Yesterday we announced the release of the new graphics drivers NVIDIA GeForce 522.25 WHQL Game Readywhich supported the GeForce RTX 4090compatibility to the first games with the technology NVIDIA DLSS 3and still there was room for improve performance in a wide variety of games running under the DirectX 12 API. In addition, these drivers support the RTX 3060 GDDR6X and the RTX 3070 Ti GDDR6X, two “new” GPUs that arrive to complement the existing market.
What the company did not know is that someone would dig into the code of their controllers finding how in the file “nv_dispig.inf” three new GPUs not yet announced were listed. Specifically, we are talking about a “new” GeForce RTX 3070Ti under the code “NVIDIA_DEV.2207”, plus a “new” GeForce RTX 3060 Ti Y GeForce RTX 3060 “NVIDIA_DEV.24C9 and NVIDIA_DEV.2544.”
If you haven’t detached yourself from here for weeks, you already know which GPUs we are talking about, and why so much “new”.
Drivers for RTX 3060 GDDR6X and RTX 3070 Ti GDDR6X
In the case of GeForce RTX 3060 Ti GDDR6X, which was even listed on the NVIDIA UK website. Regarding the original model, there are no changes at the chip level, so we will be facing the same NVIDIA GA104 with 4,864 CUDA Cores152 Tensor Cores and 38 RT Cores at the same frequency Base of 1,410MHz and a slightly higher Turbo frequency of 1,670MHz (vs. 1665MHz). As a result, we have the same FP32 performance of 16.2 TFLOPs.
Let’s go for the only difference, which is in the memory used, replacing the originals 8GB of memory GDDR6 for a few chips GDDR6X which reaches a speed of 19Gbps (versus 14Gbps). Along with the same memory interface of 256 bitwe are talking about a noticeable increase in bandwidth up to 608GB/s compared to 448 GB / s of the original model.
With her little sister, the GeForce RTX 3060exactly the same thing happens, although with a little surprise, and that is that would reduce its memory capacity from 12 to 8 GBto which is added a cut in the memory interface of 192 at 128 bitwhich would reduce the bandwidth from 360 to 240GB/s. This model would underperform the Intel Arc A770so we still do not understand why, unless its price is very attractive.
To finish, regarding the GeForce RTX 3070Tithe only difference is that it would use chips NVIDIA AD102 (from the RTX 3080 / 3090). These would have defects that do not allow their use in top-of-the-range GPUs, but isolating any problem can be reused to bring this GPU to life, which, apparently, would not have any changes at the level of specifications.
Updated list with these new GeForce RTX
|Model||Graphic chip||CUDA’s||Freq. memory||Memory||buses||Bandwidth||TDP|
|RTX 3080||GA102-200||8,704||19Gbps||10GB G6X||320 bit||760GB/s||320W|
|RTX 3070 Ti (GA102)||GA102-???||6,144||19Gbps||8GB G6X||256 bit||608GB/s||N/A|
|RTX 3070 Ti||GA104-400||6,144||19Gbps||8GB G6X||256 bit||608GB/s||290W|
|RTX 3070||GA104-300||5,888||14Gbps||8GB G6||256 bit||448GB/s||220W|
|RTX 3060 Ti (G6X)||GA104-???||4,864||N/A||8GB G6X||256 bit||N/A GB/s||N/A|
|RTX 3060 Ti (GA103)||GA103-200||4,864||14Gbps||8GB G6||256 bit||448GB/s||200W|
|RTX 3060Ti||GA104-200||4,864||14Gbps||8GB G6||256 bit||448GB/s||200W|
|RTX 3060||GA106-300||3,584||15Gbps||12GB G6||192 bit||360GB/s||170W|
|RTX 3060 8GB||GA106-???||3,584||15Gbps||8GB G6||128 bit||240GB/s||N/A|