NVIDIA AD102 GPU will be laced with 75 billion transistors. Yes, you heard it right, because according to leaker Kopite7kimi, the new flagship GPU named AD102 will have more than 2.6 times as many transistors as GA102. With that many transistors, the AD102 GPU will be just 5B less than Hopper H100 GPU, NVIDIA’s flagship data center processor, which is also based on the comparable TSMC N4 node.
The size of the NVIDIA AD102 GPU is still unknown, although a direct comparison with Hopper and Ada is a distant possibility. Both processors have a large L2 cache (50 MB for Hopper and up to 96 MB for Ada). That’s about twice the size the AD102 GPU explains where those transistors were used.
While Kopite did not confirm the exact number of transistors, it confirmed that the number is more than 75B. And now, just a week after the GTC, the official announcement will only confirm the exact number everyone is looking for.
The NVIDIA AMD 102 GPU will feature up to 18432 CDA cores, although this won’t be the case with the first wave of GeForce RTX GPUs. With this processor, GDDR 6X memory is available with support for a bus width of up to 384 bits. The RX4090 is the only other model known to have this feature with 16384 cores enabled.
Remember this. 😁Most people have made the right choice. More than 75B.
— kopite7kimi (@kopite7kimi) September 13, 2022