OK, So I signed up and started mining at Coinotron, rejects are below 1%. I don't want to switch, I like Burnsides interface, but the amount of rejects is staggering in comparison. I would gladly come back if I were shown a fix to my problem, or if you fix your problem.
Let me know....
The reason is that coinotron has stratum. To achieve that low of rejects you need to mine with this protocol inside, burnside's pool doesn't support it.
|
|
|
They're the most efficient; I calculated profitability and it was higher using higher clocks.
|
|
|
Okay, underclocking for LTC on 3x 7950s seems slightly less productive because the memory has to be run between 1200-1500MHz. I can get my rig down to 450w (~95w per card) at 800MHz core / 1250MHz RAM / 880 mV vCore (450 KH/s/card) versus 580w (~170w per card) at 1000 MHz core / 1500 MHz RAM / 1037 mV vCore (580 KH/s/card).
450 KH/s over 95w = 4.74 KH/s/w 580 KH/s over 170w = 3.41 KH/s/w
Still, that's a pretty good power efficiency improvement.
|
|
|
That's the best I can provide at the moment. I am working on getting the current stuff.
![](https://ip.bitcointalk.org/?u=http%3A%2F%2F24.media.tumblr.com%2Ftumblr_ltuq883V701qkjcneo1_500.gif&t=664&c=c_V4PEE5ppQ4ZQ)
|
|
|
As far as I can tell, yeah, I don't know how AMD keeps messing things up so much
12.10 was a nightmare too, backtracking to any previous driver didn't fix the litecoin mining issues, upgrading to a newer driver was the only way to fix it.
|
|
|
This the best I've got. It's not entirely current and I'm not a software developer so i have no idea if the defense code stuff is in there. Have fun: http://www.sendspace.com/file/6asx7lThat's a super old version that doesn't even include scrypt.c which the authors (who?) or at least the only RUcoin pool claim the chain now uses
|
|
|
you can calculate by the number of shaders in the card as long as the card uses gddr5
see thread in my sig
|
|
|
ah, okay, that makes sense. i am used to ltcmine.ru that constantly adjusts based on your hash rate (not sure if pool.itzod.ru also does this).
|
|
|
hi jgarzik, did you try http://hhtt.1209k.com/ ? it is a variable diff pool with very low fees for high diff shares, but no stratum. best regards spiccioli That pool is fixed difficulty share (32), vardiff adjusts the difficulty every x minutes for incoming shares based upon your hash rate. Share difficulty of 32 is still very low for 60 gh/s, ideally it should be in the hundreds or thousands.
|
|
|
maybe because cash today versus maybe cash in two weeks?
|
|
|
This is encouraging. Hopefully more units will roll in this week.
|
|
|
I will have to try this for LTC mining after and see if I can get big efficiency gains. Right now I pull about 200W per 7950.
|
|
|
You mean that crazy internet currency whose co-founder publicly stated it was a scam isn't worth buying or mining? Shocking 02.02.13 18:55:06 Balthazar: c4n10, ruc is a scam 02.02.13 18:55:26 Balthazar: i can say this as co-founder https://btc-e.com/chat/history/12361
|
|
|
ASICs will come no doubt, but there is a significant barrier to entry with litecoin due to the cost to develop and produce an ASIC for litecoin and the currently low value of the coin. The configuration of GPU memory and ALUs is nearly ideal for LTC mining, making the likelihood of extreme efficiency gains seen for BTC (100x) unlikely. I would guess we will see possibly a 10-fold improvement in efficiency, if that.
Probably an optimization for running on ASICs will be diminishing the memory requirement size while enhancing the number of ALU calls, which would then be trimmed down to a circuit and put on die. But even cutting the memory requirement by 10x to ~10kb would enhance the number of ALU calls 10-20 fold. The net effect is ending up with a slower but more efficient scrypt processor, if I had to guess how many cores you'd be able to fit on a 100 cm^2 die I'd wager it'd be in the hundreds and not a big improvement over GPUs in terms of speed.
The other thing to do would be to add memory and a fast memory interface... but GPU memory and GPU cores are so fast that it's difficult to compete. If you could get GDDR5 and make a fast interface for it and make <45 nm dies (yeah, right), you'd really be moving fast and efficiently (way beyond a 10x increase in efficiency), but there's such a huge barrier of entry to that that it's probably a long ways off.
|
|
|
I would prefer dynamic difficulties for shares (further lower bandwidth, some of my rigs pull 2mh/s each), other than that it is working fine
|
|
|
Was never a fan of the old "covered in wet toilet paper" logo
|
|
|
|