NVIDIA’s GeForce RTX 2050 is one of the stranger SKUs of the Ampere family. Although it uses the 20 series nomenclature, it is based on the Ampere microarchitecture, the same as the RTX 30 series linup. One of the key features of the RTX 2050 is being the first x20-class SKU to integrate both RT and Tensor cores, all the while having a TBP of just 35-40W. On the downside, it pairs the GA107 die (2,048 FP32 cores) with a flimsy 64-bit bus to just 4GB of GDDR6 memory.
The GeForce MX570 is another GPU based on the GA107. With a slashed memory budget and disabled RT, Tensor cores, it has a power draw of 15-25W. The MX550 is a similar case. Featuring the TU117 core from the Turing family, it comes with 1,024 shaders and a 64-bit bus paired with 2GB of GDDR6 memory. (Zolpidem)
As per the latest 3DMark FireStrike scores, the GeForce RTX 2050 is 2.3x faster than the MX450 with a score of 9,679. Compared to the GeForce RTX 1650, we’re looking at a lead of 23%. Keep in mind we’re talking about all the mobile variants here.
Going by these figures, the Radeon 680H integrated GPU on the Ryzen 7 6800U processor is roughly on par with the GeForce RTX 2050, or at the very least, comparable. As per AMD’s slides, the 6800U’s iGPU with 768 shaders is up to 2x faster than the MX450 which puts it on par with the RTX 2050.
Leveraging the RDNA 2 graphics architecture (including ray-accelerators), the Rembrandt APUs offer impressive graphics performance. The Radeon 680M graphics core is up to 2x faster than the preceding Vega-based Cezanne offerings in gaming workloads.
Benchmark source: ITHome