NVIDIA GeForce RTX 3050 4GB Laptop GPU vs NVIDIA GeForce GTX 1650 (Desktop)

The Nvidia GeForce RTX 3050 Laptop GPU (mobile, NVIDIA_DEV.2583, GN20-P0) is the slowest RTX 3000 mobile card and based on the GA107 Ampere chip. It offers 2048 CUDA, 16 Ray Tracing and 64 Tensor cores. The memory bus is limited to 128 Bit and supports 4 GB of GDDR6 dedicated graphics memory.The clock speed depends on the TGP variant and can range from 713 - 1530 (base) and 1058 - 1740 (boost) for the TGP variants of 35 to 80 Watt (see table below).

In early 2023 Nvidia did refresh the RTX 3050 (GN20-P0-R) with a bigger VRAM of 6GB but reduced memory bandwidth. The 3050 6GB Laptop GPU also offers 2,560 CUDA cores as the 3050 Ti mobile.

The performance should be between the old GTX 1650 Ti and the 1660 Ti Max-Q depending on the used TGP variant. Therefore, it is best suited for full HD gaming (1920 x 1080) with high detail settings in demanding games. The performance is not sufficient to enable Raytracing in most games, but the Tensor cores can be used for DLSS in some games (and get performance boost with slight quality reduction). Beware, that the low TGP variants will offer a significantly lower performance.

The faster RTX 3050 Ti offers 25% more cores at similar clock speeds and TGP settings. 

The GA107 chip offers 3,072 FP32 ALUs of which half can also execute INT32 instructions (i.e. 1,536 INT32 ALUs). The RTX 3050 however maybe won't use all ALUs on the chip. With Turing all shaders could still execute FP32 or INT32 instructions. The raytracing and tensor cores on the chip were also improved according to Nvidia. The Ampere chips also include an improved 5th generation video encoder (NVENC for H.264 and H.265) and a 7th generation decoder (for various formats now including AV1).

The GA107 chip is manufactured by Samsung in 8nm (8N), which is not quite able to keep up with the 7nm node at TSMC (e.g. used by AMD and also for the professional GA100 Ampere chip).

The Nvidia GeForce GTX 1650 for desktop PCs is a graphics card that is based on the Turing architecture (TU117 chip). Compared to the faster RTX 2000 GPUs (e.g. RTX 2060), the 1650 integrates no Raytracing or Tensor cores. The performance should be slightly faster than the old GeForce GTX 1050 Ti. Although, the mobile variant of the GTX1650 uses the same name, it is a bit slower due to lower clock rates.

The Turing generation did not only introduce raytracing for the RTX cards, but also optimized the architecture of the cores and caches. According to Nvidia the CUDA cores offer now a concurrent execution of floating point and integer operations for increased performance in compute-heavy workloads of modern games. Furthermore, the caches were reworked (new unified memory architecture with twice the cache compared to Pascal). This leads to 50% more instructions per clock and a 40% more power efficient usage compared to Pascal.

Compared to the bigger Turing chips (like the TU116 of the GTX 1660 Ti and the RTX lineup), the TU117 does not include the new NVENC encoder but an older one similar to the one used in Pascal and Volta.

The power consumption of the 1650 is specified at 75 Watt TGP (Total Graphics Power) by Nvidia. The TU117 chip is manufactured in 12nm FFN at TSMC.