It's not the first time either or it looks like it will be the last time the boys come out Finalwire We surprised us by introducing a new product to the market. In this case and after the silent auction where they are interrelated and related, the latest from NVIDIA will not be, as expected, either an RTX 2080 Ti version or a new design.
In contrast, NVIDIA didn't want to miss the low number of GPUs and it looks like the direct competition of the AMD RX 5500 to be sold initially by OEMs would be ready.
NVIDIA GTX 1650 Ti: will it be able to dominate before its rivals?
The fact is that there are no more details than what the chip used will be TU117, so from here we are going to be thinking about what we can get. The inclusion of this chip marks NVIDIA's obvious tendency toward a card that should, with benefits, be placed within the GTX 1650 SUPER and GTX 1650.
This GTX 1650 Ti will be released outside of the TU116 which features a SUPER version and in which NVIDIA marks the spot where it intends to compete. And that between the two cards currently on the market there is a huge difference depending on the model chosen, 35% on average, something that leaves the right soil to include this GTX 1650 Ti in between.
What we can get with this GPU chip TU117-400-A1 that would do 4.7 billion transistors and size of 200 mm2 produced by TSMC at 12 nm under buildings Exercise. Although we do not know the approaching waves, we can think of their SM number.
Possibly, I will come 16 SM and a complete calculation of 1024 Shaders and moving estimates of 1,700 MHz. Logic tells us it will eventually 32 ROPs, but they would see their own NTUs until 64., maintains its 64 KB L1 by SM and a complete L2 repository of 1024 KB.
I can skip GDDR6 to achieve maximum performance
Although its lower version carries GDDR5, it's likely that since this type of memory has officially installed EOL and GDDR6 offers great performance, be it NVIDIA's second choice of the new GTX 1650 Ti.
The bus would stay inside 128 bytes and the speed is more than likely to be in the end 12 Gbps, which will give us the bandwidth of 192 GB / s. Its use instead would not be very clear, as it would exceed 75 watts without reaching 100 SUPER version.
It will all depend on the clock speed of NVIDIA 6 food pins. In the end, all of these predictive information can generate performance in FP32 from all around 3.7 and 3.8 TFLOPS, which would mean winning the RX 5500 without any problems and staying above 10% of the RX 5500 XT.
We'll have to wait to see if, in fact, our estimates are correct, but nonetheless, NVIDIA doesn't give it a low list.
