Nvidia will likely release new GeForce RTX 40XX graphics cards later this year, starting with the RTX 4090.
The most powerful GeForce RTX 4090 was supposed to be released in August, but Nvidia may delay it. Because Nvidia's partners have so many RTX 30XX systems.
The renegotiation of Nvidia's deal for new 5 nm cores from TSMC was a debacle. Only Nvidia's delivery delays will continue.
Until recently, most was known about the flagship RTX-e 4090. Now, a Twitter user with the username Kopite7nymi discloses more facts, including the flagship core clock and the weaker RTX 4080 and 4070.
The flagship will include 126 stream multiprocessors (SM) and 16,128 CUDA cores, based on the AD102 core. 384-bit bus will connect 24 GB of 21,000 MHz GDDR6X memory to the core. The base core clock will be 2235 MHz and can be increased to 2750 MHz, according to the latest leaks.
This is a big improvement, given the current RTX 3090 turbo mode is set to 1695 MHz (not including proprietary versions, which are normally higher).
The AD103-based RTX 4080 would have 80 stream multiprocessors (SM) translating to 10240 CUDA cores and 16 GB of GDDR6X 21 Gbps memory coupled via a 256-bit bus. The 12GB RTX 3080 already features a 384-bit memory bus. Chip TDP is 420 W.
The RTX 4070, the weakest Lovelace-based card, uses the AD104 core. According to published information, this has 56 stream multiprocessors (SM) and 7168 CUDA cores. The memory bus will be 160 bits, while the RTX 3060 has 192. This will connect to 10 GB of GDDR6 18 Gbps RAM with a 300 W TDP.