NVIDIA is Secretly Working on a 5 nm Chip

[ARF]

Nvidia will be using capacity from TSMC and Samsung as been reported previously months ago.

Are the products taped-out or they are waiting a future tape-out in the next months?

 

[Valantar]

To be precise: as soon as a clueless news outlet reports it as "secret", it's no longer secret.

To be even more precise: just because the subject of a piece of information is not yet ready for publication does not mean that it is a secret per se, it just means that it is yet to be disclosed. Of course one could call this a trade secret, but then all chips are trade secrets if you look closely enough even when they are on the market - it's not like the designs are publicized below a very high level overview, after all. That a designer and maker of high end computer chips is working on a chip for an upcoming high end production node does thus not qualify as a "secret" IMO - it's both the most blindingly obvious thing ever (as they will be working on chips for all major nodes going forward, long before said nodes are available), and too general a piece of information to warrant actively keeping secret. What's next, is it a secret that Nvidia's design labs use electricity simply because they don't publicize it?

Why?! AMD's Radeon VII with Vega 20 on TSMC N7 node was released LAST year in January.
I am asking seriously, you are hiding from the question.

You know how product launches work, right? You announce a product when mass production has been going for long enough that the product can reach customers within a reasonable amount of time. In other words: while there are rumors galore about GPU designs both one and two years out, you don't ever get official commentary on them before the maker is ready to put them to market. Indirect information might be given out - such as updates on new architectures or related products for other markets (HPC/server/etc.) - but beyond that there won't be a peep from Nvidia until they are, at the very least, ready to give a launch event date. And so far Nvidia has had little reason to move on to the more expensive TSMC 7nm node as 12nm is cheap and still has them at efficiency parity/a small lead compared to RDNA 1, with a solid lead in absolute performance. They launched the Super lineup to compete better with RDNA, which gives them a further holdover period before they need to launch a new architecture - though competition is likely to heat up later this year. There's also the potential issue of die sizes, with Turing dice being massive which is problematic on smaller nodes. This might not be an outright issue, but it sure is good reason for Nvidia to hold off on adopting newer and smaller nodes until they have to, as even a 2080Ti shrunk to TSMC 7nm would be around the reticle limit for that node, which would both hurt yields and give little-to-no room for performance increases.

 

[Assimilator]

@EarthDog @Valantar please stop feeding that troll. Ignore list is your friend.

 

[zlobby]

Not so secretly, apparently.

 

[medi01]

Yes
AMD too
Apple too
Huawei too ...

Shocking news!

 

[TechLurker]

Personally, I just wish Nintendo just commissioned AMD for a custom SoC that has some actual graphical horsepower, and make it easier to port games to.

 

[mtcn77]

Personally, I just wish Nintendo just commissioned AMD for a custom SoC that has some actual graphical horsepower, and make it easier to port games to.

Something science fiction like is I have been speaking a lot about AMD pixel shading custom aa filters being too good.
It has been taken notice. New consoles do use them, despite not exactly.
What I want is a default custom 2x rgssaa ed-cfaa mode. It is all present if you check all settings via radeonmod.
The funny thing is, 2x rgssaa makes for 4x msaa across axis points and absolutely 0 filtering across diagonals. Ironically, this makes for the best showcase if you want ed-cfaa to shine through funny enough, since it evokes a perfect mask to map which pixels are missing samples(it could not have been any better, it is a dotted subpixel line neither sharpness or blurriness introduced by the filter, zero width transparent edges mean multisampling those edges has taken zero interpolation brute forcing handicap via aa). It is quite efficient in regard of its sample count. AMD could couple that with eqaa even, haven't tried that.

 

[renz496]

right? you are telling me companies are looking towards and working towards the future? like how GTA6 started development in 2014? no freaking way dude

but yeah onto the article, it would be interesting if Nintendo already has a contract with Nvidia for a next console.

nintendo already give the hint about this back in 2017. just after switch release they are talking how they were experimenting with AI to see how it can help them improve the experience. at that time they most likely already have access to nvidia xavier which have tensor cores. xavier is heavily build for auto application so nvidia and nintendo might work on custom tegra chip without all the hardware needed for auto application.

 

[yeeeeman]

Journalism today has become...gossip. Women gossip. Unprofessional, stupid, without any drop of respect for manufacturers, nothing.
Nvidia, AMD, Intel, Qualcomm, Apple, you name it, they are all working "secretly" on unannounced products on various nodes.
It is not interesting to write news articles to specify obvious things and to be honest, hardware scene is becoming a bit boring, because these pesky so called journalists like to scratch every manufacturers ass and see what is there.
Lets learn to enjoy the products we have today. Lets leave the products from tomorrow to be an unknown, so that when they launch they will be a surprise. Otherwise everything's just a freaking rumour...

 

[Xaled]

The only things I conclude from this news, expect to hear news of new AMD GPUs.