The range of graphics cards RTX40 from Nvidia will benefit from a new architecture Lovelace and everything seems to indicate that it will seriously increase in power at all levels.
The leaker kopite7kimi continues to make the presentations by specifying the characteristics of each model planned. The star will of course be the GeForce RTX 4090 with its AD102-300 GPU of 16,384 CUDA coresits 96 MB of L2 cache and its 24 GB of GDDR6X memory at 21 Gbps and 384-bit memory bus for a bandwidth of 1 TB/s.
Supposed rendering of the RTX 400 Ti (credit: Moore’s Law is Dead)
It specifies that the base frequency will be 2235 MHz with a boost rate fixed at 2520 MHz but which may reach up to 2750MHz if necessary (vs 1395 MHz in base and 1695 MHz in boost for the RTX 3090), allowing it to reach a processing power of the order of 90 TFLOPSwhich is twice the RTX 3090 Ti under Ampere, with a TDP of 450W.
Still unknowns in the characteristics
Next to it, the RTX4080 will use an AD103-300 GPU but with 10,240 CUDA cores and 16 GB of GDDR6X memory at 21 Gbps, with a bus reduced to 256-bit and a TDP of 420W.
The frequencies are not known and it should provide a memory bandwidth of 672 GB/s.
Finally, the RTX4070 will be lighter with an AD104-275 GPU with 7,168 CUDA cores while the memory will consist of 10 GB of GDDR6 at 18 Gbps with 160-bit bus, i.e. around 360 GB/s of bandwidth, all with a TDP of 300W.
The launch of the GeForce RTX 40 range, initially expected during the third quarter, could be postponed to the last quarter, one per month from October to December, with a RTX4060 which would make its appearance in January for the CES 2023 show in Las Vegas.
.
The post a frequency well above 2500 MHz? appeared first on Gamingsym.