NVIDIA is prepping the launch of its GeForce RTX 4060 graphics card. Based on the AD107-400-A1 GPU die, it will feature 3,072 FP32 cores across 24 SMs, a notable drop off the 3,584 (28 SMs) cores on the RTX 3060. The GPU will be paired with just 8GB of GDDR6 memory clocked at 18Gbps via a 128-bit bus. The memory will be backed by 24MB of L2 cache, and the graphics card will have a TBP of 115W.
RTX 4060 still uses PG190.
8G GDDR6 18Gbps
I will try to remain neutral about any leaks in the future. 😁😁😁Originally tweeted by kopite7kimi (@kopite7kimi) on February 13, 2023.
In comparison, the GeForce RTX 3060, based on the Ampere graphics architecture, featured 6GB and 12GB variants clocked at 15Gbps memory. As with other last-gen GeForce GPUs, the L2 cache was a paltry 3MB, albeit paired with a larger 192-bit bus for an external memory bandwidth of 360GB/s.
The RTX 4060 focuses on on-die bandwidth alongside a slimmer bus width, offsetting it with faster memory packages and a larger on-die buffer. One area where the budget Lovelace GPU excels is power efficiency. From 170W on the RTX 3060, the RTX 4060 cuts down the peak board power to just 115W. We expect it will hit retailers in the second half of 2023.