Till now, most reports have claimed that the top-end AD102 die powering the RTX 4090 will feature up to 18,432 cores across 84 SMs and 7 GPCs paired with 24GB of GDDR6X memory and 96MB of L2 cache. This monster GPU was expected to have a TGP of 600W+, nearly 2x more than the RTX 3090’s 350W. The latest rumor from Twitter claims that NVIDIA is testing an even more powerful SKU based on the AD102. Although the shader count is unchanged at 18,432, the memory has been doubled to 48GB and the boost clock has been raised by a few hundred MHz. To top it off, the maximum boost power consumption is set to a whopping 900W.
Now, it’s important to note that this GPU only exists as a test product (or on paper) at the moment, and it’s unclear whether it’ll ever come to the consumer market. If I had to bet, I’d say that NVIDIA is keeping a backup option in place in case the Radeon RX 7900 XT ends up being faster than the RTX 4090 by a large margin. This could become either the RTX 4090 Ti or the next Titan.
|GPU||GA102||AD102||RTX 4090||AD103||RTX 4080||AD104||RTX 4070|
|Arch||Ampere||Ada Lovelace||Ada Lovelace||Ada Lovelace|
|Process||Sam 8nm LPP||TSMC 5nm||TSMC 5nm||TSMC 5nm|
|TP||37.6||~100 TFLOPs?||83 TFLOPs||~50 TFLOPs||47 TFLOPs?||~35 TFLOPs||35 TFLOPs?|
|Memory||24GB GDDR6X||48GB GDDR6X||24GB GDDR6X||16GB GDDR6X||12GB GDDR6X|
|Launch||Sep 2020||Sept 22?||Sept 22?||Q1 2023?|
Last but not the least, a 900W TGP seems highly unlikely. If NVIDIA does decide to bring this SKU to the gaming market, the power consumption will have to be optimized. You can expect a peak boost consumption of 700-750W, but not more. The RTX 4090, on the other hand, should have a TGP of 600W, 100W less than the H100, and 250W more than the RTX 3090.