Question 'Ampere'/Next-gen gaming uarch speculation thread

[Ottonomous]

How much is the Samsung 7nm EUV process expected to provide in terms of gains?
How will the RTX components be scaled/developed?
Any major architectural enhancements expected?
Will VRAM be bumped to 16/12/12 for the top three?
Will there be further fragmentation in the lineup? (Keeping turing at cheaper prices, while offering 'beefed up RTX' options at the top?)
Will the top card be capable of >4K60, at least 90?
Would Nvidia ever consider an HBM implementation in the gaming lineup?
Will Nvidia introduce new proprietary technologies again?

Sorry if imprudent/uncalled for, just interested in the forum member's thoughts.

 

[Muhammed]

It says: Majority. And If I understand English language well enough, Majority means: "not all of" .

Don't dance around the subject, NVIDIA always had the majority of it's chips done at TSMC, with some low end parts done at Samsung, NVIDIA is continuing the trend in 2020.

I will repeat this. 7 nm Products from Nvidia we will see in 2020 is only HPC GA100 Chip. Gaming cards are 7 nm, and Late 2020/Early 2021 products. And Gaming cards are not Ampere architecture. it will be completely different name.

And I repeat again, you will be eating your words hard, You are so wrong on so many accounts, get ready for the storm.

 

[lobz]

Okay, that was interesting. I wonder how the Samsung order got so blown out of proportion.

As rumor has it, while on TMSC's HP optimized 7nm node variant (a bit less dense libraries) the defect density has improved a lot compared to the ultra-dense LP mobile variants, but not at Samsung, for whatever reason.

Mr. Leather Jacket is not someone that would want to get burned by not reacting in time to early bad news from the chosen process - like ever again.

 

[lobz]

Considering the only recent switch from Samsung to TSMC by Nvidia I would raise a question.

Have gaming cards even been taped out, yet?

I'd bet a lot of money on quite a few of NVIDIA designs having been taped out for quite some time.

 

[lobz]

The quoted articles say nothing about a 'switch'.

That's because I think, that 1: NVIDIA is not in a position where they should excuse themselves for anything they do or don't, so Jensen will not deliberately want to speak about any kind of hindrance, instead of saying on the launch stage 'the 7nm process presented a number of difficult challenges that our dedicated engineers have marvelously overcome' ))) and 2: even if he wanted to do that, I'm sure Samsung would not allow him to do that, if a change has taken place, it was well before it could have made serious damage, so also no reason to. Jensen may need Sammy in the future, he's burned enough bridges already, hasn't he?

 

[maddie]

@AtenRa

Page 6 - Question - 'Ampere'/Next-gen gaming uarch speculation thread

Page 6 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

That's a very old projection, mentioning GloFlo 7nm and Intels 10nm equivalent. Are you sure it's still accurate? I remember reading that ASML was falling behind in EUV shipments.

 

[Glo.]

Don't dance around the subject, NVIDIA always had the majority of it's chips done at TSMC, with some low end parts done at Samsung, NVIDIA is continuing the trend in 2020.


And I repeat again, you will be eating your words hard, You are so wrong on so many accounts, get ready for the storm.

The same way you were wrong on twitter saying that the GPU beating RTX 2080 Ti in VR benchmark was a laptop with Nvidia GPU?

Whole thread of your twitts when you are clearly wrong:

https://twitter.com/x/status/1215377031766183946

And this is Hassan's reply with a screenshot to your twitt:

https://twitter.com/x/status/1215561840304050176

Anyway.

Im not dancing about the subject. Nvidia went to Samsung FIRST to design GPUs on 7 nm process. Then it turned out to be a turd, so they went back to TSMC. But to not waste money on R&D, which was massive, they still use Samsung's process to some degree, depending on how the yield will be.

This year is only HPC for Nvidia. Which good thing. Volta needed replacement, considering that just around the corner are MI100 GPUs from AMD. Competition at its best.

 

[lobz]

Don't dance around the subject, NVIDIA always had the majority of it's chips done at TSMC, with some low end parts done at Samsung, NVIDIA is continuing the trend in 2020.


And I repeat again, you will be eating your words hard, You are so wrong on so many accounts, get ready for the storm.

Where the heck do you guys come here from? Every day more and more all of a sudden with this entitled bossy attitude?

 

[AtenRa]

@AtenRa

Page 6 - Question - 'Ampere'/Next-gen gaming uarch speculation thread

Page 6 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

That's a very old projection, mentioning GloFlo 7nm and Intels 10nm equivalent. Are you sure it's still accurate? I remember reading that ASML was falling behind in EUV shipments.

Well the absolute numbers may not directly reflect the current situation but im certain TSMCs 7nm volume will increase in 2020 and 2021.
7nm seems to be the next 28nm, it will remain in high volume production for many, many years.

 

[Muhammed]

The same way you were wrong on twitter saying that the GPU beating RTX 2080 Ti in VR benchmark was a laptop with Nvidia GPU?

WTF? Who is that? now you are truly gone off the rails!! who cares about a stupid VR benchmark?
Stick to the subject at hands, you are moving the goalposts, first you say NVIDIA is using Samsung for all chips, but when you are proven wrong you dance around the subject claiming only one chip is for 2020 release, clearly you know nothing and are just shooting in the dark.

I can't count the so many times you were wrong about Vega and it's drivers and the Vega 64 suddenly beating 1080Ti due to magical fine wine, do I need to go back and fish around?

It's quite amusing, but also tiresome.

 

[Mopetar]

Well the absolute numbers may not directly reflect the current situation but im certain TSMCs 7nm volume will increase in 2020 and 2021.
7nm seems to be the next 28nm, it will remain in high volume production for many, many years.

28nm only stuck around as long as it did because the 20nm process from TSMC had a lot of issues. We've already heard that Apple is moving to their 5nm process so it's hard to say we're in a similar situation. I think the 7nm process is pretty good and will enjoy a long life for that reason, but the situation is a lot different from 28nm where companies stayed because the next node wasn't worth using.

 

[Glo.]

WTF? Who is that? now you are truly gone off the rails!! who cares about a stupid VR benchmark?
Stick to the subject at hands, you are moving the goalposts, first you say NVIDIA is using Samsung for all chips, but when you are proven wrong you dance around the subject claiming only one chip is for 2020 release, clearly you know nothing and are just shooting in the dark.

I can't count the so many times you were wrong about Vega and it's drivers and the Vega 64 suddenly beating 1080Ti due to magical fine wine, do I need to go back and fish around?

It's quite amusing, but also tiresome.

Show my any post where I said that Nvidia is using ONLY Samsung. And show my post where I said this plainly, not the way you understood it.

You haven't proven anything, yet.

 

[Krteq]

Don't feed a troll

 

[beginner99]

Gosh AMD and NV PR guys, go get a beer together in a bar so we can enjoy the forums again. Thank you.

 

[railven]

Gosh AMD and NV PR guys, go get a beer together in a bar so we can enjoy the forums again. Thank you.

Amen! If only we had a group of users with elevated privileges that could do something, but gosh that be asking too much.

 

[lobz]

Gosh AMD and NV PR guys, go get a beer together in a bar so we can enjoy the forums again. Thank you.

Don't dance around the subject you've just changed!!!!!! Are you calling me an AMD PR guy? OR GUYS?
Joking aside, rather than throwing around this unweighted and generalizing condemnation, it might be best to at least specify those people who brought this out of you recently. I, for example, only saw one guy in the last couple of pages of this thread, who's posting in a way that makes the forum sound hostile and unenjoyable. @Glo. 's still keeping quite calm despite the tone addressed towards him multiple times. This time I like @Krteq 's advice more, not to feed the troll It's good to be reminded sometimes

 

[Mopetar]

I almost wonder if we don’t see NVidia do a die shrink of the 16xx series parts first since just doing that is probably more than good enough to compete with the current AMD products.

The low end of the market isn’t going to need or get RT anytime soon and NVidia really needs to focus their efforts on their RT tech for the next generation if they want it to be more than just a gimmick.

 

[Glo.]

I almost wonder if we don’t see NVidia do a die shrink of the 16xx series parts first since just doing that is probably more than good enough to compete with the current AMD products.

The low end of the market isn’t going to need or get RT anytime soon and NVidia really needs to focus their efforts on their RT tech for the next generation if they want it to be more than just a gimmick.

That is assuming the power qualities of TSMC's 7 nm process are as good as they advertised. Which is not true.

Secondly, I don't want Nvidia to just shrink the dies. Bring something new, better than we already have. Don't make Pascal v2 case again.

 

[jpiniero]

The low end of the market isn’t going to need or get RT anytime soon and NVidia really needs to focus their efforts on their RT tech for the next generation if they want it to be more than just a gimmick.

The console devs will pretty much determine how quickly RT gets adopted. All nVidia can do is just push them in that direction.

 

[Mopetar]

The console devs will pretty much determine how quickly RT gets adopted. All nVidia can do is just push them in that direction.

Console RT isn’t going to be any better than 2070 RT unless AMD has really worked some magic, which I’ll doubt until proven wrong with actual hardware.

Console makers always try pushing new stuff. How well did motion controls or VR ever take off as a result of consoles pushing those technologies? The answer is that they now feel like a gimmick that couldn’t live up to initial hype.

The RT that everyone wants is still probably 10 years off, particularly at mainstream prices. Currently it’s possible to sprinkle in at a high performance cost. The monetary cost is even more substantial.

 

[jpiniero]

Console RT isn’t going to be any better than 2070 RT unless AMD has really worked some magic, which I’ll doubt until proven wrong with actual hardware.

The difference is much more that they can design the game around using RT, optimize it, use tricks to maximize the effect. Plus 30 fps is quite OK in console land.

For a PS5/XBS designed game, they don't have to worry about not having HW RT unless they are also targeting the earlier gen consoles.

 

[Stuka87]

Console RT isn’t going to be any better than 2070 RT unless AMD has really worked some magic, which I’ll doubt until proven wrong with actual hardware.

Console makers always try pushing new stuff. How well did motion controls or VR ever take off as a result of consoles pushing those technologies? The answer is that they now feel like a gimmick that couldn’t live up to initial hype.

The RT that everyone wants is still probably 10 years off, particularly at mainstream prices. Currently it’s possible to sprinkle in at a high performance cost. The monetary cost is even more substantial.

AMD's implementation of RT is quite different from nVidia's, so we really don't know how it will perform. But each maker went about it in a different way.

And I agree that RT outside of some reflections and shadows is a long ways off.

 

[soresu]

The RT that everyone wants is still probably 10 years off, particularly at mainstream prices. Currently it’s possible to sprinkle in at a high performance cost. The monetary cost is even more substantial.

With a liberal use of denoising too, given even hybrid RT probably only uses very few samples per pixel for performance sake.

 

[Ajay]

AMD's implementation of RT is quite different from nVidia's, so we really don't know how it will perform. But each maker went about it in a different way.

And I agree that RT outside of some reflections and shadows is a long ways off.

How do we know AMD's implementation is so different than Nvidia's?

 

[Mopetar]

The difference is much more that they can design the game around using RT, optimize it, use tricks to maximize the effect. Plus 30 fps is quite OK in console land.

For a PS5/XBS designed game, they don't have to worry about not having HW RT unless they are also targeting the earlier gen consoles.

A lot of PC games are essentially console ports so unless they want to limit sales to a fraction of the market and have a game that runs like a slideshow on anything but the most high-end expensive GPUs, then they're going to need to write some fallback code anyway or simply just allow users to disable those effects. Even the consoles themselves are going to be somewhat limited by whichever system is the least powerful. Why put in extra RT on the Xbox/PS5 if it's just going to cause the game to run horribly on the other system that doesn't have as much capability in that area?

I suspect that there's also a bit of a learning curve in being able to effectively utilize RT in games because of the performance implications. It might be a few years before we start to see some of the first or second party exclusives start to figure that out, and they have the advantage of not having to worry about a PC port or getting the game to run on the other console. But even though those titles will be able to showcase the goodness of RT, they're also the titles that are exclusive to a console and don't get as much mass-market exposure.

 

[Stuka87]

How do we know AMD's implementation is so different than Nvidia's?

Parts of it have been in drivers since this last summer. There was a thread on here that talked about it. And we still do not know about anything in depth, only that its not the same as nVidia's system. But, so long as both are compatible with Microsoft's DXR library, it should not matter to end users. Only that they turn it on, and it works.