rumors Multiple internal sources confirm that all initial consumer Ampere products will be on Samsung 8nm

[Results45]

So......right off the bat Tom divulges that two of his 'Nvidia insiders' tentatively verified that all production of "GA102 and higher" (engineering samples aside) leading up to the launch of consumer Ampere cards this fall will on Samsung's 8nm process. This falls on the heels of Intel's troubles (and Nvidia earlier this year) securing capacity at scale with TSMC as Apple, AMD, and Qualcomm have snatched up the vast majority of it.

 

 

This seemingly includes the RTX 3080 (GA102), 3070 (GA103), 3060 (GA104), 3050 (GA106), and 3040 (GA107:

Spoiler

.

.

 

 

Maybe we'll see a "Super" refresh on TSMC next year. At least, one can hope.

[The_russian]

What are the multiple sources? I’m not watching the video but I hope the only source isn’t that guy and his secret “informers”.

[Mateyyy]

Meh, rumours rumours rumours. Please correct me if I'm wrong, but from what I've seen that guy's whole channel is based on rumours and random assumptions. I'm not trying to hate or anything, but I don't feel like what he dreamt last night or what his mate emailed him is newsworthy.

 

Regardless, even if consumer-grade Ampere was on 8nm, that doesn't mean anything. Turing at 12nm beats Navi/RDNA at 7nm in both performance and efficiency, for instance. Heck, Pascal on 16nm is about on par with Navi (a 1080Ti is quite similar to the 5700XT watt for watt).

Let's just wait for real performance metrics to tell us what we actually want to know - who the performance king is this generation.

[Korben]

Who cares? 7 or 8nm? You wouldn't see the difference anyway!

[Results45]

25 minutes ago, Mateyyy said:

Meh, rumours rumours rumours, please correct me if I'm wrong, but from what I've seen that guy's whole channel is based on rumours and random assumptions. I'm not trying to hate or anything, but I don't feel like what he dreamt last night is newsworthy.

 

Well if you skim through all his videos, most of his Intel leaks over the past year were spot-on. He also apparently corroborates his info with AdoredTV and Coreteks as well.

 

The YouTube tech leaking community is still pretty small, but I give them as much benefit-of-the-doubt as the following sites:

• Notebookcheck
• Tech Power Up
• PC Gamer
• PC Perspective
• PCGamesN
• Tom's Guide
• ArsTechnica

 

Plus, he was leaking and discussing this stuff in videos 2-4 months before other mainstream tech sites started picking up on it.

 

Go watch his videos and podcasts. He's not just some random guy "confirming" stuff he has no clue about.

[Lakobrija]

3 minutes ago, Korben said:

Who cares? 7 or 8nm? You wouldn't see the difference anyway!

I think there is some real difference in density. 7nm is a lot better according to some image I found.

101 MTr/mm2 is tsmc 7nm

61 MTr/mm2 is Samsung 8nm

[Results45]

.

2 minutes ago, Korben said:

Who cares? 7 or 8nm? You wouldn't see the difference anyway!

 

It's all about scaling, efficiency, and yields: https://semiengineering.com/5-3nm-wars-begin/

 

Which is why everyone wants a piece of TSMC before Samsung or Global Foundries despite the latter two offering cheaper prices.

[Mateyyy]

8 minutes ago, Results45 said:

Well if you skim through all his videos, most of his Intel leaks over the past year were spot-on. He also apparently corroborates his info with AdoredTV and Coreteks as well.

 

The YouTube tech leaking community is still pretty small, but I give them as much benefit-of-the-doubt as the following sites:

• Notebookcheck
• Tech Power Up
• PC Gamer
• PC Perspective
• PCGamesN
• Tom's Guide
• ArsTechnica

Just to play devil's advocate here, Comet Lake-S has been leaked for over a year, and predicting what Intel's going to come out with hasn't been too challenging lately, considering the fact that their die shrinking troubles are public knowledge.

[porina]

3 minutes ago, Korben said:

Who cares? 7 or 8nm? You wouldn't see the difference anyway!

Haven't we learnt that what one company calls x nm isn't necessarily the same as what another company calls y nm? 

 

I do wonder, even if roughly, where Samsung 8nm fits in relative to TSMC 7nm, Intel 10nm, or whatever else is around. In a bit of searching so far, Samsung 8nm is an upgrade to their 10nm process, before making a bigger jump to Samsung's 7nm process. News about 8nm first appeared around 2017. TSMC 12nm as used for Turing was apparently a refinement of their 16nm process. This also dates around 2017. It is hard to compare these different processes as they vary in different parameters. The only common one I've found between them, shows a big improvement going from TSMC 12nm to Samsung 8nm (about a node worth), Samsung 8nm to either TSMC 7nm or Intel 10nm is about a half node difference. Again, this is only on one parameter of something much more complex than a single number, so performance delta on average may be different.

[Korben]

And I insist: you wouldn't SEE the difference!

Don't take this stuff so serious... it's just rumours, speculations, nonsense whatsoever... chill a bit.

[Maticks]

i am a bit confused with Ampere... Nvidia was going on about 25% better power efficiency and shrinking the die size.

So shouldn't they be going from 8pin power connectors to 6 or 4 pin. Moving to 12 pin plugs that can likely deliver 600Watts doesn't sound like power efficiency?

 

If you are chasing lower power usage why are we going the other side of the scale here?

[Brooksie359]

53 minutes ago, Maticks said:

i am a bit confused with Ampere... Nvidia was going on about 25% better power efficiency and shrinking the die size.

So shouldn't they be going from 8pin power connectors to 6 or 4 pin. Moving to 12 pin plugs that can likely deliver 600Watts doesn't sound like power efficiency?

 

If you are chasing lower power usage why are we going the other side of the scale here?

Power efficiency is not the same as power draw. If that were the case then an old 75 watt gpu would be the same power efficiency as a new 75 watt gpu. 

[pas008]

1 hour ago, Maticks said:

i am a bit confused with Ampere... Nvidia was going on about 25% better power efficiency and shrinking the die size.

So shouldn't they be going from 8pin power connectors to 6 or 4 pin. Moving to 12 pin plugs that can likely deliver 600Watts doesn't sound like power efficiency?

 

If you are chasing lower power usage why are we going the other side of the scale here?

I'm thinking those 600w 12pin is going to be for those hpc and very high end hopper MCM design cards

Probably be killers but spendy

 

[rogerwilco91]

That’s fine with me, as long as RTX performance doesn’t take a significant hit at 1440p or 4K, I am ready for Ampere’s launch. Sept can’t come soon enough!

[Arika]

it was inevitable that TSMC was going to run out of capacity. so this isn't that surprising.

[pas008]

29 minutes ago, Arika S said:

it was inevitable that TSMC was going to run out of capacity. so this isn't that surprising.

Believe this was planned from the beginning considering samsung supposedly under cut tsmc

But i guess 8nm though which is most likely fine

 

https://www.tomshardware.com/news/samsung-nvidia-7nm-ampere-tsmc,39583.html

 

Think there was news topics here also if memory serves me correct

[GDRRiley]

19 hours ago, Korben said:

Who cares? 7 or 8nm? You wouldn't see the difference anyway!

it does matter. Samsungs 8nm is way behind TSMC 7nm

[5x5]

21 hours ago, Korben said:

And I insist: you wouldn't SEE the difference!

Don't take this stuff so serious... it's just rumours, speculations, nonsense whatsoever... chill a bit.

You WILL see the difference in heat and power. It's one thing for your card to be using 250W, it's a whole different thing I'd it's at 350W

[Master Disaster]

22 hours ago, Korben said:

And I insist: you wouldn't SEE the difference!

Don't take this stuff so serious... it's just rumours, speculations, nonsense whatsoever... chill a bit.

Total role reversal...

 

The German person makes the joke and everyone else totally misses it

 

I got it at least.

[Results45]

10 hours ago, GDRRiley said:

it does matter. Samsungs 8nm is way behind TSMC 7nm

 

Yeah, it's a slightly improved version of their 10nm. Basically equivalent to Intel's 14+++ process: https://fuse.wikichip.org/news/1443/vlsi-2018-samsungs-8nm-8lpp-a-10nm-extension/

 

 

 

 

Thing is, I don't get we haven't heard anything about Nvidia and Intel buying up Samsung 7nm/5nm capacity. Cuz they've been building out capacity for the past year: https://www.anandtech.com/show/15538/samsung-starts-mass-production-at-v1-a-dedicated-euv-fab-for-7nm-6nm-5nm-4nm-3nm-nodes

[leadeater]

12 hours ago, Results45 said:

Thing is, I don't get we haven't heard anything about Nvidia and Intel buying up Samsung 7nm/5nm capacity. Cuz they've been building out capacity for the past year: https://www.anandtech.com/show/15538/samsung-starts-mass-production-at-v1-a-dedicated-euv-fab-for-7nm-6nm-5nm-4nm-3nm-nodes

Probably all allocated to NAND, Memory and Mobile SoCs.

[justpoet]

17 minutes ago, leadeater said:

Probably all allocated to NAND, Memory and Mobile SoCs.

Samsung has been pushing Exynos for not just mobile CPU, including on device AI cores similar to what Apple has done for a while, but also as a 5g modem platform.  Those are already shipping on their 7nm.  Of course there's various memory that they've announced using EUV for now too, but what most people don't see is that they're also in the automotive industry with a different Exynos line that has been on 8nm as well, and not just for infotainment or self driving, but for things like your normal dashboard cluster readouts…and the most often forgotten chipset in computing circles is optics/sensors.

 

So I would expect their next node production lines to initially be fully used by those items as they ramp up their next generation of stuff.

[Guest]

Still going to knock the park out of what AMD can offer, so why not use the cheaper fab?

[Spindel]

Mehh thought this was about Ampere computings ARM CPUs, not some boring graphics card

 

EDIT:// I think it’s time to add a rule to this forum that you ned to have Nvidia ampere in the headline for Nvidia related stuff (and Ampere computing for the ARM stuff, if anyone posts anything about that)

[mr moose]

Typical, someone makes a relatively amusing joke about node size and half the forum trips over themselves to correct it.

 

 

@Korben I "SEE" what you did there.