Today’s RTX 40 Super rumor comes from a very reliable source (the most). Kimi has shared his thoughts (via Harukaze) on the GeForce RTX 4070 Super and 4080 Super, including the core and memory specifications. According to him, the GeForce RTX 4080 to 4080 Super upgrade will be similar to the previous 80-class refresh. Ergo, the RTX 4080 Super will feature the full-fledged AD103 GPU with 10,240 FP32 cores with the same 256-bit bus but with faster 16GB of GDDR6X memory.
The upgrade would result in a minor (~10%) performance uplift over the RTX 4080. Perhaps if the pricing is reduced to $999, it would make more sense. Either way, NVIDIA is under no pressure in this segment.
| GPU | NVIDIA RTX 4070 Ti (ASIC) | NVIDIA RTX 4080 FE | NVIDIA RTX 4070 FE |
|---|---|---|---|
| GPU Die | AD104 | AD103 | AD104 |
| SMs | 60 | 76 | 46 |
| CUDA cores | 7,680 | 9,728 | 5,888 |
| RT Cores | 60 | 76 | 46 |
| Tensor Cores | 240 | 304 | 184 |
| Base clock | 2,310 MHz | 2,205 MHz | 1,920 MHz |
| Boost clock | 2,640 MHz | 2,508 MHz | 2,475 MHz |
| Memory | 12GB GDDR6X | 16GB GDDR6X | 12GB GDDR6X |
| TBP | 285W | 320W | 200W |
| L2 Cache | 48MB | 64MB | 36MB |
| MSRP | $799 | $1199 | $599 |
The GeForce RTX 4070 Super will be more important as AMD has potent offerings in this price range. The Radeon RX 7700 XT and 7800 XT offer unbeatable value for money, especially after the latest price cuts. The 70-class Super will allegedly be a cut-down RTX 4070 Ti.
A cut-down AD104 die with roughly 7,000 FP32 cores spread across 55 SMs would confer a healthy 10-15% performance bump over the base 4070 while keeping the memory untouched. A price point of $499 to $549 would be ideal. Here’s hoping it doesn’t cost more.
