NVIDIA 4nm AD104 "Ada" Silicon Pictured, Half the Die-area of AD102

[btarunr]

Here's the first picture of the 4 nm "AD104" silicon powering what would have been the $900 GeForce RTX 4080 12 GB, and upcoming RTX 4070-series graphics cards. The third largest GPU based on the "Ada Lovelace" graphics architecture, the AD104 looks tiny. This is because it has roughly half the die-area of the AD102, estimated to be around 295 mm² (compared to 608 mm² of the AD102), which means its transistor count should be less than half, with older reports pinpointing it to 35.8 billion. The RTX 4080 12 GB was supposed to max out the AD104 silicon, enabling all 60 streaming multiprocessors (SM) physically present.

The AD104 with 60 SM hence has 7,680 CUDA cores, 60 RT cores, 240 Tensor cores, 240 TMUs, and 80 ROPs. NVIDIA has generationally narrowed the memory interface (compared to the GA104 and TU104), down to 192-bit GDDR6X. Its predecessors such as the GA104 feature 256-bit wide memory interfaces. NVIDIA is overcoming the memory bus width "deficit" by giving SKUs based on the silicon higher memory speeds (21 Gbps or more); and architecture-level improvements such as larger on-die caches. NVIDIA is reportedly planning to launch an AD104-based SKU early January 2023. VideoCardz reports that could be the RTX 4070 Ti, a re-branding of the RTX 4080 12 GB.



View at TechPowerUp Main Site | Source

 

[P4-630]

Not too bad if they sell it at 399,-....

 

[THANATOS]

Not too bad if they sell it at 399,-....

That's too unrealistic considering they at first wanted to sell It for 899,-

 

[freeagent]

I predict 799

 

[P4-630]

That's too unrealistic considering they at first wanted to sell It for 899,-

Always bid low...

 

[Vayra86]

A nice x70 indeed, that would be 500 maximum, Nvidia. 499- that is. Fair, sensible. Higher? Gtfo

 

[P4-630]

A nice x70 indeed, that would be 500 maximum, Nvidia. 499- that is. Fair, sensible. Higher? Gtfo

One can only hope....

 

[blu3dragon]

3060 had a 192 bit memory bus, 276mm2 die size and launched at 329 USD.

8nm Samsung vs 4nm tsmc.

GA104 (3070/3060 ti) was 392 mm2

 

[THU31]

They sold the GTX 1080 (full GP104, only 314 mm2) for $599, and that was 6 years ago, so what can we possibly expect with this considering current wafer costs. This generation seems like a skip.

 

[Vayra86]

One can only hope....

Hope or dream, otherwise Im simple enough to not press buy, honestly. This gpu is right on the edge of what could still be called upper midrange. Paying more is just not happening lol

 

[N/A]

Very similar to 103 but shorter by 22%, same height. That is most likely where it came from and how the memory channel got dropped. Can't explain how 10240 Cuda can be translated to 112 rops. 10752 and 7680 are 7 and 5 GPC with 16 Rops each.

 

[Easo]

Yeah, you can forget this going for 500 USD/EUR/GBP. It will be more.

 

[defaultluser]

I predict 799

yeah, by the time card is launched in January, only a fool would pay $900 for this POS

 

[_Flare]

If the 4070Ti gets a higher MSRP than the 699 USD of the 3080 10GB and the 3070Ti 16GB just dont buy it.

 

[Why_Me]

If the 4070Ti gets a higher MSRP than the 699 USD of the 3080 10GB and the 3070Ti 16GB just dont buy it.

Is there such a card?

 

[Minus Infinity]

Fun fact, 4070 class cards were originally going to use AD103 up until quite recently. Why the change? So let's see how the money grubbing Huang will price the real 4070 Ti as anything over $649 is ludicrous, but $250 under the la la land 4080 12GB. AMD will certainly be chuckling when they unleash the 7700 and 7800XT to fight in this class.

 

[defaultluser]

Fun fact, 4070 class cards were originally going to use AD103 up until quite recently. Why the change? So let's see how the money grubbing Huang will price the real 4070 Ti as anything over $649 is ludicrous, but $250 under the la la land 4080 12GB. AMD will certainly be chuckling when they unleash the 7700 and 7800XT to fight in this class.

no, its not

NVIDIA has a long history of doing this only until demand is met- see the desktop rtx 3050, which officially uses a 106 cut, but the laptop Ti part uses the exact same 2560 core count 107!

I guarantee you the 4070ti will have a similar cut 103 PART TOO, early-on just not in any official spec sheet

 

[Minus Infinity]

no, its not

NVIDIA has a long history of doing this only until demand is met- see the desktop rtx 3050, which officially uses a 106 cut, but the laptop Ti part uses the exact same 2560 core count 107!

I guarantee you the 4070ti will have a similar cut 103 PART TOO, early-on just not in any official spec sheet

So you are telling us that Nvidia has scrapped all the 4080 12GB cards that use AD104 and will sell 4070 Ti's with cut down AD103. Nope no way. They will reserve AD103 for a future product depending on how much 7800XT kicks the 4070 Ti's arse.

 

[THU31]

I still think it would be stupid to release a Ti card in the initial line-up. What will they call the refreshed card a year later? 4070 Super? How is someone supposed to know if Super is better than Ti.

And they will have to do a refresh.

But I guess they did it with Turing. 2080 and 2080 Ti came out together, and a year later there was a 2080 Super, but it was a refresh of the normal 2080, not the Ti. It makes no sense to me. 4070 Ti Super?

 

[defaultluser]

So you are telling us that Nvidia has scrapped all the 4080 12GB cards that use AD104 and will sell 4070 Ti's with cut down AD103.

Absolutely - assuming the yields on the big core start out poor, the 4070 ti makes sense as the initial cut

once yields improve, we could see a 4080 SE during product refresh (now that thjey haver high enouh yields of Ada 104 to make aall the deman for 4070 Ti plus cut 4070, and the yields on 103 is similarly improved)