Most (or all?) cards that are shipped currently seem to have Hynix memory, really bad...

Guys, please stop asking always the same question! No, no one could, can or will be able to tell which memory is used on which card!

Even getting a card with Samsung memory does not necessarily mean that you got a "good" card. For example I have two Gainward GTX 1060 6GB, both Samsung.
Both do ~450 H/s stock in XMR. With memory overclocked +800, one does 540 H/s while the other one hardly gets the 500 H/s. This means although they both have Samsung and running the same memory speed, they yield different results.

To conclude, you cannot say that a certain card, memory type or overclock will guarantee a certain speed. It totally depends on the card, maybe memory have different timings, moreover you can't really say whether the GTX 1060 ECC is already eating up some of your overclock until you do lots of tests with different speeds. Maybe its dfferent with ETH, but with XMR, it's like that obviously.

But not with Hynix or do you have better results with Hynix?
Whether I run the cards with 1507 or 1911 core makes no difference in Mh/s, so I prefer to fix them at 0.8V and 1709 Mhz to have lowest power consumption.

Got three ZOTAC GTX 1060 Mini 3GB these days, one Samsung, 2 Hynix.

Samsung doing 19.x at stock speeds, Hynix doing 15.x at stock. So if you get a Hynix card, get rid of it.
Have no experience with Micron chips so far, but I guess they are somewhere in between, maybe closer to Samsung than Hynix.

Seeing you insulting other people just because they do not believe anything you tell without giving a single proof for it is enough for me to know what kind of person you are.

It's almost impossible to get 27 from a GTX 1060, even with +1000 memory you won't have more than 24-25.
So for me it's a typical "pic or didn't happen"!

Samsung But I have a 2nd card (same model) that already struggles with +700, currently trying whether +650 is stable.

Really strange, one of my GTX 1060 can mine XMR with 1511 MHz @ 0.8V and mem +800 MHz without issues...

So how far can it be pushed without clock increase and how far with clock increase in your case? Although I do not really see the link between core clock and memory clock, but only nVidia knows how their voltage regulation works internally...

You meant "you can't" didn't you? Otherwise, you sentence makes no sense..

The memory amount is not relevant for the hash rate but the memory performace, and here 3GB and 6GB cards are the same.

Why do you only have 14 MH/s stock, thats a shame. You should have 19 MH/s stock and 22-23 overclocked.

What patch are you talking about? Source?

If you fix the GPUs at a lower voltage, you should be able to stay below 600W.

I ordered 3 GTX 1060 with 3GB for 200€. 6GB version sells for 280+. Even if the 3GB will sell worse in a few months, I do not think that the 6GB is worth the money. However, I do not mine ETH but XMR, for ETH the 3GB could become a problem if the blockchain keeps their growing rate ^^

I would try to get a used GTX 1060. The cheaper, the better. The more you spent on the card, the longer it will take until you reach the ROI.
As clock rate is irrelevant, the cheap models are fine as you do not need to overclock and hence do not need best colling solution.
Which RAM is used is just luck. Even on check cards from Palit and PNY you can have Samsung chips. Most 1060 seem to have Samsung chips.

Speed. Yes the 9Gbps card would be faster but imho its more efficient to overclock a normal 1060 and save the money.

ETH mining only scales with memory not with core, so the TDP has no influence on the result.

Hi guys,

I noticed that Monero scales 100% with memory clock rate and 0% with core clock rate. With a GTX 1060, +100 Mem brings about +11 H/s. But it does not matter whether you clock rate is 1507 or 2000, doesn't change anything. So it's possible to fix the GPU clock @0.8V and 1534 Mhz and the mem to +700 to get about 530 H/s at only 40W. Quite good.

However, I was wondering whether it might make sense to mine something else in parallel that is more core intense to get most out of the card. Is thery any coin that might be worth mining that does mainly depend on core clock performance and uses less memory? Or are all these coins already so ASIC heavy that GPU mining does not make a sense?

Best regards

Stefan