A little birdie has told us that NVIDIA’s next-gen GeForce RTX 4080 and 4070 graphics cards will both feature 16GB VRAM buffers. The key difference is that the former will leverage Micron’s PAM4 equipped GDDR6X memory chips, the latter will retain GDDR6, albeit faster 16Gbps modules [guesswork]. The RTX 4080 will be based on the AD102 die, although there is a chance NVIDIA might want to downgrade it to the AD103 core. This would push the RTX 4080 Ti and 4090 a tier above the 4080, rather than just being halo products with horrible performance per dollar.
|Arch||Turing||Ampere||Ada Lovelace||Ada Lovelace||Ada Lovelace|
|Process||TSMC 12nm||Sam 8nm LPP||TSMC 5nm||TSMC 5nm||TSMC 5nm|
|TP||16.1||37.6||~90 TFLOPs?||~50 TFLOPs||~35 TFLOPs|
|Memory||11GB GDDR6||24GB GDDR6X||24GB GDDR6X||16GB GDDR6||16GB GDDR6|
|Launch||Sep 2018||Sep 2020||Aug-Sep 2022||Q4 2022||Q4 2022|
It’s worth noting that prior to Ampere, the RTX/GTX x80 and x70 GPUs were based on the Gx104 die. The x80 Ti and x90/Titan were the sole SKUs to leverage the top-end Gx102 die. With Ampere, NVIDIA used the GA102 die for the RTX 3080, 3080 Ti, 3090, as well as the 3090 Ti, leaving marginal performance deltas between GPUs ranging from $699 to $1,499. The GA104, on the other, was used to feed the RTX 3060 Ti, RTX 3070, and the RTX 3070 Ti.
Regardless, NVIDIA will be equipping its higher-end RTX 4070, and 4080 graphics cards with 16GB of memory via a 384-bit, and 256-bit bus, respectively. This will be backed by 96MB, and 64MB of L2 cache, respectively. The RTX 4090 will likely pack 24GB, and the 4080 Ti will be paired with 20GB of GDDR6X memory. In the mid-range segment, the RTX 4060 should pair the AD104 die with 12GB GDDR6 memory via a 192-bit bus, and the RTX 3050 will likely stick to 8GB.