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.

 

[lobz]

What are they going to do, broadcast it from Jensen's bedroom?

Nope, no virtual broadcast at all. Because of a non-virtual disease. That's what I need help understanding.

 

[Stuka87]

Add in that AMD relies more heavily on Global foundaries compared to Nvidia on Samsung and Nvidia could be a bigger spender at TSMC.

This is not currently true. All of AMD's current products are on 7nm, which makes them TSMC only. Older GPU's like Polaris, which are still in production, are also TSMC.

There are some older embedded CPU products that are still manufactured at GF, but these are a small percentage of AMD's overall product line.

 

[mopardude87]

In case anyone is thinking of buying just cause, well here ya go. Nothing but gospel truth out of Tech Jesus. This needs to be heard and seen EVERYWHERE. Gonna hike up the prices so damn much i will be forced into my secondary hobby till this thing clears over. People need to relax seriously before they ruin it for EVERYONE.

 

[Glo.]

Are you sure of that?

Yes. I am sure of what I have written.

 

[Veradun]

Nope, no virtual broadcast at all. Because of a non-virtual disease. That's what I need help understanding.

I was wondering as well, then someone pointed me in the right direction.

To brew a quality webcast you need a lot of people on site and that is the concern. It's then safer to go "take these slides and pr statement".

Not that keynotes are any different than that anyway, they're just cozyer

 

[joesiv]

In other news:
"(...)in light of the spread of the coronavirus, it is (NVIDIA) deferring plans to deliver a webcast keynote(...)"

Can someone please explain to me, what in the frozen hell does a biological virus have to do with a webcast?

It's an interesting question. Perhaps Jenson was going to drop a bunch of product announcements, and release dates, but with virus, there are going to be delays. I wouldn't want to make a good deal of a new fancy smanshy top end GPU and then say that it might be available in a couple months, but knowing that it's unknown how much impact the virus will have on supply chain and dates in general.

With regards to NVIDIA being a 3rd tier, I have no clue, but I was under the understanding from comments from AMD, that next gen process technologies are a collaboration. AMD has invested in the 7nm research with TSMC, that with a bunch of cash (obviously), would get them higher in the queue. Those that invest less, or none, have to pick up the remaining of allocations at auction. Perhaps Nvidia didn't invest early in 7nm, so TSMC has less obligation to them when it comes to allocations.

 

[tajoh111]

This is not currently true. All of AMD's current products are on 7nm, which makes them TSMC only. Older GPU's like Polaris, which are still in production, are also TSMC.

There are some older embedded CPU products that are still manufactured at GF, but these are a small percentage of AMD's overall product line.

The central IO chip of every Zen 2 chip is still made at Global foundaries.

And at the end of January this year, just half of AMD revenue came from 7nm products.

 

[lobz]

I was wondering as well, then someone pointed me in the right direction.

To brew a quality webcast you need a lot of people on site and that is the concern. It's then safer to go "take these slides and pr statement".

Not that keynotes are any different than that anyway, they're just cozyer

Thanks for some sensible food for thought

 

[Stuka87]

The central IO chip of every Zen 2 chip is still made at Global foundaries.

And at the end of January this year, just half of AMD revenue came from 7nm products.

AMD doesn't report monthly sales, so I am going to assume you are referencing their 2019 financials report where Lisa Su stated "half of the chipmaker's revenue now comes from 7nm products."

And it would make sense for this to show half of their revenue was from 7nm products, as the first half of the year had basically no 7nm products outside of Vega20. And chips like those found in the XBox and PS4 are not 7nm. And some of the 7nm GPU's launched late in the year.

 

[tajoh111]

AMD doesn't report monthly sales, so I am going to assume you are referencing their 2019 financials report where Lisa Su stated "half of the chipmaker's revenue now comes from 7nm products."

And it would make sense for this to show half of their revenue was from 7nm products, as the first half of the year had basically no 7nm products outside of Vega20. And chips like those found in the XBox and PS4 are not 7nm. And some of the 7nm GPU's launched late in the year.

4th quarter 2019. Which was just a couple months ago and their most recent report.

AMD's 4Q/FY19 Financial Report: Revenue up 50% from 4Q18, Debt Halved

www.anandtech.com

ryzen 2 + Navi were already released.

 

[uzzi38]

GA102 just taped out on Samsung's 8LPP. NG Tegra is on the same node.

Source is Kimi.

 

[lobz]

GA102 just taped out on Samsung's 8LPP. NG Tegra is on the same node.

Source is Kimi.

While I have zero actual evidence of the contrary, I must say, all this sounds incredibly unlikely.

 

[Glo.]

So high-end Nvidia GPUs are due in 9 months at best. CES 2021 release?

 

[uzzi38]

So high-end Nvidia GPUs are due in 9 months at best. CES 2021 release?

I think they'll squeeze for holiday period this year. Also lets them show off 'leadership' graphics performance vs the also-releasing-then-consoles

 

[mopardude87]

I want this Ampere but with my new 1080p monitor and no games of interest besides maybe Warzone, i doubt i will have a reason to swap my 1080ti. Losing interest in the gaming hobby anyways, maybe its time to work on the old mopars a bit more and let this hobby simmer? Been so many disappoints i have lost the will to invest in it.

Shame too cause Ampere could be epic, perhaps a upgrade will thrill me again? I got restraint luckily this time about,i am willing to be sensible. Went back to 1080p cause i don;t feel like investing in thousands for 4k. Not caring that much even about being back on a freesync 1080p panel. Its good enough as is the 1080ti for now.

 

[jpiniero]

GA102 just taped out on Samsung's 8LPP. NG Tegra is on the same node.

Source is Kimi.

I have serious doubts about that one. You would think after being burnt once already they would go with TSMC rather than port to another Samsung node. Plus I have to think it would be way more work to go from an EUV node to a non-EUV node.

 

[Veradun]

I have serious doubts about that one. You would think after being burnt once already they would go with TSMC rather than port to another Samsung node. Plus I have to think it would be way more work to go from an EUV node to a non-EUV node.

To move production you need open slots tho and I highly doubt TSMC have that many in the next six to twelve months

 

[jpiniero]

To move production you need open slots tho and I highly doubt TSMC have that many in the next six to twelve months

nVidia would have had to made the decision to move away from Samsung's 7 a long time ago. So that shouldn't have been an issue. If they do end up sticking with Samsung for the most part it would be for other reasons than capacity issues at TSMC.

 

[Stuka87]

nVidia would have had to made the decision to move away from Samsung's 7 a long time ago. So that shouldn't have been an issue. If they do end up sticking with Samsung for the most part it would be for other reasons than capacity issues at TSMC.

I would not be surprised if Samsung gave them a killer deal. TSMC isn't giving many deals, because there is huge demand.

 

[Bouowmx]

Samsung 8 nm (based on 10 nm): https://fuse.wikichip.org/news/1443/vlsi-2018-samsungs-8nm-8lpp-a-10nm-extension/

Transistor density is 51.8 M/mm^2 with high density (HD), and 61.1 with ultra high density (UHD).

With 7/8 of HD density (0.875 * 51.8 = 45.3 M/mm^2), an A102 chip with 1.5x transistors of TU102 (1.5 * 18600 M = 27900 M) would have area of 615 mm^2.

Like a distant memory, do recall that Samsung 10 nm was used for Qualcomm Centriq (ARM server CPU). It had 18000 M transistors in 398 mm^2.

 

[Krteq]

GA102 just taped out on Samsung's 8LPP. NG Tegra is on the same node.

Source is Kimi.

Wait a second. Wasn't xpea trying to told us that (according to his "sauce") GA102 has been taped-out on TSMC's 7nm EUV in the middle of last year

Samsung's 10nm is a BS IMO. Nothing in that tweet makes sense.

 

[Richie Rich]

Wait a second. Wasn't xpea trying to told us that (according to his "sauce") GA102 has been taped-out on TSMC's 7nm EUV in the middle of last year

What's so funny? We know Zen3 was taped out at N7+ or N7P in the middle 2019 too.

 

[Glo.]

Wait a second. Wasn't xpea trying to told us that (according to his "sauce") GA102 has been taped-out on TSMC's 7nm EUV in the middle of last year

Samsung's 10nm is a BS IMO. Nothing in that tweet makes sense.

Nothing in the rumors about next gen Nvidia GPUs, apart from GA100 chip, makes sense. Every rumor contradicts previous one.

 

[Krteq]

Exactly

What's so funny? We know Zen3 was taped out at N7+ or N7P in the middle 2019 too.

Well, in case of Zen 3... that information about tapeou on improved TSMC 7nm process was from official source - AMD itself.

In case of GAxxx... just a bunch of random contradictory forum/twitter posts.

 

[happy medium]

https://twitter.com/x/status/1237748855426310144

I dont think this is official data.