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

Page 106 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

dlerious

Golden Member
Mar 4, 2004
1,781
722
136
Hmm... was randomly looking at Newegg and 4090 supply is very low. I imagine a lot of that is because they are probably sending every 4090 they have to China right now, but still.
Looks like Huawei has chips that are alternatives to the Nvidia A100 according to this story

 

blackangus

Member
Aug 5, 2022
69
95
51



WAHHHHTTTT????? :eek:

This could be the end of AMD and Intel in the GPU race if nGreedia buys Diamond Foundry and uses these wafers exclusively for its GPUs. Of course, I'm just thinking of a worst case scenario. Hopefully, this will never happen.
Well who would care?
Its Nvidia so it would 3x the performance at 10x the cost!
 

gorobei

Diamond Member
Jan 7, 2007
3,663
990
136
there are some diamond/graphene based tape/fabrics with massive thermal conductivity that can transport a ton of heat from a ihs(they are used in smartphones and other space constrained electronics that dont have big heatsink fin arrays, but the heat still has to go somewhere to be radiated/convected.

getting heat off of the substrate around the chip is fine but unless you alter how everything is packaged, i dont see how you get the substrate heat to a fin array any better than getting it off of the gpu ihs.
 

jpiniero

Lifer
Oct 1, 2010
14,571
5,202
136

Now it's official that they will be presenting at CES.
 
  • Like
Reactions: Mopetar and Ajay

MrTeal

Diamond Member
Dec 7, 2003
3,568
1,696
136
If this works like Turing where the Supers generally slotted in at the same MSRP as the regular ones, those rumoured specs would make for a nice upgrade in the upper midrange.

4080 Super at 4080 MSRP is pretty meh given that it's only 5% more shaders and you can already buy a 4080 for $100 under MSRP in the US.
4070 Ti Super is a stupid name, but would be a pretty decent GPU. 256-bit bus would help with its drop-off at higher resolution, and would really force a price drop on the 7900XT given it'd faster than it with the Nvidia feature set and cachet at the same price.
4070 Super at $600 is a big upgrade, should be almost 4070 Ti level of performance. Newegg has a 4070 at $530, so it's not as much of a no brainer as it seems in that market as just comparing MSRP, but it'll look great in reviews. It'd pull it close to the 7800XT @ $500 in cost per frame in raster, so there's not a whole lot of reason to buy that card unless it drops or you just can't swing the extra $100.
 
  • Like
Reactions: Tlh97 and Ajay

Ajay

Lifer
Jan 8, 2001
15,402
7,830
136
4070 Ti Super is a stupid name, but would be a pretty decent GPU. 256-bit bus would help with its drop-off at higher resolution, and would really force a price drop on the 7900XT given it'd faster than it with the Nvidia feature set and cachet at the same price.
So, is the 4070 Ti Super a cut down AD103 die? I assume with a 256b bus that it will have 16GB of RAM (Yes/no?).
 

jpiniero

Lifer
Oct 1, 2010
14,571
5,202
136
4070 Ti Super is a stupid name, but would be a pretty decent GPU. 256-bit bus would help with its drop-off at higher resolution, and would really force a price drop on the 7900XT given it'd faster than it with the Nvidia feature set and cachet at the same price.

I have a feeling that the 4070 Ti Super will have a higher MSRP than $799.
 

Aapje

Golden Member
Mar 21, 2022
1,370
1,836
106
Kopite7kimi said 4070 Ti Super can be either AD103-275 or AD102-175. In the latter case that's one heck of a cut.
There's no way it will be a 102. The production costs are way higher for that chip and they can easily sell them.
 

Mopetar

Diamond Member
Jan 31, 2011
7,826
5,968
136
There's no way it will be a 102. The production costs are way higher for that chip and they can easily sell them.

Or it's a bottom bin salvage part that they'd otherwise have to throw away. NVidia has done similar things in the past towards the end of a generation. There were 1060s that used the GP-104 die even though it was basically cut in half just because after two years they built up enough of them to create a version of the 1060 based on it.

If the quantity is low (or they just price it high enough to keep the demand in check) they can always offer it as an OEM-only part.
 

Heartbreaker

Diamond Member
Apr 3, 2006
4,226
5,226
136
I'd say it's 50/50. AD104 is selling for 550-800 right now. AD103 die is ~25% bigger, so I can see nvidia charging $800 for the cut down part (45% higher price than $550) and be happy.

This is NVidia. They don't have to go that far, so they won't. They will probably price Super parts a little higher to help clear, or coexist with the current parts.
 
  • Like
Reactions: blckgrffn

Kepler_L2

Senior member
Sep 6, 2020
329
1,160
106
Or it's a bottom bin salvage part that they'd otherwise have to throw away. NVidia has done similar things in the past towards the end of a generation. There were 1060s that used the GP-104 die even though it was basically cut in half just because after two years they built up enough of them to create a version of the 1060 based on it.

If the quantity is low (or they just price it high enough to keep the demand in check) they can always offer it as an OEM-only part.
Also the 2060 KO using TU104.
 
  • Like
Reactions: Mopetar

MrTeal

Diamond Member
Dec 7, 2003
3,568
1,696
136
While the 2060 KO was only 62.5% of the full die shaders it was the only TU104 die with a 192-bit instead of 256-bit bus. It was a pretty limited release and a good way to dump otherwise good silicon with a bum MC. Same with the GP104 1060's, everything else in the stack higher had the full 256-bit bus, so that was the only place to dump defective MC dies.

AD102 with half the shaders off and two MC's disabled is weird as a 4070 Ti Super, since basically the same die could go into a 4080 Super. Your vapor chamber needs to be big enough to still contact the AD102 die on a cull part which is 60% bigger than AD103, so that cost would be easier to eat on a $1200 card instead of a $800 card.
 

Aapje

Golden Member
Mar 21, 2022
1,370
1,836
106
Or it's a bottom bin salvage part that they'd otherwise have to throw away. NVidia has done similar things in the past towards the end of a generation. There were 1060s that used the GP-104 die even though it was basically cut in half just because after two years they built up enough of them to create a version of the 1060 based on it.

If the quantity is low (or they just price it high enough to keep the demand in check) they can always offer it as an OEM-only part.
The 4090 is a pretty deep cut, so they won't have that many units which fail to reach that level. The main issue is going to be chips with partially defective IO. However, they recently released a salvage part to take care of that, the A5000.

That is a 256 bit bus AD102 variant with a third of the chip disabled, making it only slightly faster than AD103.

If enterprises don't buy enough of those $4k cards, that are effectively just slightly faster 4080's with double the VRAM, for more than triple the cost, I can see them wanting to sell these chips as 4080 TI cards. But not right now.
 
Last edited:

jpiniero

Lifer
Oct 1, 2010
14,571
5,202
136

Seemingly confirmation that the 4070 Non-Super will remain being sold. Probably get an official price cut to $549 or possibly even $499.
 

MrTeal

Diamond Member
Dec 7, 2003
3,568
1,696
136
Seemingly confirmation that the 4070 Non-Super will remain being sold. Probably get an official price cut to $549 or possibly even $499.

At $500 for the 4070 they'd have to cut the 16GB 4060Ti to at least $450, but realistically remove one of the two SKUs altogether. In a better world maybe we'd see the 8GB 4060 Ti disappear and the 16GB at $400, but you don't get 75% gross margins just giving away RAM for free.
 
  • Haha
Reactions: Tlh97 and blckgrffn

jpiniero

Lifer
Oct 1, 2010
14,571
5,202
136
At $500 for the 4070 they'd have to cut the 16GB 4060Ti to at least $450, but realistically remove one of the two SKUs altogether. In a better world maybe we'd see the 8GB 4060 Ti disappear and the 16GB at $400, but you don't get 75% gross margins just giving away RAM for free.

Yeah I am thinking the 16 GB model will go away.
 

Kepler_L2

Senior member
Sep 6, 2020
329
1,160
106
At $500 for the 4070 they'd have to cut the 16GB 4060Ti to at least $450, but realistically remove one of the two SKUs altogether. In a better world maybe we'd see the 8GB 4060 Ti disappear and the 16GB at $400, but you don't get 75% gross margins just giving away RAM for free.
Maybe Jensen is making so much money with AI he's feeling generous for once and will give gamers a good deal :D
 
  • Haha
Reactions: igor_kavinski

MrTeal

Diamond Member
Dec 7, 2003
3,568
1,696
136
That's just it, he's so busy with AI that he does not have time for discounts.
He probably just asks ChatGPT if he should sell more cards to gamers, but the answer keeps coming back "More flops to me. So close..."
 

coercitiv

Diamond Member
Jan 24, 2014
6,175
11,805
136
He probably just asks ChatGPT if he should sell more cards to gamers, but the answer keeps coming back "More flops to me. So close..."
Well, let's ask Copilot!

Q: Should Nvidia lower price of products?
That is a difficult question to answer, as Nvidia’s pricing strategy depends on many factors, such as its target market, competitive advantage, product differentiation, and cost structure. According to some sources, Nvidia has adopted a skimming pricing policy for its products in the consumer market, meaning that it charges a high price for its innovative and qualitative products, and then gradually lowers the price over time as the product becomes more common and faces more competition1. This strategy allows Nvidia to maximize its profits in the early stages of the product life cycle, and also to recover its research and development costs quickly. However, this strategy also has some drawbacks, such as alienating price-sensitive customers, attracting more competitors, and creating a negative image of being overpriced.

If Nvidia decides to lower the price of its products, it might be able to gain more market share and increase its customer loyalty, especially in emerging markets where the demand for GPUs is growing rapidly. However, lowering the price might also have some negative consequences, such as reducing its profit margin, diluting its brand value, and triggering a price war with its rivals, such as AMD and Intel. Therefore, Nvidia should carefully weigh the pros and cons of lowering its prices, and consider the impact of its pricing decisions on its long-term goals and vision.

Q: Let's pick a conclusion with the limited data available, should Nvidia lower prices?
Based on the limited data available, I would say that Nvidia should not lower its prices, as it might lose its competitive edge and profitability in the long run. Nvidia has established itself as a leader in the GPU market, and its products are known for their high quality and performance. By lowering its prices, Nvidia might compromise its brand image and reputation, and also invite more competition from its rivals. Nvidia should instead focus on maintaining its innovation and differentiation, and creating value for its customers. This way, Nvidia can justify its premium pricing and retain its loyal customer base. Of course, this is just my opinion based on the information I have, and there might be other factors that Nvidia should consider before making its pricing decisions.

There we have it, the flops know what's good for them :p