New indiscretions around the GeForce RTX 4090 and GeForce RTX 4070 are encouraging. Their specifications should allow a jump in performance compared to the current GeForce RTX 3090 and 3070 Ampère.
This information provided by Kopite7kimi focus on two references”Ada Lovelace“, the GeForce RTX 4090 and RTX 4070. The main difference between these two cards will be at the GPU level. The first will use a GPU based on the AD102 while the RTX 4070 will use a chip from the AD104 GPU.
GeForce RTX 4090 and 4070, a performance leap
The GeForce RTX 4090 is supposed to use the same GPU as its big sister “Ti” but in a slightly less muscular configuration.
The AD102 GPU should be equipped with 144 SMs, an increase of 71% compared to the current existing GA102. It would bring 18,432 CUDA cores to life. Ada Lovelace would also offer a larger L2 cache. Thus the AD102, the flagship GPU would be equipped with a maximum of 96 MB of L2 cache (the L2 cache of the GA102 is 6 MB).
The card would be equipped with 24 GB of GDDR6X memory calibrated at 21 Gbps operated through a 384-bit bus enough to ensure a bandwidth of more than 1 TB / s. All this suggests that the TBP will reach 600W for the so-called “reference” version. Some custom models will likely be even more power hungry. It is not impossible for cooling systems to evolve into a hybrid version combining AirCooling and Watercooling
On the GeForce RTX 4070 side, the AD104 would carry 60 SMs or 7680 CUDA cores, all accompanied by 48 MB of L2 cache. The equipment should offer 12 GB of GDDR6 calibrated between 16 and 18 Gbps and operated through a 192-bit bus. The bandwidth should thus reach 400 GB / s. The TGP would be around 300W. Compared to the GeForce RTX 3070 Ti whose TGP is 290 W, the increase would be limited for performance at the height of a GeForce RTX 3080 Ti whose TGP is 350 W.
The “Ada Lovelace” generation is normally expected for the second half of 2022.
GeForce RTX 40 Series lineup – Rumors
|Model||GPU||SMs / CUDA Core||VRAM / Bus||VRAM Frequency / Bandwidth||TGP||PCIe slots|
|GeForce RTX 4090 Ti||AD102-350||144 / 18432||48 Go/ 384-bit||24 Gbps / 1.15 To/s||~900W||2x 16-pin|
|GeForce RTX 4090||AD102-300||>144 / 18432||24 Go / 384-bit||21 Gbps / 1.00 To/s||~600W||1x 16-pin|
|GeForce RTX 4080||AD103-300||>84 / 10752||16 Go / 256-bit||21 Gbps / 672 Go/s||~350W||1x 16-pin|
|GeForce RTX 4070||AD104-400||>60 / 7680||12 Go / 192-bit||18 Gbps / 432 Go/s||~300W||1x 16-pin|