And it is that a brief specification sheet has been leaked where said graphics card has been named and specified with some information of interest. As we say, what should be the RTX 3090 SUPER will not be called that and would enter for the first time in history as a Ti series and its changes are going to be very important.
NVIDIA RTX 3090 Ti, the most extreme monodie GPU ever
Extreme graphics cards have been and will be, as manufacturers always try to push everything to the limit, but this GPU will be something else and does not seem to have much room for maneuver. Although the launch date remains for January 2022, what is really interesting is not this but its characteristics that have put this data aside.
It is true that it will have the GA102 chip in full and fully unlocked as we anticipated a week ago and therefore it will remain within the Ampere architecture, but its news will boost its performance much higher than we thought. Starting with the most noteworthy in this section: new 21 Gbps GDDR6X memory, which will mean a greater bandwidth and evidence that the current RTX 3090 was at the limit in this regard.
Bandwidth is essential to supply data and textures to the die, so more muscle requires more bandwidth and since the bus cannot be increased since it would require a new chip and design, the most viable option has been to increase the speed of the memories based on new chips that will now be 2 GB per drive.
PCB redesign, higher power consumption, new power connector and PCIe 5.0?
The new modules will imply a greater consumption of the card, which due to the unlocking and at this point would go to the 450 watts, being the most “greedy” monodie GPU ever. The new 21 Gbps GDDR6X modules mean that NVIDIA could now choose to install them on one side of the PCB rather than two, which could lower the temperature by better dissipating them with air from the fans.
High consumption and high speeds are going to cause two things: a redesign of the PCB and a new power connector, which will now be a 16-pin Micro-Fit I should bear a whopping 800 watts (Be careful with the PSUs here, they are more than important figures for the peaks). Due to the energy consumption and the high temperatures that VRAMs must have, the PCB has to be new in order to adapt to the new energy requirements.
We understand that the aesthetics will be similar seen from the outside, but the changes introduced must be of a certain depth, although nothing has been confirmed. Finally, it is speculated that with the new 16-pin connector NVIDIA could launch this card being PCIe 5.0, since there are rumors that this connector will be in charge of accompanying the new bus.
Therefore and summarizing, the NVIDIA RTX 3090 Ti will have a GA102 with 82 SM and 10752 Shaders, 24 GB from GDDR6X at 21 Gbps, 84 RT Cores, 336 TMU, 128 ROPs and 336 Tensor Cores for a bus of 384 Bits and a bandwidth of 1008 GB / s. Will AMD be able to answer this? Will you trust RDNA 3 and give up the performance crown?