NVIDIA GeForce RTX 3050 Ti Laptop GPU vs NVIDIA GeForce GTX 1650 Ti Mobile vs NVIDIA GeForce RTX 3070

The Nvidia GeForce RTX 3050 Ti Laptop GPU (for laptops, NVIDIA_DEV.2583, GN20-P1) is a faster variant of the RTX 3050 mobile card and based on the GA107 Ampere chip. The chip offers 2560 CUDA, 20 Ray Tracing and 80 Tensor cores. The clock rate depends on the used TGP-version (from 35 - 80 Watt).

The performance (of the high TGP variants) should be similar to the old RTX 2060 and therefore most suited for 1080p gaming with high detail settings. 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 a performance boost with a slight quality reduction). Beware, that the low TGP variants will offer a significantly lower performance.

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 Ti 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 Ti Mobile is a dedicated mid range graphics card for laptops based on the Turing architecture. It uses the same TU117 chip as the desktop version but for the mobile GPU all 1024 shaders are enabled. Compared to the faster RTX 2000 GPUs (e.g. RTX 2060), the 1650 Ti integrates no Raytracing or Tensor cores. 

There are two packages of the GTX 1650Ti, the GB4D-128 with a 128 Bit memory bus and a TGP of 50 Watt and the GB4B-256 with a 256 Bit memory bus and a TGP of 55 Watt (compatible with the faster RTX cards). The clock rates seem to be identical for both versions, therefore, the 256 Bit version should be faster.

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 Ti for laptops is specified at 50 - 80 Watt TGP (Total Graphics Power) by Nvidia and therefore 5 - 25 Watt higher than the GTX 1650. The power efficient Max-Q variant of the GTX 1650 Ti is rated at 35 Watt TGP.