[Hardware] NVIDIA CMP 30HX and 40HX Confirmed to Be Turing Based And Not Ampere

[EC HTN^]

 

To say the NVIDIA CMP Series has been met with great enthusiasm would likely be a misguided statement for the gaming community. But in an effort to separate gamers and miners for access to graphics cards NVIDIA recently announced their CMP or Crypto Mining Processor add-in cards to target hash happy miners. They didn't go very far into details but many were quick to assume these were more Ampere based graphic cards and would be sucking the life from the gaming silicon. But, thanks to Videocardz catch we find that at least for the CMP 30HX and 40HX it won't be the case.

 

 

When the cards were first announced I had my suspicions on what cores they would be used by simply comparing existing cards stock hash rate and the memory capacity. My original thoughts on the 30HX and 40HX were that the 30HX would be based on the RTX 2060 TU106 core because that's about where it hits on the hash rate, but I failed to consider the GTX 1660 SUPER and it having the same memory configuration, but it checked out. The 2060 Super and 2070 both lands about the same hash rate and power draw so that was a fairly easy assumption and according to the 461.72 it appears to ring pretty true

 

 

Based on the details from the driver string here we can see that the CMP 30HX is based on the TU116 like the GTX 1660 SUPER and the CMP 40HX is based on the RTX 2060 SUPER. If these cards turn out to be as tunable as their GeForce counterparts we can expect the 30HX to hit right around 30MH/s at 85-90w and the 40HX to land between 40-42MH/s at around 125w, both figures make it much more attractive than the rated speeds and power consumption. But of course, price and availability speak volumes.

While it's good to see these two taking advantage of older 12nm Turning dies that can be made on a less constrained node they'll still need VRAM modules to function and we can basically chalk up the CMP 50HX and 90HX being Ampere based GA102 cores, but most likely with either massive defects for GeForce purposes or the Tensor/RT Cores either not used or no good. Only time will tell.