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

Page 57 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

repoman0

Diamond Member
Jun 17, 2010
4,544
3,471
136
So it turns out there is nothing special about the sense pins -- just a two bit binary signal that could be easily implemented totally within the cable with no "help" from the PSU. I am assuming Cablemod and other vendors are smart enough to hook them up properly depending on how many PCI-e connectors the cable has. Wouldn't even need a chip/IC to do it.

 
Reactions: Leeea

biostud

Lifer
Feb 27, 2003
18,402
4,965
136

Alleged 4080 16 GB benchmarks. Gets 13977 in TSE Graphics. That's only about 30% faster than the average 3080 Ti. Even Port Royal is also around 30% faster than the average 3080 Ti.
It is all about DLSS 3
 
Reactions: Leeea

repoman0

Diamond Member
Jun 17, 2010
4,544
3,471
136
So basically the 4090 is the only real upgrade for 3080+ owners. Could have figured as much from the specs. The pricing on both 4080 variants is pretty nonsense .. might as well just go with a cheaper 3080 or 3090, or step up to a 4090.

DLSS3 is a joke to me, not interested in anything that adds latency.
 

Stuka87

Diamond Member
Dec 10, 2010
6,240
2,559
136
GN: "I see your 4090 unboxing, and raise you cutting a 4090 cooler in half":

I really enjoyed this video. I really like when GN puts out these very technical videos, or they get to interview an engineer, and not some sales person.

nVidia's executives may be some questionable people, nVidia's engineers are very good at what they do.
 

SteveGrabowski

Diamond Member
Oct 20, 2014
7,121
5,998
136
Let's not ignore Alchemist GPUs completely. They could be quite good mid-range cards if the drivers get there in a few months. I am rooting for Intel to stick it to the other 2 companies.

The 16GB A770 is definitely tempting, but as well as I imagine it'll play Starfield, the two 2023 games I'm most excited for are Like a Dragon Ishin and Like a Dragon Gaiden, and not sure those will be big enough sellers to get the kind of careful optimization in the drivers that games like Starfield or Forspoken or Final Fantasy 16 will get.
 
Reactions: Leeea

Saylick

Diamond Member
Sep 10, 2012
3,385
7,151
136
Look on the bright side. Even if they cost $3000 dollars the box will be big enough to live in so you can sell your house if you have trouble affording one.
And even if the box has the thermal insulation of... well, a cardboard box, the RTX 5090 will surely keep the interior nice and warm. Maybe even toasty.

EDIT: Looks like some more info came out regarding OC'ing the 4090:
 
Last edited:
Reactions: Tlh97 and Leeea

jpiniero

Lifer
Oct 1, 2010
14,841
5,456
136

3dmark leaked results from Videocardz. Says that 4090 results could be closer to 70-80% faster actually. That wouldn't be terrible if it translated to games. They have both a stock card at the 450 W TDP and an overclocked AIB card which is 2-3% faster.
 

Leeea

Diamond Member
Apr 3, 2020
3,698
5,432
136

3dmark leaked results from Videocardz. Says that 4090 results could be closer to 70-80% faster actually. That wouldn't be terrible if it translated to games. They have both a stock card at the 450 W TDP and an overclocked AIB card which is 2-3% faster.
Intel ARC did really good in 3d mark also, did translate to real world performance.

I do believe the rtx 4090 will be an exceptionally well performing card.

80% more performance on a 256 bit bus? possible with that cache, but seems unlikely.
 

yottabit

Golden Member
Jun 5, 2008
1,375
240
116
Intel ARC did really good in 3d mark also, did translate to real world performance.

I do believe the rtx 4090 will be an exceptionally well performing card.

80% more performance on a 256 bit bus? possible with that cache, but seems unlikely.

4090 has the full 384 bit bus
4080 16 GB has 256 bit bus
4080 12 GB has 192 bit bus

So I think the 4090 numbers are viable to hold up through 4K. why I’m so keen to see how much uplift the 4080 16 GB has at 4K specifically. I get the sense the 4080 12 GB will be a great 1080p card, the 16 GB will be a great 1440p card, but they may have some dropoff in their (peak) gains over previous gen at 4K

also why it makes the prospect of a 4080 Super or Ti with a bit more memory bandwidth and maybe a price adjustment and interesting prospect this time

I’m also pretty curious as to the undervolting performance of the 4090 and hope that is explored in launch day reviews
 
Reactions: Tlh97 and Leeea

yottabit

Golden Member
Jun 5, 2008
1,375
240
116
@yottabit

LOL at trying to sell the fake 4080 as a 1080p card at this point.

That’s honestly what I’m expecting. I’m not really trying to defend Nvidia. Has anyone seen the 12 GB pop up anywhere yet as an AIB? It almost smells like something that will be mostly for OEMs like Alienware etc. mentioned they won’t be doing a founders edition.

The counter argument would be, with the 10x larger L2, maybe Nvidia knows what they are doing and the memory bandwidth is sized to match the cores even at higher resolutions. I just find that unlikely

I’m sure the 12 GB will be fine for 1440p too. Just theorizing that if it might show better (comparative to last gen) performance at 1080p. I don’t see how the memory bandwidth of the 12 GB model could not be a bottleneck at 4K / VR even with the larger cache. And it’s not like it has THAT many less shaders than the 16 GB.

I’m really curious if the 16 GB model will keep the same +40% trend over the 3080 (shown at 1440p), when running at 4K. That’s the big question mark to me
 
Reactions: Leeea

jpiniero

Lifer
Oct 1, 2010
14,841
5,456
136
The counter argument would be, with the 10x larger L2, maybe Nvidia knows what they are doing and the memory bandwidth is sized to match the cores even at higher resolutions. I just find that unlikely

I think it was meant to be in combination with GDDR7. If the 16 GB had 28 gbps GDDR7 instead, that would give it the same memory bandwidth as the 3080 12 GB and 3080 Ti.

A Full AD103 with that should be decently faster enough for a 4080 Super next year.
 

gorobei

Diamond Member
Jan 7, 2007
3,713
1,067
136
lol, wan show podcast went thru a size comparison of gtx1000, rtx3000, and rtx4000. the gtx1080 as size reference is hilarious (video indexed to time)
the weight of those things will lead to a boom product market: gpu support brackets.
they were niche before, but now they will be mandatory if not integrated into the case to begin with.

 
Last edited:

gorobei

Diamond Member
Jan 7, 2007
3,713
1,067
136
I really enjoyed this video. I really like when GN puts out these very technical videos, or they get to interview an engineer, and not some sales person.

nVidia's executives may be some questionable people, nVidia's engineers are very good at what they do.
while the execs are doing the legally required job of maintaining shareholder value/stock price, it would be nice if they could actually learn to lose gracefully.

they are pushing the limits of residential power infrastructure (circuit breakers and wiring) and the amount of engineering required to cool these monstrosities is well past ridiculous. the amount of metal/weight in the cooler is hitting 5 pounds(2186 grams) which is twice the weight of some of the cpu air coolers back in the core2 days that would bend motherboards and cause crashes. and dont get me started about pushing past 3 slots thickness.

all of this is because they cant transition to chiplets and stay in their required profit margins without losing to amd in performance. if nv cards had to stay in a traditional 2 or 2.5 slot volume they would be either hotter or noisier with how hard they are pushing the silicon voltage/freq curve. so they have to push past that size limit because they need to increase the total cooler fin surface area to reject the increased heat from pushing the voltage.

if they did stay in the realm of reasonable power draw/size, they would probably lose to amd in perf. and losing would have financial consequences. but being 2nd for a product cycle or two until they get chiplets working properly would be a far more reasonable course than pushing monolithic dies and tripping circuit breakers (or worse) potentially starting fires in wall wiring of older homes.


while microsoft execs learned the lesson of cheaping out on cooling with the xbox360 RROD, this seems like going too far towards the other end of the scale.

but yes, it was cool to see the engineer go into the math of the fan/finstack parameters.
 

TESKATLIPOKA

Platinum Member
May 1, 2020
2,429
2,914
136
I have to wonder what will Nvidia ask for the full AD106, when the full AD104 is already $899.
AD106 is supposedly ~190mm2 while AD104 is 294.5mm2.

AmpereDie Size [mm2]SMCudaTMUROPCache [MB]Memory bus [bit]Vram [GB]
GA107?202560803221284
GA1062763038401204831928
GA10439248614419296425612

I think Ada based full chips could look like this.
Ada LovelaceDie Size [mm2]SMCudaTMUROPCache [MB]Memory bus [bit]Vram [GB]
AD104294.5607680240804819212
AD106~19040512016040-48 ?321288
AD107?243072963216-24 ?96-128 ?6-8 ?

The full AD106 should be RTX 4060Ti or maybe RTX 4070?
I expect 3070-3070Ti level of performance with ~200W TDP, and their official MSRP was $499-599. I think, Nvidia will price It at $499, maybe $449 If we are lucky, in other words $50-100 more than RTX 3060Ti. The problem will be the big gap in price between full AD106 and AD104. In my opinion, only one cut down SKU won't be enough to fill the gap.

The fight AD106 vs N33 will be very interesting.


edit: AD106 with 40SM should perform a lot better than 3070-3070ti.
 
Last edited:
Reactions: Tlh97 and maddie

jpiniero

Lifer
Oct 1, 2010
14,841
5,456
136
The 4070 is very likely a cut AD104 - 7160 shaders, maybe 160 bit bus. I have doubts you will even see that soon, let alone AD106.

After all, the 3060 Ti GDDR6X is going to be announced soon.
 

CakeMonster

Golden Member
Nov 22, 2012
1,428
535
136
The 4090 resale value should probably worry some people, because when you upgrade to 5xxx or 6xxx, any potential used buyer of your 4090 is not just going to worry about a power draw disadvantage, but they can't literally fit it in 50% of cases either.
 
Reactions: Tlh97 and Leeea
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |