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

[McHuj]

+ Quote

And on Sale(shelves) in May-Jun...?

If it taped out in march, it's at least 3-4 months before the silicon comes back and another ~6-12 months before product.

 

[w0lfram]

So 2020 Holiday season...? 'ish….

 

[Samwell]

So 2020 Holiday season...? 'ish….

Earlier, 1 year from Tapeout till Sale is normal. As probably the big HPC chip taped out first, add 3 months to it, so May-June for HPC Ampere.
But the smaller chips coming after shouldn't need more than 1 year and will probably get released in a similar timeframe.

 

[DavidGraham]

During a press conference in Seoul, Nvidia Korea chief Yoo Eung-joon said that its production of advanced GPUs will be conducted by Samsung Electronics with its 7-nanometer extreme ultraviolet process.


http://www.koreaherald.com/view.php?ud=20190702000692#cb

 

[AlNom]

"conducted"

 

[bridgman]

"conducted"

You're thinking "semi-conducted" would have been more appropriate ?

 

[AlNom]

You're thinking "semi-conducted" would have been more appropriate ?

I think it's highly appropriate for Ampere.

 

[Kaotik]

Just to point out that they didn't actually confirm wether all of their 7nm will come from Samsung or just some, they've used Samsung before too at least for GP107

 

[Deleted member 13524]

Just to point out that they didn't actually confirm wether all of their 7nm will come from Samsung or just some, they've used Samsung before too at least for GP107

I think the GP108 (used in the very popular MX150) is also made on Samsung.

In the meanwhile, I've been seeing reports saying that TSMC's vanilla 7nm have just gotten a "free" 10% performance upgrade, and that process' yields have jumped significantly.
If 7nm DUV became so good, it makes me wonder what's the point with 7nm+ and 6nm.

 

[Kaotik]

I think the GP108 (used in the very popular MX150) is also made on Samsung.

Yep​

In the meanwhile, I've been seeing reports saying that TSMC's vanilla 7nm have just gotten a "free" 10% performance upgrade, and that process' yields have jumped significantly.
If 7nm DUV became so good, it makes me wonder what's the point with 7nm+ and 6nm.

or is that "free 10 performance upgrade" really the '6nm'?

 

[Deleted member 13524]

or is that "free 10 performance upgrade" really the '6nm'?

It's TSMC's 2nd gen 7nm.

I got it wrong though, it's reportedly over 5% and not 10%.

https://fuse.wikichip.org/news/2408...ls-2nd-gen-7nm-and-the-snapdragon-855-dtco/2/

TSMC also developed a 2nd generation of their 7nm process. This is an optimized process which uses the same design rules and DUV and is unrelated to 7nm+ which is EUV-based. This process is entirely design-compatible with the first generation but enjoys additional power and performance enhancements. For their second generation process, TSMC made some additional optimizations.

• Fin profile Optimizations
• Epi Optimizations
• MOL resistance Optimizations
• FEOL capacitance
• Metal gate Optimizations

All in all, the 2nd-generation 7nm process is said to deliver over 5% improvement in performance. Additionally, at the same leakage, at high frequencies, the second-generation 7nm process has improved the Vmin by 50 mV. Qualcomm has stated that the second-generation 7-nanometer node will be used for their next-generation cellular 5G modem which will deliver 2x peak data rate over their first-generation 5G modem.

Still the difference to 6nm will now be smaller and 7nm DUV yields are said to be really high, on the level of 16FF+.

 

[Kaotik]

It's TSMC's 2nd gen 7nm.

I got it wrong though, it's reportedly over 5% and not 10%.

https://fuse.wikichip.org/news/2408...ls-2nd-gen-7nm-and-the-snapdragon-855-dtco/2/

Still the difference to 6nm will now be smaller and 7nm DUV yields are said to be really high, on the level of 16FF+.

That still sounds awfully lot like the '6nm' which incidently uses same design rules as 7nm DUV.

 

[Deleted member 13524]

That still sounds awfully lot like the '6nm' which incidently uses same design rules as 7nm DUV.

6nm uses a number of EUV layers.
This 2nd gen 7nm is DUV only.

 

[Deleted member 13524]

BTW, TSMC is now claiming that 7nm is cheaper than 16nm, in cost per transistor.
Again, this is 7nm DUV. We're not into EUV territory yet.

https://twitter.com/x/status/1146417926481874944

 

[trinibwoy]

BTW, TSMC is now claiming that 7nm is cheaper than 16nm, in cost per transistor.
Again, this is 7nm DUV. We're not into EUV territory yet.

https://twitter.com/x/status/1146417926481874944

Not much of a cost improvement though. And given transistor counts are increasing much faster than density is improving there's only one direction for end consumer prices to go.

 

[xpea]

BTW, TSMC is now claiming that 7nm is cheaper than 16nm, in cost per transistor.
Again, this is 7nm DUV. We're not into EUV territory yet.

https://twitter.com/x/status/1146417926481874944

Very skeptical about this claim. And it doesn't take into account the huge increase in design cost and mask set

 

[del42sa]

Very skeptical about this claim. And it doesn't take into account the huge increase in design cost and mask set

true, but it´s Bits and Chips so take it with the same relevance as WCCFTECH

 

[Deleted member 13524]

Whoever wrote the tweet didn't author the graphs nor TSMC's claims.

 

[A1xLLcqAgt0qc2RyMz0y]

NVIDIA Next Generation ‘Ampere’ 7nm Graphics Cards Landing 1H 2020

https://wccftech.com/nvidia-next-generation-ampere-7nm-graphics-cards-landing-1h-2020

 

[McHuj]

Do we even know if Ampere is a successor to Volta or Turing? I don’t think NVIDIA has ever indicated.

I hope it’s for consumer and not enterprise. I hope they at least double RTX performance.