They are rumors and leaks, yes, there is nothing confirmed, okay, but if we stick to the sources from which they arrive, there are reasons to worry if what was said finally comes true. And it is that although AMD and NVIDIA are going to compete on equal terms thanks to TSMC’s 5 nm node, but… All that glitters is not gold, as we warned months ago.
Over 400 watts on a GPU and with a new lithographic process
You have to go back a few years to dual GPUs to see how a single PCB could consume a huge amount of watts with the graphics card fully standard. At that time we were talking about a covert SLI or CrossFire, now with AMD we have to talk about the first MCM GPU in history for desktop (the Instinct MI200 is the first in history officially).
Meanwhile, the data that we are going to offer from NVIDIA corresponds to a monolithic GPU, but with monstrous performance like AMD. To be specific, we are talking about Navi 31 and AD102, which have already undergone the final touches in their architecture and design, so the engineers have the data on the table, fresh, and at the same time not very satisfied with the data. of consumption.
To give us an idea of the dimensions of the commented chips, it is rumored that Navi 31 would be a dual core / matrix GPU, with a few 600/650 mm2, all without counting the MCD. So it is estimated that the total size of the GPU itself would be around 800 mm2, a real monster.
NVIDIA would win the consumer battle again
In contrast, the AD102 being a monolithic “traditional” GPU would have an already more common size of 600 mm2. The problem in consumption is not the size logically, but what can be included in this area with a lithographic process as advanced as that of TSMC 5 nm.
We talk about 15360 Shaders for Navi 31 and nothing less than 18432 Shaders for AD102, which according to the leaks would imply about 450/480 watts for the AMD GPU and about 420/450 watts for the NVIDIA GPU.
Logically the question is, how do you hope to dissipate such a quantity of heat with an area that has not grown too much? At the moment no image of the heatsink of either of the two cards has been leaked, but if they were already throwing the house out of the window with these latest designs, what remains to be seen must undoubtedly be the panacea.
The downside of this is that considering the price of TSMC’s 5nm wafer (rumored to be over $ 17,000 right now) and the incredible heatsink they must have (R&D, materials and qualities), what is certain is that they will not be cheap, although they supposedly double the performance compared to the RTX 3000 and RX 6000.
For now, let’s have a little salt and be patient, because there is still almost a year left until its launch, at least in theory.