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

Page 118 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Ottonomous

Senior member
May 15, 2014
559
292
136
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.
 

TESKATLIPOKA

Platinum Member
May 1, 2020
2,362
2,854
106
If it was SS7, I would expect it the die to be smaller and draw much less power. Almost has to be SS8 (or a custom version) at max density possible.
Maybe smaller die size, but power draw can be this high, If the clockspeed and with It the voltage is higher than what 2080Ti had, although the clockspeed looks the same to me.
 

beginner99

Diamond Member
Jun 2, 2009
5,210
1,580
136
I still believe they used the same 7nm at TSMC as with A100

If it actually is 7nm then yes, they have to because else Mr. Leatherjacket saying "most of 7nm will be from TSMC" will clearly be wrong if they use Samsung 7nm for GA102 and lower. SO eithers it's TSMC 7nm or Samsung 8nm (and hence not 7nm products)
 

A///

Diamond Member
Feb 24, 2017
4,352
3,154
136
Samsung could rename a specific line tuned for NVidia and call it 7nm -whatever-. It's not like these names mean anything nowadays. BTW, this isn't my theory. Just grabbed it from Twitter and it sort of made sense. IDC either way. I'm going solid mid range because I'd rather spend the money on a processor and mobo.
 
Last edited:
  • Like
Reactions: spursindonesia

blckgrffn

Diamond Member
May 1, 2003
9,128
3,069
136
www.teamjuchems.com
Yeah, I mean we can’t forget how BTC and ETH mining thoroughly disrupted the pricing landscape for video cards for several years, arguably all the way back the 5870 AMD cards.

What’s interesting now that I ponder it a bit is how that “value” still seems baked into these newer cards, despite mining any worthwhile crypto is extremely unprofitable in GPU at this point unless you have free power.
 

lobz

Platinum Member
Feb 10, 2017
2,057
2,856
136
Samsung could rename a specific line tuned for NVidia and call it 7nm -whatever-. It's not like these names mean anything nowadays. BTW, this isn't my theory. Just grabbed it from Twitter and it sort of made sense. IDC either way. I'm going solid mid range because I'd rather spend the money on a processor and mobo.
But then Jensen will have lied about TSMC getting the vast majority of their 7nm GPU orders.
 

A///

Diamond Member
Feb 24, 2017
4,352
3,154
136
But then Jensen will have lied about TSMC getting the vast majority of their 7nm GPU orders.
Is it if half their sales end up being DC/scientific parts like the A100 build on TSMC 7nm? Majority is such a vague word because you can twist it to mean 70% or greater or simply 52% over 48%.
 

uzzi38

Platinum Member
Oct 16, 2019
2,635
5,984
146
Weird theory, what if the product stack is split between 7LPP and 8LPP? And to go into a bit more detail, GA102 on 7LPP with everything else 8LPP?

That should still fulfil Jensen's statement, as minimal volume would be on 7LPP compared to N7.

Have been thinking about this the last couple of days, and suddenly this report came to mind:


Which should explain why the Exynos 9825 and 9820 were nearly identical in perf. Because they weren't particularly truthful about properly using 7LPP in the first place. Using 7LPP should show an actual performance improvement over what the 9825 showed.

Just a theory. I'm looking forwards to whatever the truth is in a couple of days time.
 

jpiniero

Lifer
Oct 1, 2010
14,622
5,237
136
Weird theory, what if the product stack is split between 7LPP and 8LPP? And to go into a bit more detail, GA102 on 7LPP with everything else 8LPP?

I was trying to talk myself into them using the SS7 version of GA102 (or even GA101!) for just Quadros, but that seems a stretch.
 

jpiniero

Lifer
Oct 1, 2010
14,622
5,237
136

500W.

Five Hundred whatts.

No way in hell this is only 1700 MHz, it clocks way past that under load, with those clock speeds.

Remember that the actual frequency has been typically decently higher than the advertised max speed in games. 500 sounds like nVidia is serious about not letting anyone know the real performance and is running the card at max power consumption. 3x8-pin max is 525.
 

Glo.

Diamond Member
Apr 25, 2015
5,711
4,559
136
Well, some AIB of RTX 2080 Ti already were using 300W of power during gaming.

500W air-cooled? No way.
Yep, that also did occur to me, that 500W versions HAVE TO be water cooler.
 

mohit9206

Golden Member
Jul 2, 2013
1,381
511
136
Samsung could rename a specific line tuned for NVidia and call it 7nm -whatever-. It's not like these names mean anything nowadays. BTW, this isn't my theory. Just grabbed it from Twitter and it sort of made sense. IDC either way. I'm going solid mid range because I'd rather spend the money on a processor and mobo.
Why not call it 5nm or 3nm if they can call it anything to market it better?
 

uzzi38

Platinum Member
Oct 16, 2019
2,635
5,984
146
Didnt we learn nothing from the last twitter leaker? 350W is the official number even leaked by Gainward.
That was specific to clocks under official specifications.

Many AIB cards come with factory overclocks which are NOT covered by the 350W TGP listed by Gainward.
 

Krteq

Senior member
May 22, 2015
991
671
136
What about Zotac?
vickq87yu53d36hyuqkoq.jpg
 

sontin

Diamond Member
Sep 12, 2011
3,273
149
106
There will be so many versions of AIB cards. My Gigabyte 2080TI Gaming OC has a TDP of 300W with 360W powerlimit. That is 50W/40W more than the reference TDP.