RedGamingTech has provided us with a heavy dose of rumors and very early leaks regarding RTX 50 and a possible Ada refresh. These are rumors and not fully confirmed even by the source itself, so take these leaks with a grain of salt.
RTX 40 Super
As per the rumors, NVIDIA is planning to reinstate the RTX 40 lineup with a slight caveat. Leakers have denied the existence of an upcoming Ada refresh. However, RGT’s claim mentions two specific SKUs, not the entire lineup.
When Kopite verified that Ada Titan was axed, the leaker seemingly mentioned that we may see more variants of the RTX 4070 and the RTX 4060. Based on this, RGT has some specifications of these ‘planned’ SKUs and they are indeed exciting.
The RTX 4060 Super will now make use of the AD104 GPU, a huge leap from the original AD107. A design of 42 SMs equates to 5376 CUDA cores which is 75% more than the original RTX 4060.
Hold on to your seatbelts because it gets better. The VRAM configuration has been upgraded to 12GB, making this 4060 Super an excellent 1440p option. The TDP should remain in the 150–200W range since the 4060 Super is inches away from the RTX 4070, specifications-wise.
The RTX 4070 Super on the other hand packs 8704 CUDA cores, 47% higher than the base model. The additional 4 GigaBytes of memory can enable 4K gaming, although don’t expect RTX 4090 levels of performance. On paper, the RTX 4070 Super is dangerously close to the RTX 4080. Given the price difference, we may see a few price drops if these leaks are true.
Ada-Next GB102 Packs 204 SMs
Since RTX 50 is more than 1 year away, we cannot guarantee the legitimacy of these claims. However, if true, Ada-next is an absolute monster given the raw power it packs.
GB102, the highest-end consumer GPU which will power the RTX 5090 should see a total of 204 SMs. This will be supported with 96MB of L2 cache, similar to AD102. The bus width differs from what we covered a few days back. Although, RGT claims that a 512-bit configuration may be on the cards, allowing the 5090 to ship with 48GB of GDDR7 memory.
Worst-case scenario, over a 384-bit bus, expect 24GB or 36GB of G7 VRAM.
It has been rumored for a long time that Ada-next will bring about a change in the SM structure, the basic unit (for simplicity) in NVIDIA GPUs. As of Ada Lovelace, each SM packs 128 FP32 or CUDA cores. This number can change with Ada-next, nullifying our calculations, so we should keep this in mind before we make performance estimates.
RTX 50 Release Date
The next generation of GeForce consumer GPUs by NVIDIA is set to launch in 2025. NVIDIA has missed its two-year cadence by a small amount. This opens room for a possible refresh and more market adjustments. Do you think NVIDIA plans for an RTX 40 Super refresh? Tell us in the comments.