Earlier @Kopite7kimi had exposed the legendary RTX 4070 Ti specifications, but the new revelation can be directly applied to the RTX 4070 – with the same number of CUDA cores / video memory, but the TGP power consumption is much lower.
It was previously reported that the RTX 4070 may use a ‘full blood version’ AD104 GPU core with 60 sets of SM units / 7680 CUDA cores, supplemented by 48MB L2 cache / 160 ROP raster units.
Compared with the RTX 3070 Ti using the GA104 GPU core, the number of cores has increased by 25%, and the cache has been increased by 12 times.
Although the GPU frequency has not yet been determined, considering the adoption of TSMC’s 4N process (skip 7nm, go straight to 5nm), we expect AD104 to be between 2.0 – 3.0 GHz.
For reference, previous-generation Ampere GPUs usedSamsung8nm process, but it is essentially only a partial optimization based on the 10nm process node.
In terms of video memory, the RTX 4070 is expected to be paired with 192-bit @ 12GB of 21Gbps GDDR6X chips.
In terms of power consumption, the nominal TBP parameter of the RTX 4070 may be reduced from 400W of the RTX 4070 Ti to 300W.
However, this value may be derived from non-public custom SKUs, Founders Editon (public PCB is PG141-SKU331) may be between 255 – 285W.
In the 3DMark “Time Spy” benchmark test, the RTX 3070 is expected to achieve a running score of 11,000 – comparable to Ampere’s flagship RTX 3090 Ti.
As for other RTX 4070 specs, it is also expected to include various modern NV feature sets, such as the latest Gen 4 Tensor / Gen 3 RT cores and the latest NVENC / NVCDEC codecs.
Plus modern RTX graphics features like DLSS, Reflex, Broadcast, Resizable-BAR, Freestyle, Ansel, Highlights, Shadowplay, and G-SYNC.
(screenshot viaWCCFTech)
Finally, given that performance is expected to be on par with the $1999 RTX 3090 Ti, the RTX 4070 series graphics cards are likely to be priced further up from the RTX 3070 series, reaching the $499-$549 range.
