Despite the fact that GPU is the acronym for Graphics Processor Unit or Graphics Processing Unit, the reality is that nature makes them excellent for work tasks that work in parallel where they are several times more efficient than a CPU. Among these We have seen GPUs make an important contribution to scientific computing, but if there is an application where GPUs also excel, it is in mining, where the NVIDIA RTX 3000 wins the battle against the AMD RX 6000.
With the arrival of new generations of graphics cards, not only the most enthusiastic gamers are interested in them, but also those who want to set up complex mining stations using these GPUs, but it seems that when it comes to mining, the superiority of NVIDIA does not admit of discussion.
NVIDIA RTX 3000 outperforms AMD RX 6000 in mining
The author of the leak is kopite7kimi, who in a Twitter thread has dropped that the Radeon RX 6800 has better performance than the RX 5700 but worse than the RTX 3080 ,
Although he has not given the performance figures of the AMD card, he has confirmed that they are much lower than the 80 MH / s in Ethereum, Keep in mind that the RTX 2080 Ti reached 57 MH / s so if we have to make an estimate by eye the RX 6000 would be below 70 MH / s .
Why superiority of the RTX 3000 over the RX 6000 in mining
There are a number of elements that make the NVIDIA RTX 3000 superior when it comes to mining compared to the AMD Radeon RX 6000 and they have to do with the way in which both AMD and NVIDIA have raised their architectures. If you’ve ever wondered how mining farms hunted down the few RTX 3000s as soon as they went on sale, there are plenty of reasons for it.
The first of these is the fact that each SM unit has about 128 FP32 ALUs inside it instead of the 64 of the RTX 2000 and the AMD RX 5000 and RX 6000. This means twice as many floating point operations per cycle of clock and although the RTX 3000 range has a higher power consumption, if we average watts per TFLOP , then we will realize that the performance of the latest generation of NVIDIA under that metric is much higher .
The other element is the Tensor Cores, these units are systolic arrays for the AI, but they have the peculiarity that they can also execute low precision SIMD instructions. Actually, what NVIDIA does is transfer the calculation in FP16 and Int8 to the Tensor Cores where they have a huge comparative advantage over AMD, so if the decoding algorithm performs operations with these precisions and is optimized for the Tensor Cores then NVIDIA GPUs have a huge advantage.
AMD’s bad strategy for mining
Actually, what would be a beast to mine is its compute-only GPU, the future CDNA , which is the real successor to the Vega at 7nm . But AMD apparently would have decided to recycle the excess stock of the RX 5700 to create a low-cost mining-only card, much less powerful than its RX 6000 but cheaper and designed so that the mining world does not leave out of stock to your RX 6000.
In any case, in relation to power / consumption , which is one of the most important elements for mining, NVIDIA’s offer, even with much more expensive cards and higher consumption, gains in performance / consumption, since one of the most important in mining is the frequency of mining at a specific energy consumption.
When the AMD Vega 14 nm were a commercial fiasco they were able to relocate it in the mining market, but with Vega 7 nm they controlled the distribution very well. Therefore, it does not appear that the mining stations will have access to the CDNA architecture about to be presented.
Source> Twitter (Kopite7Kimi)