Topic: If you optimize/make kernels... (Read 4053 times)

The man has a point. I released some optimizations to XMR for this reason - selling a 6x faster miner is good, but you may want to knock it down to 2x - 3x or you'll kill the coin.

He is rather kind to us, unlike most miners

What you're asking is for devs to deliberately cripple the speed of the miners so that most people use less power. You expect no one will make a faster miner? Or do you not care as long as it's not released to everyone?

The problem with private kernals is that the market might crash. If the public kernals are too slow, nobody will make a profit, and people stop mining. Only the big whales are left.

after all may-be it is a good idea... 
Lets convince everyone to use crippled miners, while we keep using the other...
actually that guy is the genius 

Yet it doesn't matter, because everyone is operating as slow as molasses speed, which works just fine and dandy. It saves power and it still works fine due to difficulty. You don't get a bigger piece of the pie when everyone is operating on the same kernels (all else being equal).

I was talking about hash/watt. It's a ratio. That could be at 1GH/s or 1H/s

that's what he's asking... he want less hashrate and low power usage 

LoL djm!

Maybe we have a communication issue here due to esl, I'm arguing for more efficient algos. Which is completely the opposite of the side you're arguing. Unless you ARE arguing for more efficient algos (from a watt/hash or 280x/watt perspective), in which case we're on the same page.


Yup, we were operating under the idea of what would happen if it was be publicly available. Once again I think there is a esl barrier here somewhere.

It's more efficient up till the point of when it's released, then it's less efficient.

Except that his kernel is not publicly available, which, if you use it, gives you a larger part of the pie...  
Get it ? No ? Well, I give up  

You have been trolling since 1 week about the fact you want less power for more hashrate.. (or less power for the same hashrate or whatever), I think it is time for you to let go and start thinking we are just a bunch of idiots...

You tell us now that the efficiency is bad, which isn't even true.
We don't control the market, GPU mining isn't really profitable at the moment, we can't do anything for that. sorry.
sorry but you have to live with it as anybody else.

There is no magical button (nor magical algo either) on a gpu which will make your card work 10x faster for 1/2 the power cost (which would require 20x faster algo... ).
Performing an operation on the gpu cost power, there is a limit beyond which you can't go... (because the gpu still have to solve the hash and performs the necessary operations...).

And again, what you are asking can be done by tweaking your card setting (and Claymore probably did a tweak which does it for you... or just reduce the intensity or something like that...)

So 3 answers possible to "help"' you:

* pump altcoins, this way they will become profitable again
* learn GPU programming and come up yourself with your 20x faster algo since you really seems to know a lot about what is bad in the current kernels...
* and I forgot the 3rd...  oh yes, learn how to tweak your gpu to obtain what you have been asking us to do for you (but refusing to do so far)

and "Buying thousands of dollars worth of gear isn't a option for most people", still don't understand to what you are referring....
But if you expect huge profit with 0 investment on your part, I haven't seen anything around so far which look like that....

Buying thousands of dollars worth of gear isn't a option for most people. You aren't even making any sort of meaningful argument as to why efficiency is bad.


Undervolting does not change the underlying operations of a algorithm. I can undervolt all I want, but that does not make a algorithm more efficient, it makes that card more efficient. It's another avenue of gaining efficiency, but it's not the same as a algorithm becoming more efficient.

I, just like other people I'm sure, have already taken this avenue and we're STILL below power costs right now. You can point fingers all you want, but more efficient algorithms are more helpful to everyone right now. I know you guys are just being dumb and you're so used to driving hashrate because bigger is better, but it isn't... Not when everyone is running the same kernels.

Your modified kernel is LESS efficient then what is currently publicly available. Difficulty works like a pie, if everyone is using the same 280x with the same kernel they'll all take the same size percentage piece regardless of what hashrate X is, only with your kernel it will be at 17% more power.

You could reduce clocks and volts (and people do this already), but it's still at 17% more power. That's a percentage. It works against everything. Core voltage doesn't determine overall TDP as well, as we've seen with scrypt already.

I've already seen this done with Claymore and for awhile it worked till kachur starting pushing hashrates of CryptNote and then they both ended up at about the same amount of hash/watt and the %s were the same, but now consumed about 20 watts more per 280x then they used to. Coins dynamically change difficulty.

yawn.....  boring .....
again you want less power, then undervolt... you'll get what you ask for... then when will get 50% hashrate gain, you will still be able to do that.
or buy low power cards... or buy ltcgear shares

edit: actually I did it on sk-1024 (coinshield) depending how you write the algo you increase the power consumption by 20~30% with no speed increase (I chose the other one obviously...).  Voila, just mine coinshield, you'll get what you ask for...

You can only tune so much. You can't cut power usage by 50%. One of the benefits of x11 for instance is reducing power usage, from scrypt. Scrypt ran things full boar and I'm sure you remember how much heat that put out and used.

I've already tweaked, tweaking doesn't help that much. Everyone tweaks right now. If someone can increase hashrate by 50%, they can reduce power usage and keep the hashrate the same as it is right now.

I'm not even certain why some people are against this. Everyone should be happy and agree with reduced power usage, it'll save everyone money or god forbid they'll make money (not counting loans).

What? We're talking about the most efficient excavator and drills here. Mining as fast as possible for the smallest overhead. Efficiency doesn't mean people will necessarily mine slower, it could be much, much faster, as long as hash/watt is lower.

That completely puts aside that everyone ends up on the same kernel, so they all mine at the same speed and consume the same amount of power (unless they're using private kernels which don't count). Increasing efficiency when we're sitting at power equilibrium will benefit everyone. Meaning everyone right now is mining at the price of power (it's actually lower now as everyone is using their GPUs as space heaters and driving prices ridiculously south).

There is opportunity cost for getting into GPU mining. More efficient kernels isn't going to lead to another Bitcoin Nov GPU boom. People are fucking retarded if they're buying GPUs to make a couple cents a day in profit or break even with power (which isn't even profit). That doesn't take into account the price of hardware, what you could do with that money if you didn't buy, or depreciation. You could put your money in the bank for example and make more in a year.

You talk about big industry, but no one would buy into GPU mining right now or in the future. That ended around last spring and smart people sold their GPUs for almost what they bought them for and made money. If you took out loans and hung onto hardware past last spring, you're still very negative.



Efficiency is hash/watt, which is what we're talking about. Your 'private' kernels don't count towards anything here as they aren't available to everyone and if they were, it'd be less efficient as everyone would be hashing at the same (faster) speed, but using 17% more power. Your kernel is only more power efficient till the public gets it, then it's less efficient.

Really it's not even hash/watt as hashrate does not correlate to a linear increase in profit, it would be closer to 280x/watt as each 280x unit will always mine the same amount at the same power levels without tweaking (all else being equal).

Hashing more when everyone is using the same kernel just means the difficulty goes up, but no one makes more. Coins self adjust to new hashing power, they don't adjust to power costs. That is a man added attribute that occurred due to the decline in price of BTC and would only result in a rise in difficulty if more people bought GPUs, which no one would right now unless they're retarded or know some whales that say BTC is going to hit 1000 at christmas.

Furthermore:

If your wish is granted, and kernel developers made GPU's hash at a lower wattage, the ONLY thing that will do is incentivize miners with high electricity costs to get into mining.

All those people that turned off their rigs because the power cost was too high (such as yourself, maybe) would turn them back on, and difficulty would go up again. Essentially, the same exact thing would happen regardless of how they improve the kernel. Making low power kernels at the expense of hashrate would actually prevent optimizing for cheapest possible power.

Are you going to tell companies that make oil drilling equipment to make weaker and low powered drills? Are you going to tell companies that make gold mining equipment to make weaker and low powered excavators? No, that would be laughable. Power is simply not the bottleneck for these industries, and it isn't a bottleneck for crypto mining either.

The biggest cost is in the UNIT, not in the POWER used to RUN THE UNIT. Therefore, these companies will always produce the best possible and most efficient drills, excavators, and kernels in terms of production per unit.

That is just the nature of the business. If you want to save on power, you have to find a cheaper utility company.

but still you can do that by tweaking the parameter of your card...
The problem doing this is that there will always someone telling that we give low speed miners while we keep the fast one for us (or some group of people) which in that case would be true.
You can always undervolt, underclock (which will reduce consumption since it will require less voltage to run etc...) etc...
Modifying the kernel here isn't a good solution.

Optimizing kernels usually means people are optimizing for more 'speed', not power efficiency. Gains are almost always in speed. Claymore used to have a low power mode that raised efficiency of the cards at the cost of speed (or a little bit less speed). It doesn't really work well as of 9.1, but it used to reduce the power consumption of a 280 by about 20w per card. The watt/hash was much better and because we're mining around the cost of power, I came out ahead often times even though I mined less.

It makes perfect sense. Power isn't free. Increasing hashrate for public kernels does nothing to help anyone as we all end up taking the same size slice of the pie (when everyone is on the same kernel). Better power efficiency means miners can actually turn a profit as we'd no longer be stuck at the price of power.

I never said it would be faster... (more calculation means less speed)
But low power algo are usually algo with memory access bottleneck.
Meaning the gpu spend most of its time waiting to read or write to memory before being able to process the next operation.

That why I think it what you ask doesn't make much sense.
Everybody want optimized kernel and optimizing a kernel means finding a way to have the gpu doing faster calculation full time with no latency.

Then again, the best way to have low power usage, is to tweak on your side the gpu setting by either changing the tdp (which at least for nvidia is
different from the temp target and really act on power) or change the thread concurrency/intensity