NVIDIA is supposedly working on an RTX 40 “Super” Refresh with an expected Q1 or Q2 2024 launch. According to rumors, a 20GB variant of the GeForce RTX 4080 based on the AD102 is on the horizon. It’ll allegedly feature a 320-bit bus and additional SMs, slotting it between the RTX 4080 and 4090. From what I can surmise, this is likely the RTX 4080 Ti rather than an RTX 4080 Super. There’s only room for one SKU in there.
GPU | Nvidia RTX 4060 Ti | NVIDIA RTX 4070 | AMD Radeon RX 7800 XT | AMD Radeon RX 7700 XT |
---|---|---|---|---|
GPU Die | AD106 | GA104 | Navi 32 | Navi 32 |
SMs | 34 | 46 | 60 | 54 |
Cores | 4,352 | 5,888 | 3,840 | 3,456 |
RT Cores | 34 | 46 | 60 | 54 |
Base clock | 2,310 MHz | 1,920 MHz | 1,295MHz | 2,171MHz (Game) |
Boost clock | 2,535 MHz | 2,475 MHz | 2,430MHz | 2,544MHz |
Bus Width | 128-bit | 192-bit | 256-bit | 192-bit |
Bandwidth | 288GB/s | 504GB/s | 624GB/s | 432GB/s |
Memory | 8GB GDDR6 | 12GB GDDR6X | 16GB GDDR6 | 12GB GDDR6 |
Memory Clocks | 18Gbps | 21Gbps | 19.5Gbps | 18Gbps |
TBP | 160W | 200W | 263W | 245W |
LL Cache | 32MB L2 | 36MB L2 | 64MB L3 | 48MB L3 |
MSRP | $399 | $599 | $499 | $449 |
Two new variants of the GeForce RTX 4070 are in the pipeline: A Super based on the AD103 GPU with 16GB memory and a 256-bit bus and a downgrade with GDDR6 memory (instead of GDDR6X). It is unclear whether the RTX 4070 Super will feature an increased core count over the existing model. Regardless, a memory bus upgrade should provide an ample uplift at higher resolutions.

There’s no word on a potential RTX 4060 Super, but I reckon the downgraded GDDR6 variant of the 4070 may take its place. Out of all the Ada Lovelace GPUs, the GeForce RTX 4070, and the RTX 4090 have been the most successful. With the launch of the Radeon RX 7800 XT, AMD has intensified the competition in the midrange segment. Consequently, we’re getting two variants of the RTX 4070 to tackle the two newcomers.