A more powerful GeForce RTX 4080 is rumored to debut in early 2024. It is anticipated to use a scaled-down version of the AD102 GPU featured in the RTX 4090, though the official moniker, Super or Ti, has not yet been verified. It is said to be comparable in price to the RTX 4080 and has a power limit of less than 450W. According to Videocardz, MEGAsizeGPU posted these specifics on Twitter.
The AD102, which now powers Nvidia’s top-tier GPU, the AD103, is reported to be used in the rumored GeForce RTX 4080 model. A maximum of 80 SMs with shaders, tensor cores, and ray tracing units make up the AD103. It has a 256-bit memory bus width. The current RTX 4080 uses a variation of the AD103 with only 4 SMs turned off, allowing the remainder of the chip to function properly. This restricts the potential benefit of designing a speedier product using this chip because an AD103 that is fully enabled and operating at the same clock rate as the RTX 4080 would only slightly increase speed.
The AD102, however, has greater potential. The maximum number of SMs it can support is 144, and its total bus width is 384 bits. There is already one with 128 activated SMs in the RTX 4090. Nvidia might choose an AD102 chip with about 100 SMs for a hypothetical RTX 4080 Ti.
With a clock speed comparable to the RTX 4080 but much less power consumption, this chip is currently in use in the RTX 5000 Ada Generation workstation card. The Ti variant might potentially provide up to a 25% improvement in performance over the base RTX 4080 if it features a 320-bit memory interface and 20GB of VRAM. As a result of the RTX 4080’s much higher price compared to earlier 80-class devices, Nvidia hasn’t had as much success with it. The RTX 4080 was $400 more expensive than the GeForce RTX 3080, which had a $799 MSRP at launch.
Except for the RTX 2080, which included both Ti and Super, the Super moniker was only ever used once in the Turing series. If Nvidia reintroduces the Super name, it could potentially confuse consumers who already own a Turing-based Super model and are seeking an upgrade.