NVIDIA’s RTX 4070 will be the fourth member of the Ada Lovelace family with a mid-2023 launch window. Unlike the first three SKUs, it will feature the same shader configuration as its predecessor. Based on the AD104 core, we’re looking at a core count of 5,888 spread across 46 SMs paired with 12GB of GDDR6 memory. The bus will be 192-bit wide, backed by roughly 32MB of L2 cache.
Reference RTX 4070 has a Base clock of 1920 and 2475 boost
first AIB to finish a 4070 is Colorful with their Ultra W OC line, boost of 2505 MHz
There is also 2 reference models of the 4070
Some additional info regarding the operating clocks of the RTX 4070 has surfaced, courtesy of @T4CFantasy, one of the editors of the TPU GPU database. Matthew claims that the midrange Lovelace part will have a base clock of 1,920MHz and a boost of 2,475MHz. Factory overclocked chips will raise the ceiling to 2.5GHz, with Colorful being the first to be ready with their Ultra W OC lineup. (Pinnaclemontessori.com)
These numbers indicate that the RTX 4070 will be substantially slower than the RTX 4070 Ti. We’re looking at a shader deficit of 1,792 paired with slower GDDR6 memory. The L2 cache will also be 30-35% lower, and finally, the boost clocks have decreased by over 100MHz. This gives us a GPU approximately 25-30% slower than the RTX 4070 Ti at a price point of $549 (speculation).
At the time of writing, the GeForce RTX 3080 is priced at $699, placing it roughly on par with the RTX 4070 (perhaps a little faster) for $150 less. The TBP will also be lower at around 200W, offering notably better performance per watt. NVIDIA is essentially making slightly faster GPUs than their previous-generation parts and selling them at marginally lower prices.