In the property, the main thing of each video card is a video memory, in some video cards it is more expensive, while others are cheaper, and it depends not on the model, but on the occasion.
 When you buy a video card you never know what memory you get a video card. People try to determine the type of memory by the batch number, other algorithms, I personally prefer not to bother with this and I do not advise you. Why? Now a huge deficit of video cards, large players massively buy video cards from factories, the same video cards that came out instantly buy up the bugs. You want to buy a video card and get a cache or you will sit and wait until the party with the desired number comes. Choosing for you. Personally, I consider the "landing" on the CPU not feasible. Increasing complexity, the cost of components, the need to purchase a motherboard for every 1, 2 processors. Why 1, 2 Because there are motherboards that support installing multiple processors. Video cards are times more productive than processors, for profit, it is better to choose coins for "mining", so I think at this moment, this option should not be considered.
That is, if a person purchased this ASIC, he would fly into a round sum of money, and if he bought 10-100 such?

At that moment, when he was in pre-order, everything was fine with profitability, but after the release ...

The bottom line is that video cards have not lost profits at all. At the time of writing, the profitability is maintained at the same level. Usually there were no settlements without them, corrections, and so on.

The constant growth of complexity, the unstable exchange rate, if you break the repair yourself or throw away, all this is present in asikov.

Of course, there are certainly more interesting options for asics, for example, Antminer S9. One of the most popular, if not the most popular currently ASIC, which owns the SHA256 algorithm. But when buying this asic as you are taking a video card, you take risks. In the first case, the complexity of the layer is always static 512. This means that we will send each layer with a complexity of 512. Accordingly, if for example, on the complexity of 256 we would send a 20 layer, then on 512 complexity we will send 10 layers, of course, all this is about + - the difference may be, but the bottom line is that the higher the complexity, the more difficult our equipment. Accordingly, if you have Highend equipment, then the complexity should be.
How to determine if some other pool is suitable for landing? Personally, I look at the following 2 criteria. Hashtime Pool and Hashtime Network. If we see that the pool's power is half the full power of the network, then such a bullet can safely lose the tokens. If we see that this pool is only 1% of the total power, then we will receive tokens on it extremely unstable, we may find one block per day, but maybe we will not find anything. It's like in the solo mining, roulette.