Despite the news that NVIDIA has yet to finalize the specifications of the GeForce RTX 4070 “Ada Lovelace” GPU, well-known tipster @Kopite7kimi took to Twitter to release two possible sets of parameters. Initially, there were rumors that the RTX 4070 would be paired with 10GB @ 160-bit memory, and soon after “updated” with a 12GB @ 192-bit option, but it seems that both have not been ruled out by the green factory.
The first set of specs says the RTX 4070 has up to 7680 CUDA cores, complemented by 12GB @ 192-bit of 21Gbps GDDR6X memory (the previous generation RTX 3070 had 8GB @ 256-bit of 14 Gbps GDDR6 memory).
The internal SKU code is “PG141-SKU340/341” with a nominal TBP power consumption of 285W – a 50% increase in VRAM and a reduction in memory particles from 8 to 6 due to the doubling of capacity.
The first set of specs says the RTX 4070 has up to 7680 CUDA cores, complemented by 12GB @ 192-bit of 21Gbps GDDR6X memory (the previous generation RTX 3070 had 8GB @ 256-bit of 14 Gbps GDDR6 memory).
The internal SKU code is “PG141-SKU340/341” with a nominal TBP power consumption of 285W – a 50% increase in VRAM and a reduction in memory particles from 8 to 6 due to the doubling of capacity.
The second set of specs says the RTX 4070 has 7168 CUDA cores, complemented by 10GB @ presumed 160-bit wide 21Gbps GDDR6X memory (or 5 VRAM chips). The internal SKU code is “PG141-SKU336/337” and the nominal TBP power consumption is 250W.
The tipster also pointed out that the 3DMark Time Spy Extreme scores for SKU340 / 341 are expected to exceed 11,000, while SKU336 / 337 are designed to reach at least 10,000 points.
(Screenshot via VideoCardz)
In summary, the second configuration is likely to be the successor to the RTX 3060 Ti “Ampere GPU”. However, given that NVIDIA usually releases new products from high to low, it is expected that the RTX 4070, which is targeted at the performance segment, will not arrive until late 2022, or early 2023.
