Something for over a year has been with us NVIDIA GTX 1650, the time when as we have provided 4 different models according to the company it is necessary to update them for various reasons. Released with GDDR5 and TU117 on hand, upgraded to GDDR6 keeping the same chip and subsequently to TU116 and finally TU106.
The NVIDIA GTX 1650, without any noticeable change in performance, is one minus
NVIDIA has completely calibrated the functionality of its different models, so that we have what we have, benefiting from the same FPS. As we saw at the time, performance improvements were due first to the competitive advantage of the RX 5500 XT and second, to having a single multi-model chip, thus increasing the benefits per machine and simplifying the use of matrices there.
However, although it did not affect the card performance itself, it did affect the adjacent category: NVENC entries. Different chips carry different types of chips and this has been one of the most controversial features of the card. Why was the Taring GPU for the structures it was extracted from Volta?
To put it in context, i TU117 It was the only chip that relied on the old Volta installer, while the GTX 1650 models included TU116 or the highest TU106 they already have a rating-based desoder installed
VVta’s NVENC works in the same way as Pascal’s
GeForce GTX 1650 TU106 | GeForce GTX 1650 TU116 | GeForce GTX 1650 GDDR6 | GeForce GTX 1650 | |
---|---|---|---|---|
GPU | Test (TU106) | Test (TU116) | Test (TU117) | Test (TU117) |
NVENC encoder | Exercise | Exercise | Time | Time |
Shaders | 896 | 896 | 896 | 896 |
NTUs | 56. | 56. | 56. | 56. |
ROP | 32 | 32 | 32 | 32 |
Clock Base | 1,410 MHz | 1,410 MHz | 1,410 MHz | 1,485 MHz |
Increase Clock | 1,590 MHz | 1,590 MHz | 1,590 MHz | 1,665 MHz |
Memory Clock | 12 Gbps | 12 Gbps | 12 Gbps | 8 Gbps |
VRAM | 4GB GDDR6 | 4GB GDDR6 | 4GB GDDR6 | 4GB GDDR5 |
Bus | 128 bytes | 128 bytes | 128 bytes | 128 bytes |
The bandwidth | 192 GBps | 192 GBps | 192 GBps | 128 GBps |
L2 | 1 MB | 1 MB | 1 MB | 1 MB |
TDP | 75W | 75W | 75W | 75W |
Transistors | 10.8 billion | 6.6 billion | 4.7 Billion | 4.7 Billion |
Location | 445 mm² | 284 mm² | 200 mm² | 200 mm² |
According to NVIDIA itself, the Volta-based NVENC performs the same as Pascal’s, but not the test-based ones, from Huang’s writing to the improvement of this until 15% efficiency, not to mention the improvements included in the design itself.
So, and although many manufacturers specify it as such, there are some who make no effort beyond calling the GTX 1650 model otherwise. If you are willing to buy one of these GPUs, the good thing is that you get yourself well informed about the model in question because of the above.
A good indication would be that it includes GDDR6 and not GDDR5, so the first cut will pass, because in the worst case scenario TU117. The last point that resolves the doubts is more complex, because we will have to know the calculation of the transistor or matrix size. If you have 6.6 billion or more transistors and the chip has room 284 mm2After that we will make sure that NVENC works basically.
Unfortunately, the only way to know without the manufacturer’s specifications is clear, because the other features are the same, blurring the differences between GDDR6 and GDDR5, and the frequencies logically.