According to Twitter leaker Kopite7kimi, NVIDIA's new flagship graphics card, the AD102, will feature up to an insane 75 billion transistors. Compared to the GA102, the AD102 will offer 2.65 times more transistors than the previous generation.
NVIDIA Goes Crazy With Transistor Density on 5nm, Ada Lovelace AD102 GPU Expected To Pack Over 75 Billion Transistors
The NVIDIA AD102 GPU based on the Ada Lovelace architecture will offer five billion less than the company's Hopper H100 (80 Billion transistors), the company's data-center CPU utilizing the same TSMC 4N node. While the size of NVIDIA's new AD102 GPU is still a mystery, it isn't easy to compare the new GPU with the Hopper and Ada series since one chip is designed for the gaming segment and the other is designed purely for the server (AI / HPC) market.
It is known that both processors will offer a significantly sized L2 cache. The L2 cache for the Hopper H100 GPU is 50 MB, and the AD102 GPU is expected to pack up to 96 MB of cache. This information may suggest why NVIDIA needed the extra transistors for the AD102 GPU.
Remember this. 😁Most people made the right choice. More than 75B.
— kopite7kimi (@kopite7kimi) September 13, 2022
The transistor count has not been officially revealed, but the leaker on Twitter guaranteed that the count would be above 75 Billion. We will probably need to be patient until the company's announcement next week during the GTC 22.
The new NVIDIA AD102 GPU will offer 18432 CUDA cores, but this will not land during the first installment of releases for the GeForce RTX GPUs. The GPU will feature some fast GDDR6X, supporting a bus width of 384-bit. Right now, only one graphics card has revealed itself: the GeForce RTX 4090 GPU, offering 16384 enabled cores. If rumors are true, the company will include a possible GeForce RTX 4090 Ti or maybe a new TITAN GPU.
About three months ago, rumors about the reference card leaked, stating the cooler design would offer a four-slot design and triple fans mounted within the front shroud of the GPU. The rumored design of the GPU is reminiscent of custom GPUs from other manufacturers for the GeForce RTX 3090 Ti that offers an identical design. With most of these custom designs being "air-cooled," the company is unlikely to add liquid cooling unless the TGP surpasses 600W.
After NVIDIA's announcement at GTC next week, the new next-gen graphics card series will release in the last quarter of this year. The new series will compete with AMD and Intel with the company's discrete lines.
|NVIDIA Flagship Gaming GPUs|
|Node||TSMC 4N||SAMSUNG 8N||TSMC 12nm||TSMC 12nm||TSMC 16nm|
|Die Size||TBC||628 mm²||754 mm²||815 mm²||471 mm²|
|Trans. Density||TBC||45.1M TRAN/mm2||24.7M TRAN/mm2||25.9M TRAN/mm2||25.5M TRAN/mm2|
|Memory Bus||GDDR6X 384-bit||GDDR6X 384-bit||GDDR6 384-bit||HBM2 3072-bit||GDDR6X 384-bit|
News Source: Videocardz