NVIDIA Wasting TSMC 7nm Capacity on Crypto Miners? NVIDIA Potentially Preparing Ampere A100 CMP HX Mining Card
22 March 2021

“Everyone wants Israeli companies” – Jensen Huang

It would appear that NVIDIA’s greed certainly nose no bounds with the company already putting focus on its GeForce RTX Ampere graphics cards being sold directly to miners on top of rebranding previous generation RTX Turing trash for its CMP HX lineup of mining products which are actually godawful for mining.

As you may know by now NVIDIA’s latest generation of GeForce garbage is being produced on Samsung’s godawful 8nm mobile processing node which is godawful for large desktop trash in the form of Ampere however NVIDIA has put the big bucks towards TSMC’s vastly superior 7nm processing node when it has come to their data center segment such as the NVIDIA A100.

NVIDIA’s A100 is surely a compute powerhouse coming in with 9.7 TFLOPs of FP64 and 19.5 TFLOPs of FP32 compute power excluding bullshit Tensor cores, being built on TSMC’s 7nm processor surely gives the A100 a serious boost in efficiency over Samsung’s garbage 8nm processing node which NVIDIA runs beyond its sweet spot intentionally to reach the rated frequency of its GeForce products.

It should also be mentioned that NVIDIA’s A100 contains a staggering 6912 CUDA Cores with 40MB of L2 cache whereas the GeForce trash get zilch. And it comes in lovely variants of 40GB of HBM2 memory providing a staggering 1.6TB/s memory bandwidth or a 80GB variant for supercomputers utilizing HBM2E memory which bumps that bandwidth up to 2TB/s effectively.

Needless to say that NVIDIA’s A100 would certainly be a crypto miners wet dream.

And NVIDIA certainly wants to make that dream a reality (for profit of course) because according to renown leaker kopite7kimi NVIDIA could be planning to repurpose what little TSMC 7nm production it has to spare A100 products into its CMP HX lineup.

It’s a smart move for the company if you actually think about it, why charge $10,500+ for equipment that Jensen Huang practically has to give away most of the time when instead NVIDIA could just as easily repackage A100 PCIe as a cryptocurrency mining monster for probably $4000 easy.

This is just another act in the saga of NVIDIA pretending to care about gamers. With the bogus hash rate limitation of ETH for the RTX 3060 for the company to then “accidentally” release a insider BETA driver update that unlocks that restriction just a couple weeks later. With the top-end CMP 90HX producing just 86 MH/s on Ethereum NVIDIA still falls short of the like of AMD’s Radeon VII which can easily hit hash rates anywhere from 90 to even 100 MH/s even.

NVIDIA’s A100 Tensor GPU is a compute powerhouse with a massive surplus of HBM2 memory on board boasting staggering memory bandwidth figures, if miners could get their hands on such hardware which is all but guaranteed to blitz pretty much anything for Ethereum mining they’ll pay any price, and being built on the TSMC 7nm process no less, the A100 comes with a TDP value of just 250W making it vastly more efficient than the likes of GA102 when it comes to mining.

With a slated release window of Q2 for the CMP 50HX and 90HX it’s anyone’s guess whether or not NVIDIA will follow through with plans to introduce A100 into their CMP HX fold considering just how indecisive Jensen Huang is surrounding the 20GB RTX 3080, 16GB 3070 and 20GB 320-bit RTX 3080Ti which were certainly on the table but have since been shelves or changed entirely.

However if NVIDIA actually have the audacity to pull something like this off which I certainly wouldn’t put it past them, NVIDIA has always been a company built on deception, lies and most certainly profit. Though I just couldn’t hold back my laughter that NVIDIA would seriously stagger its data center production with the company’s limited 7nm orders with TSMC by focusing on selling A100 to cryptocurrency miners also.


blog comments powered by Disqus