What makes a GPU better at mining than another?

AngeNoir0324

Notable
Apr 28, 2017
329
0
810
What I want to know is in the title, but more specifically between the current gen Nvidia and AMD cards/chips. Obviously power consumption makes them better than the last generation, but still.

What does the AMD cards have that the Nvidia ones didn't? More VRAM, CUDA/STREAM cores?

Thanks for shedding some light
 
Solution
How well the design of the card matches the algorithm in use by the cryptocurrency. Not every algorithm is the same. Some value throughput, others are more sensitive to latency, others just need brute force etc. Being good at one type of calculation often comes at the expense of another type though, so there isn't really a "best" architecture per se. And newer cryptocurrencies such as Ethereum are deliberately being designed to perform poorly on certain architectures, namely ASIC's.

AMD cards have tended towards the brute force approach for the last few generations, while Nvidia has spent a lot of effort on memory compression and tiled rasterization and so forth to try to be as efficient as possible. In games, it's allowed them to do...
How well the design of the card matches the algorithm in use by the cryptocurrency. Not every algorithm is the same. Some value throughput, others are more sensitive to latency, others just need brute force etc. Being good at one type of calculation often comes at the expense of another type though, so there isn't really a "best" architecture per se. And newer cryptocurrencies such as Ethereum are deliberately being designed to perform poorly on certain architectures, namely ASIC's.

AMD cards have tended towards the brute force approach for the last few generations, while Nvidia has spent a lot of effort on memory compression and tiled rasterization and so forth to try to be as efficient as possible. In games, it's allowed them to do more with fewer resources and lower power. AMD often had more pure processing power available, but was held back by drivers or memory. That's helped with artificial workloads like mining, but has hurt them this generation with the reliance on HBM2.
 
Solution

Mupples

Honorable
May 10, 2013
15
0
10,510
Every core on an AMD gpu can operate indepdently within it's ACE unit, Asyncronous Compute Engine units. Because of this it has unparrellel compute performance but sacrifices in other areas. NVidia tethers it's compute engines to the clusters.

It is simply that every core can be leveraged for compute operations and memory bandwidth is need as every core is trying to an accessive amounts of read and writes.