Quite awhile ago actually when cryptocurrency was at the peak of its bull run it was already speculated that NVIDIA weren’t just happy giving away all the shitty Samsung 8nm GeForce trash to cryptocurrency miners but apparently they’ve felt the need to take its real Ampere, its top dog A100 built on TSMC’s 7nm processing node and also sell them to crypto miners as well.
But not it seems such rumors are finally coming into fruition as an alleged CMP 170HX has surfaced out with laughably cucked specifications and an amazing ETH hash rate to boot.
The GA100 based CMP 170HX surfaced on Zhihu by “Codefordl” and showcases various images of card itself followed by your typical GPU-Z and actual Ethereum mining hash rates.
The supposed CMP 170HX shares the same fanless design that most compute oriented enterprise equipment tends to come with, they’re designed to be slotted into server racks with chassis fans doing all of the heavy lifting.
However when we get into the GPU-Z specifications it’s pretty obvious that glorious leader Jensen Huang has decided to make use of defective A100 units by severely cucking them with the CMP 170HX and ensuring that nobody would want one of these for any other workloads besides cryptocurrency mining.
According to Codefordl the specifications of the CMP 170HX are severely reduced compared to the A100 of which it’s cut from, the CMP 170HX only comes equipped with just 70 of the maximum 128 SM’s that GA100 silicon is capable of giving out, which means the CMP 170HX comes with 4480 CUDA cores which is down from the actual A100 Tesla accelerator which offers 6912 CUDA cores from its 108 SM’s.
Now of course I’m not talking about that rubbish on its retarded GeForce cousins that have “doubled” FP32 units giving them arbitrarily high CUDA core counts on paper, all that bullshit is marketing fluff.
Aside from that another staple feature from the A100 that you’ll find cucked to oblivion on its CMP brother is the HBM2E memory where in the case of the A100 it came with either 40 or 80GB in total…. here with the CMP 170HX it’s just 8 GB. I get that NVIDIA is selling absolute trash for top dollar as CMP products but this is just fucking absurd, it has just 8GB of HBM2E memory across a 4096-bit memory interface so at least it’s better than the bullshit Titanic V.
It comes with a base and boost frequency of 1140 and 1410MHz.
Aside from all this the CMP 170HX only supports PCIe Gen 1 and is limited to only 4 lanes, while it may physically be 16 lanes this is probably to ensure that somebody isn’t smart enough to poach one of these “bad boys” for other intentions besides crypto mining, as the PCIe limitation would prove to be quite the hinderance for all things besides mining ETH.
So how exactly does it perform when mining Ethereum? VERY well.
As the CMP 170HX is limited to a single 8-pin power connector straight away it blows all the other CMP shit out of the water in terms of raw hash rates and efficiency, at 250W out of the box the CMP 170HX is capable of producing 164 MH/s which trounces just about anything else on the market.
Especially when tuned down to around 190W for maximum efficiency the CMP 170HX will gladly continue pushing 158 MH/.
Without a doubt the CMP 170HX is hindered mostly by its 8GB of HBM2E memory which just shits itself in terms of effective bandwidth compared to the 40/80 GB configs of working A100 products, but NVIDIA CMP products are made entirely for maximum profit as shown previously with $600 USD 1660 Supers.
NVIDIA themselves stated not long ago actually that this whole GPU shortage would continue well into 2022, and that much is obvious if Jensen Huang is looking for new alternatives revenue streams by harnessing defective GA100 cores and packaging them as the gouged trash that is their CMP HX lineup.
Large scale mining firms will buy this CMP 170HX is droves, their expendable income is about as endless as the kikes tyranny on the world.