The GeForce RTX 4070 Super will likely be one of NVIDIA’s bestselling GPUs in the coming months. Based on the AD104 die, it features 7168 shaders across 56 SMs, 22% higher than the base 4070. Previously, the memory bus and cache were listed the same at 192-bit and 36MB, respectively. However, an update on NVIDIA’s update indicates that the L2 cache has also been scaled up to 48MB, increasing the effective bandwidth to 504GB/s.
Following in AMD’s footsteps, the LLC (Last Level Cache) has become a vital part of modern GPUs, alleviating the strain on the memory bus. By storing more and more data on-die, the latency is significantly reduced without wider buses.
Using a large LLC also reduces power consumption by eliminating frequent memory accesses, leaving the bus with more downtime (at least theoretically). Following this update, the RTX 4070 Super looks almost like the 4070 Ti. At 1080p, the two might perform the same. NVIDIA has essentially brought the 4070 Ti down to $599 and the 4080 to $799, with a marginally faster variant at $999. Brings back memories.
