The next generation from NVIDIA may provide more good reasons to hesitate than before for consumers
The stocks of the current generation models are still loading the shelves of stores around the world, but there is no doubt that the next generation of GeForce RTX cards from NVIDIA is already on its way – with possible teasers for the time when we will get the first significant details about it and especially a huge amount of leaks that tell us about different versions and modifications that we expect in the framework of its senior models and intermediate models.
At the top end, the RTX 4090 stands with an AD102 chip and a huge amount of processing units when it seems that there is a wide agreement among those who deal with technological rumors and prophecies that there will be use of 128 processing clusters with 16,384 CUDA cores and 24GB of built-in dynamic memory similar to models The elite of the current generation, although with a higher operating frequency that will provide a bandwidth of up to 1.15TBps – but in all three levels below the RTX 4090 there seems to be quite a bit of conflicting information about the hardware that will be offered, where as time passes the chances increase that we will eventually see some Different hardware configurations under the same model name, as in the current generation.
In the RTX 3000 generation, we received a pair of versions of the GeForce RTX 3080, with different volumes of dynamic memory and also different amounts of active processing units, with the 12GB version (launched at a later stage) being closer in performance to the GeForce RTX 3080 in some cases Ti confirms the original version with 10GB of GDDR6 – and it is possible that in the RTX 4000 generation we will have a pair of different GeForce RTX 4080 models right from the opening whistle.
According to current information, the RTX 4080 will also be based on the AD103 core with 76 active processing clusters (out of 84 clusters on the silicon chip) and 9,728 CUDA units combined with 16GB of GDDR6X memory at an impressive 23Gbps speed – and on the physically smaller AD104 core With 60 active processing clusters, 7,680 CUDA units and also 12GB GDDR6X memory at 21Gbps speed and a narrower 192-bit interface. It is not clear why NVIDIA can decide to give two cards with completely different characteristics and the potential for performance differences of over 20 percent the same model name, but after we have already received an illustrative precedent in the current generation, it is probably impossible to dismiss this option outright.
The same AD104 core may form the basis of a pair of GeForce RTX 4070 models in addition to the RTX 4080 12GB, with the senior of them likely to benefit from 56 available processing clusters out of 60 physically present on the core (which means that the RTX 4080 12GB will challenge NVIDIA and -TSMC and will require perfectly active cores, which may certainly be a big challenge in terms of output with the transition to a new and innovative 4 nm manufacturing process), plus 10GB of GDDR6X at a basic 21Gbps speed and a 160-bit interface that will provide an effective bandwidth of approximately 420GBps.
At the bottom end, there are the first bits of information about the AD106 and AD107 cores with 36 active processing clusters and 24 active processing clusters respectively – with one of the main question marks here regarding the chances that these chopped bits of silicon will also be based on the same advanced N4 process, or will they settle for the process Less advanced and above all cheaper for NVIDIA and its partners than TSMC or maybe Samsung.
NVIDIA’s annual online GTC event starting in a few days may provide us with the first important official details about the new generation of graphics cards – so we hope to provide you with more juicy updates on this topic shortly.