.

nvidia-investor-day-turing-pascal-45-slide-1030x574.png


https://wccftech.com/nvidia-turing-...-than-pascal-in-first-eight-weeks-of-revenue/
 
Underselling compared to the mining boom, last week NVIDIA announced that Turing sold 45% better than Pascal in the first 8 weeks of launch.
That graph looks like revenue, as in $$ not units. Given the price of Turing cards were considerably higher than Pascal wouldn't this be a given? Or am I reading this wrong? Also combined with both 2080ti and non-ti at launch, compared to only 1080 for Pascal?
 
Last edited:
Several factors (if memory serves)
  • Turing had a price premium over Pascal '80 vs '80, '70 vs. '70 and so on.
  • Pascal was launched '80 and '70, much lower ASP than Turing which launched '80Ti and '80
  • Shortage of Pascal cards in the first couple of weeks (due to then-new 16nm process?)
 
Status
Not open for further replies.
Back
Top