NVIDIA recently announced their CMP or Crypto Mining Processor add-in cards to target miners. Now it confirms that the first two CMP models will not be based on Ampere architecture. According to sever Videocardz, the latest 461.72 driver has confirmed that the 30HX and 40HX mining processors are indeed based on old Turing architecture. The CMP 40HX would therefore feature a custom TU106 processor, while 30HX would be equipped with TU116 based GPU.

Officially NVIDIA only revealed the power rating and memory configurations of new CMP models. Only the 30HX and 40HX models are to be released in the first quarter with 50HX and 90HX not expected till at least April.
| 30HX | 40HX | 50HX | 90HX | |
|---|---|---|---|---|
| Ethereum Hash Rate(1) | 26 MH/s | 36 MH/s | 45 MH/s | 86 MH/s |
| Rated Power(2) | 125 W | 185 W | 250 W | 320 W |
| Power Connectors(2) | 1x 8-pin | 1x 8-pin | 2x 8-pin | 2x 8-pin |
| Memory Size | 6GB | 8GB | 10GB | 10GB |
| Starting Availability | Q1 | Q1 | Q2 | Q2 |
The CMP HX family only includes the bare minimum for cryptocurrency mining, which means that it doesn't have display outputs. Based on the power consumption data, the 6GB of memory, and 26 MH/s hash rate, it appears that the CMP 30HX is based on the TU116 (Turing) GPU, while the CMP 40HX with its 8GB of RAM, 185W TDP, and 36 MH/s hash rate is reportedly based on the TU106 GPU as per VideoCardz.
These are graphics cards which are dedicatedly using for professional mining.