Two NVIDIA GPUs take center stage: the AD106 or the AD107, as you can use either of them. There are already rumors that the AD106 is intended for a future Ti model, but it could also be used for the notebook models. For reference, we have the GA107, which gave life to the desktop RTX 3050 and didn’t perform well enough in ray tracing to recommend this GPU.
NVIDIA RTX 4050, the specifications you can expect to see
Well, the NVIDIA RTX 4050’s manufacturing process could be TSMC’s 4N, which is an optimized version of Taiwan’s 5nm, whose GPU would have a surface area of 150mm². This AD107 can have up to 3 GPCs consisting of 4 TPCs and 8 SM (streaming multiprocessor). Each SM will have 4 sub-cores, but what has changed is the FP32 and INT32 configuration:
- Each sub-core will have 128 FP32 units.
- Together with the INT32 units, the total number is 192.
Adding up the ends (24 SM -> 8 per GPC) results in a total number of 4608 CUDA cores, which are distributed as follows:
- 3072 FP32 units.
- 1536 INT32 units.
Another new feature is the increase in cache memory compared to the RTX 3000: about 50%. So, we can expect an increase to 16 MB of L2 cache, considering that the RTX 3050 was equipped with 2 MB of L2. Beyond that, we’ll see a 50% increase in ROPs, which is expected to be a big increase when combined with the 4th generation Tensor Cores and 3rd generation RT Cores.
Regarding the VRAM memory, if nothing changes, we will continue with the 8 GB GDDR6 at 20 Gbps, which would use a 128-bit bus, giving a bandwidth of about 320 GB/s. In this sense, it improves memory speed and bandwidth.
The price to pay for all this would be a higher power consumption than the RTX 3050: about 150 W TGP.
Introduction and pricing
And this is where the bad news comes in. As it stands today, the NVIDIA RTX 4050 is expected to launch in 2023 at a starting price of around $250. This means that all ranges confirm an increase of 100 euros, as these GPUs used to be available for significantly less.