GeForce GTX 16 series prototype has PCIe 4.0 enabled
What you see above is a prototype TU116 (Turing) model, likely from 2018, as indicated by the GPU die information. This is an engineering sample, meaning it’s not a retail card, even though it resembles the RTX 2060 series. As we remember, the first-generation GeForce cards with RTX features were divided into TU200 and TU100 series. The latter lacked Tensor and RT cores, which is why they were released as the GTX 16 series.
The prototype has no markings, so the exact chip is unknown, but it appears similar to TU116. It’s likely an early version of the GTX 1650 SUPER, a mid-range card that would not be a practical choice for a 2025 system since it only supports 4 GB of memory. We can identify it as a GTX 1650 because of the memory configuration and the presence of a TU116 GPU. The only other 4 GB models in the GTX 16 series were the GTX 1650 non-SUPER and GTX 1630, both based on TU117. The only major change in specs is 1152 CUDA cores , instead of retail’s 1280.
Source: GOKForFree
While this sample does not function, it reportedly carries the PG160 SKU 18 marking, whereas the retail GTX 1650 SUPER used PG166 SKU 24. The prototype reportedly had an interesting difference from the production version—it supported the PCIe 4.0 x16 interface. All retail GeForce Turing cards, however, were limited to PCIe 3.0.
Source: GOKForFree
This leak also confirms that the never-released reference design indeed resembled the RTX 2060 series. TU116 and TU117 cards were exclusively produced by board partners, so official reference designs were never publicly released beyond a few renders.
Source: GOKForFree
