AMD Navi 23 ‘NVIDIA Killer’ GPU Rumored to Support Hardware Ray Tracing, Coming Next Year

melmac · 6 Oct 2020 at 13:22

LePhuronn said:
7nm
7nm+
7NP
6nm

Can't remember if was Anand or Tom's who listed them...

6nm is 6nm process, not 7nm?

they have the following
N7
N7P which is just a slightly improved 7nm process.
N7+ which is 7nm with EUV.

EDIT: N7+ is not compatible with the other 2 or the 6nm process Kentman mentioned.

Sargatanas2511 · 6 Oct 2020 at 13:25

melmac said:
6nm is 6nm process, not 7nm?

they have the following
N7
N7P which is just a slightly improved 7nm process.
N7+ which is 7nm with EUV.

6nm is based on 7nm node and is an evolution, not a new node. Just like Samsung's 8nm is an improved 10nm process.

LePhuronn · 6 Oct 2020 at 13:26

melmac said:
6nm is 6nm process, not 7nm?

Apparently it's the same design rules as N7 so you can port it
https://www.anandtech.com/show/1422...echnology-7-nm-with-higher-transistor-density

Whether it's part of the 7nm family or a node unto itself I guess is the question.

melmac · 6 Oct 2020 at 13:27

Sargatanas2511 said:
6nm is based on 7nm node and is an evolution, not a new node. Just like Samsung's 8nm is an improved 10nm process.

It's not counted as a 7nm node though by TSMC.

melmac · 6 Oct 2020 at 13:29

LePhuronn said:
N6 is N7+ with a higher transistor density

N6 is not N7+.

N6 is a 6nm version of N7.

LePhuronn · 6 Oct 2020 at 13:31

melmac said:
N6 is not N7+.

N6 is a 6nm version of N7.

Anand:

TSMC states that their N6 fabrication technology offers 18% higher logic density when compared to the company’s N7 process (1st Gen 7 nm, DUV-only), yet offers the same performance and power consumption. Furthermore, according to TSMC N6 'leverages new capabilities in extreme ultraviolet lithography (EUVL)' gained from N7+, but does not disclose how exactly it uses EUV for the particular technology. Meanwhile, N6 uses the same design rules as N7 and enables developers of chips to re-use the same design ecosystem (e.g., tools, etc.), which will enable them to lower development costs. Essentially, N6 allows to shrink die sizes of designs developed using N7 design rules by around 15% while using the familiar IP for additional cost savings.

Sounds like a tasty little hybrid of a few things. Same design rules as N7, +18% density, uses EUV techniques learned from N7+

gpuerrilla · 6 Oct 2020 at 13:31

Its like on the nvidia thread before I asked them to rename it.

james.miller · 6 Oct 2020 at 13:34

N6 is EUV. it's closer to N7+ than it is to N7. What's debatable is is '6nm', because as far as i know nothing has shrunk except the space between the transistors. But, it's called N6 so there you go. I think it's probably not important in the grand scheme of things whether TSMC called it N6 or n7++ as quick porting from n7/n7+ to n6 is one of the advantages TMSC were touting - so it must be very similar anyway.

LePhuronn · 6 Oct 2020 at 13:34

So @melmac care to hazard a guess which processed AMD are using for all their 2020 products? Many people are suggesting AMD have supply issues because "everything is 7nm", but that's not technically the case. All we do know if AMD are not producing Zen 3 on N7.

melmac · 6 Oct 2020 at 13:35

LePhuronn said:
Apparently it's the same design rules as N7 so you can port it
https://www.anandtech.com/show/1422...echnology-7-nm-with-higher-transistor-density

Whether it's part of the 7nm family or a node unto itself I guess is the question.

IT doesn't really matter. The difference between it and N7 seems to be tiny. The article you linked to even mentions that companies would be better off using the 7nm+ or 5nm process

Just seems to offer a little bit of cost savings for companies who have N7 products that don't want to develop new IP for N7+. Might be good option for AMD to offer RDNA 1 cards at an even cheaper price next year.

LePhuronn · 6 Oct 2020 at 13:36

james.miller said:
N6 is EUV. it's closer to N7+ than it is to N7. What's debatable is is '6nm', because as far as i know nothing has shrunk except the space between the transistors. But, it's called N6 so there you go. I think it's probably not important in the grand scheme of things whether TSMC called it N6 or n7++ as quick porting from n7/n7+ to n6 is one of the advantages TMSC were touting.

What's the interesting point is the ability to take N7 designs directly to N6 production. Isn't that what AMD did with the RX 590? Took a GloFo 16nm Polaris, made it on GloFo 12nm because they're design compatible and then clock the lads off it?

LePhuronn · 6 Oct 2020 at 13:40

melmac said:
IT doesn't really matter. The difference between it and N7 seems to be tiny. The article you linked to even mentions that companies would be better off using the 7nm+ or 5nm process

True true, I'm thinking about TSMC's manufacturing capacity. Would N6 be produced on different machines to N7? I'm spitballing the notion that the console APUs are on N6 so they don't get in the way of Zen 3 and RX 6000 production, y'see. And in turn if Zen 3 is on either N7+ or N7P, with RX 6000 on the other, then arguably there's no manufacturing backlogs, other than TSMC producing enough wafers of each "type" to fulfil demand for the products on that "type".

Sargatanas2511 · 6 Oct 2020 at 13:41

melmac said:
It's not counted as a 7nm node though by TSMC.

Because companies never rebrand or advertise things in stupid ways right?

KentMan · 6 Oct 2020 at 13:42

LePhuronn said:
So @melmac care to hazard a guess which processed AMD are using for all their 2020 products? Many people are suggesting AMD have supply issues because "everything is 7nm", but that's not technically the case. All we do know if AMD are not producing Zen 3 on N7.

In all honesty AMD could respin 5700/X onto 6nm and even the 5600XT etc and have them as entry level against the RDNA2 cards, who would be mid and high end.

KentMan · 6 Oct 2020 at 13:44

LePhuronn said:
True true, I'm thinking about TSMC's manufacturing capacity. Would N6 be produced on different machines to N7? I'm spitballing the notion that the console APUs are on N6 so they don't get in the way of Zen 3 and RX 6000 production, y'see. And in turn if Zen 3 is on either N7+ or N7P, with RX 6000 on the other, then arguably there's no manufacturing backlogs, other than TSMC producing enough wafers of each "type" to fulfil demand for the products on that "type".

Wouldnt bet against AMD respinning Xbox / Playstation stuff onto a lower node at some point, say in 2 years time, much like they did with Xbox one to Xbox One X etc, pretty sure Microsoft would go for that to prolong the life cycle of the consoles.

melmac · 6 Oct 2020 at 13:44

LePhuronn said:
So @melmac care to hazard a guess which processed AMD are using for all their 2020 products? Many people are suggesting AMD have supply issues because "everything is 7nm", but that's not technically the case. All we do know if AMD are not producing Zen 3 on N7.

For GPUs they have to be using 7nm+ The performance and power saving difference between N7 and N7P seems to be very small. Not sure how AMD could get their "up to 50% performance improvement" using N7P.

As to the supply issue, does it matter what process is been used and by whom? TSMC can only manufacture so much. Surely it also depends on how much production space AMD has ordered? And then how good yield are from that production run.

With all the stuff that's been released at the same time, will it surprise anyone if there are supply issues?

melmac · 6 Oct 2020 at 13:47

james.miller said:
N6 is EUV. it's closer to N7+ than it is to N7. What's debatable is is '6nm', because as far as i know nothing has shrunk except the space between the transistors. But, it's called N6 so there you go. I think it's probably not important in the grand scheme of things whether TSMC called it N6 or n7++ as quick porting from n7/n7+ to n6 is one of the advantages TMSC were touting - so it must be very similar anyway.

according to everything I have read that's not the case. You can't "port" from N7+ to N6 ( and indeed it seems you would be taking a step back if you did) But you can port easily from N7, N7P to N6.

But, I am happy to be corrected on this!!

Soapysage · 6 Oct 2020 at 14:15

This pretty much foreshadows what AMD has done with Big Navi doesn't it? Nintendo did it years ago with the Gamecube using 1T-SRAM, if correct, David Wang of AMD RTG had a hand in that tech with the Gamecube.

https://www.nintendoworldreport.com/editorial/516/gamecubes-memory-advantage

gpuerrilla · 6 Oct 2020 at 14:45

Just glad its sounding positive, a genuine contender to the 3080 is all we need to start changing the landscape.

LePhuronn · 6 Oct 2020 at 15:46

melmac said:
As to the supply issue, does it matter what process is been used and by whom? TSMC can only manufacture so much. Surely it also depends on how much production space AMD has ordered? And then how good yield are from that production run.

With all the stuff that's been released at the same time, will it surprise anyone if there are supply issues?

Well that's exactly what I'm saying. AMD have a little over 25% of TSMC's entire 7nm production capacity, but how that breaks down between the different processes we don't know. A lot of people are saying "well there will be supply issues and price gouging and nobody's getting Navi until next year by which time everybody has Ampere" like it's a guarantee, I'm merely offering the conversation point that it may not be the case because there are 4 different processes available to AMD to (potentially) pick and choose from. An example of which I gave earlier with the consoles being produced on N6 with Zen 3 and Big Navi on something else. Console supply would only be limited by TSMC's capacity to make N6 wafers, not sharing resources with Zen 3.