NVIDIA plans to launch its final Ada Lovelace GPU for gamers next year. The GeForce RTX 4080 Ti (not Tie) is allegedly on track for an early 2024 release. According to @Zed__Wang, it’ll leverage the AD102 die (as does the RTX 4090) and target a TBP of under 450W (~400W?). At around $1,199, it should cost roughly the same as the RTX 4080, which should have been cheaper to begin with.
The GeForce RTX 4090 leverages a partially disabled AD102 die with one of the GPCs and a TPC from two GPCs fused off. Ideally, you’d expect the RTX 4080 Ti to come with two disabled GPCs and be done with it, but I suppose we’ll have to wait and see.
If the RTX 4080 Ti does feature the AD102 minus two GPCs, you’re looking at 15,360 FP32 cores spread across 120 SMs, 1,024 shaders, or eight SMs less than the RTX 4090. The performance deltas between the RTX x90 and x80 Ti generally tend to be minimal, or so it has been.
On the memory side, expect up to 20GB of GDDR6X memory coupled with a 320-bit bus, resulting in an external bandwidth of under ~900GB/s. The core clocks should be slightly higher than the GeForce RTX 4090, at around 2.55GHz or higher. A TBP of 400W should be more than enough to sustain those clocks.
And then there’s the pricing. The GeForce RTX 4080 launched at an MSRP of $1,199 instead of $699 like its predecessor. NVIDIA marketed it as the successor to the RTX 3080 Ti, although the gains are hardly comparable to a next-generation product. With the RTX 4080 Ti, Jensen will claim that the prices of higher-end GPUs are being reduced, but existing RTX 3080 Ti owners will know what’s up.
