Discussion Ada/'Lovelace'? Next gen Nvidia gaming architecture speculation

Page 3 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

Mopetar

Diamond Member
Jan 31, 2011
7,797
5,899
136
Going wider probably wouldn't do much for making Ampere better. If DLSS takes off or improves, it actually means that cards should reduce shader count because what's the point of trying to render at 4K if you can just up-sample to 4K and get far better FPS. Throw out shaders and pack in more RT hardware instead.
 
  • Like
Reactions: GodisanAtheist

GodisanAtheist

Diamond Member
Nov 16, 2006
6,719
7,016
136
Going wider probably wouldn't do much for making Ampere better. If DLSS takes off or improves, it actually means that cards should reduce shader count because what's the point of trying to render at 4K if you can just up-sample to 4K and get far better FPS. Throw out shaders and pack in more RT hardware instead.

- DLSS is a really killer feature for mobile or portable devices if it allows less hardware (lower power) to do the same or more work. Smaller screen on the portable side of the console will help diminish any impact in visual quality in a "performance DLSS" mode, while the docked higher power mode can fall back to a "quality DLSS" setting.
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
Ampere+ is exactly what it looks like. Theoretical SS 5 density is almost double that of SS8 however so they could easily just jack up the SM counts by 25% or more.
I had read that SS N5 was having problems* ramping up to HVM, has that changed?

*Not sure what the cause was.
 

fleshconsumed

Diamond Member
Feb 21, 2002
6,483
2,352
136
The two questions are, will it be bottlenecked by CPU just as badly as 30 series, and will we be able to buy it at all?
 

Mopetar

Diamond Member
Jan 31, 2011
7,797
5,899
136
The two questions are, will it be bottlenecked by CPU just as badly as 30 series, and will we be able to buy it at all?

Unless they go back to using a hardware scheduler, it's going to result in additional CPU overhead that naturally means it will bottleneck at the CPU sooner than it otherwise would. All that it really means is that the point will move a bit down the stack towards mid-range cards when gaming at 1080p using an older CPU.

The second is anyone's guess because it really depends on what the crypto market looks like. If nothing changes then the answer is effectively no unless you're willing to camp out, set up a bot of your own, or just pay the scalper's price. Alternatively the market could tank and the real question will be whether to buy Lovelace or to scoop up a cheap used Ampere card instead because there will be a massive flood of cards hitting the market as soon as mining isn't profitable.
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
The two questions are, will it be bottlenecked by CPU just as badly as 30 series, and will we be able to buy it at all?
I haven't followed the recent bottleneck issue, though recent driver versions are causing framerate drops in one of my games.
Whether or not we'll be able to buy rests mostly on the price of ethereum. Though, arguably, if NV keeps the same shader counts, die size would go down and dice per wafer would increase leading to more supply - so long as yields don't tank.
 

moonbogg

Lifer
Jan 8, 2011
10,635
3,095
136
Lovelace should make a fantastic mining product. I'm sure these cards are already spoken for by huge mining firms. Just saying.
 

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
Lovelace should make a fantastic mining product. I'm sure these cards are already spoken for by huge mining firms. Just saying.
I just hope that NV and AMD do build into hardware something that kills hash rates on Gaming Cards.
 

Magic Carpet

Diamond Member
Oct 2, 2011
3,477
231
106
- DLSS is a really killer feature for mobile or portable devices if it allows less hardware (lower power) to do the same or more work.
Can’t wait for the new Switch, personally.

The new Switch iteration will support Nvidia’s Deep Learning Super Sampling, or DLSS, a novel rendering technology that uses artificial intelligence to deliver higher-fidelity graphics more efficiently. That will allow the console, which is also set for an OLED display upgrade, to reproduce game visuals at 4K quality when plugged into a TV, said the people, who asked not to be identified because the plan is not public.
 

jpiniero

Lifer
Oct 1, 2010
14,511
5,159
136
Was thinking... if AMD is really doing 5 nm+chiplets, Lovelace isn't going to cut it.
 

maddie

Diamond Member
Jul 18, 2010
4,723
4,628
136
I dont think the workload matters so much - the card is rated at 350W max, and I created a workload to push everything to 100%. I reached a max power draw of 350W, and saw the ratio between GPU and RAM powerdraw when operating at max, and clearly the RAM has capacity to draw considerably more juice than the GPU when both are flat-out. (at stock).

I cant imagine there would be a scanario where I would get higher powerdraw on either the GPU or RAM by reducing the powerdraw of the other component(s) so I think my test is reasonably reliable.
One extremely easy way to reality check is to look at the cooling. If the ram uses as much power as you believe [more than the GPU], then the heatsink would be designed for this. Not even important to know the actual values.
 
  • Like
Reactions: coercitiv

Timmah!

Golden Member
Jul 24, 2010
1,396
604
136
Was thinking... if AMD is really doing 5 nm+chiplets, Lovelace isn't going to cut it.

I saw some utube videos and read some speculation articles about dual-tile 7900xt, apparently 2,5 faster than current high-end....some hellbent Lovelace with it still single monolithic chip, or whatever Nvidia will compete with, wont stand a chance....
I wonder if Nvidia would really let that happen.
 

dr1337

Senior member
May 25, 2020
311
516
106
I saw some utube videos and read some speculation articles about dual-tile 7900xt, apparently 2,5 faster than current high-end....some hellbent Lovelace with it still single monolithic chip, or whatever Nvidia will compete with, wont stand a chance....
I wonder if Nvidia would really let that happen.
Well a giant monolithic GPU doesn't have to solve all of the scheduling and latency issues from an MCM design. I don't think its likely for AMD to release a bad GPU but its certainly possible for lovelace to tango against only 10,000 shaders. If nvidia can increase utilization of its ALUs in lovelace, with 18,000 shaders, it definitely wont be a walk in the park for AMD. But on the other hand if lovelace scales the same way ampere does, there is a chance the giant monolithic die could disappoint I think.
 

maddie

Diamond Member
Jul 18, 2010
4,723
4,628
136
I saw some utube videos and read some speculation articles about dual-tile 7900xt, apparently 2,5 faster than current high-end....some hellbent Lovelace with it still single monolithic chip, or whatever Nvidia will compete with, wont stand a chance....
I wonder if Nvidia would really let that happen.
Seeing that designs are finalized years in advance, what exactly can they do? Pricing is the only available variable in the end.
 

jpiniero

Lifer
Oct 1, 2010
14,511
5,159
136
Seeing that designs are finalized years in advance, what exactly can they do? Pricing is the only available variable in the end.

Surely the gaming product after Lovelace is chiplet based. Maybe they will just cancel Lovelace and pull that in?
 

Timmah!

Golden Member
Jul 24, 2010
1,396
604
136
Surely the gaming product after Lovelace is chiplet based. Maybe they will just cancel Lovelace and pull that in?

Thats what i meant. That they might try to accelerate development to get their own chiplet based arch development to compete with AMD. Even if, say 6 months later than AMD.
 

maddie

Diamond Member
Jul 18, 2010
4,723
4,628
136
Surely the gaming product after Lovelace is chiplet based. Maybe they will just cancel Lovelace and pull that in?
Thats what i meant. That they might try to accelerate development to get their own chiplet based arch development to compete with AMD. Even if, say 6 months later than AMD.
That sounds too simple. If money alone was the issue, then maybe. I keep thinking of serial vs parallel code.
 
  • Like
Reactions: Timmah!

gorobei

Diamond Member
Jan 7, 2007
3,654
980
136
Surely the gaming product after Lovelace is chiplet based. Maybe they will just cancel Lovelace and pull that in?
Thats what i meant. That they might try to accelerate development to get their own chiplet based arch development to compete with AMD. Even if, say 6 months later than AMD.
according to rumors from maybe a year or two ago, hopper was/is chiplets and was supposed to come next. but then nv announced they were going monolithic for the rtx4000 series. their chiplet whitepaper came out around 2017, so 2020/21 was reasonable for chiplets coming out.

but with covid, tsmc having more customers and no open capacity forcing nv to samsung, and samsung's 8(10)nm node not being close to tsmc; i could see nv having to go to plan b.

ampere on samsung didnt deliver the performance so they had to pump the voltage. nv dies are near the reticle limit for litho machines so they arent going to get a lot of good dies even from a good yielding tsmc 5nm. with the component shortages, the complex packaging would just get more expensive. and tsmc has no reason to cut nv a break on pricing with 0 unused capacity. all of those things would make multichiplet unprofitable for nv and their team green tax.

sticking with samsung gets them much better fabrication costs, and since there is no word on samsung being able to do complex packaging that rules out chiplets. it is unlikely they would be able to pull chiplets forward while costs are so high. its worth it for amd to push it now to take the performance crown. if rtx4000 is close enough they will simply move the goalposts to raytracing and directed learning while losing in raster fps.
 

soresu

Platinum Member
Dec 19, 2014
2,617
1,812
136
and since there is no word on samsung being able to do complex packaging that rules out chiplets.
There is this 'I-Cube4' packaging announced a couple of weeks ago, and there was this 'X-Cube' thing for 3D packaging too.

So even if they don't have a logic chiplet specific option yet they clearly have complex packaging in the pipeline.
if rtx4000 is close enough they will simply move the goalposts to raytracing and directed learning while losing in raster fps.
Yeeeeaaaahhhhh.

Pretty sure they moved those goalposts back with Turing - at the very least from a PR messaging perspective.

If not they wouldn't have made such a HUGE deal about it with hardly any games in the pipeline when it was released.

Their concentration on 8K capability marketing for the Ampere high end also implies they already have some fear on the raster side of things when there is minimal uptake of 8K TVs so far (vs 4K).
 
Last edited:

Dribble

Platinum Member
Aug 9, 2005
2,076
611
136
I saw some utube videos and read some speculation articles about dual-tile 7900xt, apparently 2,5 faster than current high-end....some hellbent Lovelace with it still single monolithic chip, or whatever Nvidia will compete with, wont stand a chance....
I wonder if Nvidia would really let that happen.
You could basically find a variant of this sentence for pretty well every gpu release in the last 10 years.
 
  • Like
Reactions: Mopetar

Mopetar

Diamond Member
Jan 31, 2011
7,797
5,899
136
Surely the gaming product after Lovelace is chiplet based. Maybe they will just cancel Lovelace and pull that in?

It'd cost a lot to scrap it entirely and even if they try to accelerate its replacement they can't magically ensure that happens as quickly as they want either, particularly because moving to a chiplet design has plenty of things that could go wrong or cause delays.

Even if AMD does surpass them in raw performance, Nvidia will still likely have better RT performance and I'm guessing they'll really push that and their next generation of DLSS. As long as that let's them crank out at least one graph with bigger bars there are going to be plenty of people who convince themselves it's the better card.

They can also cut prices quite a bit. Margins can go back to pretty good from obscene and if they launch first people will be praising the cards and company for offering such a great value.