@kopite7kimi, one of the most reliable and influential NVIDIA leakers, has revealed the specifications of NVIDIA’s upcoming GeForce RTX 4070 graphics card. In recent years, Kopite was the first to reveal the existence and specifications of nearly all NVIDIA GPU families, so we have little doubt that this leak is also accurate. We have not only the CUDA core counts but also many more details, so let’s get started.
The RTX 4070 Ti and RTX 4070 are both based on the AD104 GPU, with the 4070 being a binned form of the RTX 4070 Ti GPUs. For those unfamiliar with binning, all GPU manufacturers create the higher-spec-ed SKU, but not every SM or CUDA core is functional by the end of the lithographic process.
As a result, the perfectly working dies are sold as the higher-end SKU, while the dies with less functional sections have the problematic bits lasered off or deactivated with straps and are offered as a lower-segment product. This is how AMD and Intel function, and it is a component of the yield sciences involved in semiconductor manufacturing.
The RTX 4070 will be a scaled-down AD104 die with the IDs PG141-SKU336 and PG141-SKU337, according to Kopite7Kimi. The precise GPU variant name will be AD104-250-A1. The GPU will have the same transistor count as the RTX 4070 Ti, however, it will only have 5888 CUDA Cores enabled over 46 SMs. It will have the same VRAM as the 4070 Ti, with 12GB of GDDR6X memory and the same 192-bit bus width.
The memory will also be clocked at 21 Gbps, bringing the total bandwidth to 504 GB/s. This will be NVIDIA’s first card with a TGP of 250W as the default.
The die size will be the same as the RTX 4070 Ti, 295mm2. The entire NVIDIA Ada Lovelace series is built on TSMC’s 4nm technology.
We can now confidently reverse engineer the anticipated FP32 performance of this card based on the core count parameters supplied by Kopite. The card should have 30.7 TFLOPs of nominal FP32 performance (5888 CUDA Cores * 2 * 2.610 GHz = 30.7 TFLOPs). Because this is a card from the same IHV and family, we can utilise FP32 performance to directly compare rasterization performance between GPUs, and based on this, it appears to be approximately 76% as fast as an RTX 4070 Ti.
Of course, performance is only as important as pricing, and depending on how NVIDIA markets this, it may become a crowd favourite. Based on NVIDIA’s pricing philosophy thus far (and assuming no price reduction), the NVIDIA RTX 4070 might have an MSRP of roughly $699, albeit this is purely speculative on my part.
Also Read:
