I basiclly stumbled onto this just with playing with cgminer. I read in one post the best thing to do was to adjust mem speeds down to 300. Well this may keep things cool but hash rates suck. I have a couple old 5830's one with 2 pipe cooler and one with the 5 pipe cooler. Keeping the one with the 2 pipe cooler cool is a task. The ratios you mention here are they just from experimenting or is there definitive math proof for the numbers?
For the 7770, yes https://bitcointalk.org/index.php?topic=117221.msg1302319#msg1302319From what I can tell this holds true for my 7950s too, but I don't have as much quantitative data. Someone needs to make these same curves for the 5xxx and 6xxx series too, since they use a different architecture. It'd be nice if someone could, I've wanted to roll out a program to do this automatedly for some time but have been working or coding other things.
|
|
|
The small block size is actually kind of nice for miners, since it forces clients to include bigger fees if they want their transactions to go through in a timely manner. I'm not hugely opposed to it right now, the problem is moreso the SD spam that's been plaguing the network.
|
|
|
If the GPU I want to use for mining is not used as first device the miner doesn't see it. In my case I want to use the integrated GPU in my I5 for desktop and mine with my Radeon 5770, but it doesn't work. With the guiminer for Bitcoins this problem doesn't occur, so I dunno if it's my fault or there is a little bug.
With cgminer or reaper? Right now there's a bug with cgminer and you need to add the extra flag, That should fix it. Reaper seems to be unaffected because it enters the platform number. This will be fixed in 0.03 alpha, which should come out this weekend (want to see if any other bugs come out first).
|
|
|
Your aggression is way too high. Try with aggression of 12 and increase it by 1 until you hit max hash rate.
It looks like your display driver crashed. You can try it with lower aggression and increase it like he said, and you can also adjust your core and mem clocks to different values. Alternatively you can also try higher or lower thread_concurrency values and see if they help -- every 7xxx card I've seen can do 18+ aggression if the core and memory clocks are set correctly along with thread_concurrency.
|
|
|
Yup, it was reaper. I will download the newest version and try again.
Cool, let me know if it works okay now.
|
|
|
Actually, that would be perfect as it would cap SD flow at 3%*MaxBlockSize/AvgBlockSize of Bitcoin's capacity leaving the rest for everyone else. Nice.
Right. There is no easy way around it for SD; they have to pay someone more. Not only this, but if SD has tons of spam transactions like they do now, they'll never fit into 6 blocks per day and will have to start overflowing into the next block, leaving users for days without any confirmation of unsuccessful gambling.
|
|
|
RAM REQUIREMENTS ON-MOTHERBOARD The equivalent amount of system RAM as for the sum of all the vRAM used by the GPUs is required when mining with reaper. For instance, 3x 5970s mining with thread_concurrency values of 8000 each would require 3x 500MB = 1500MB system ram (plus additional RAM for the OS). The memory requirements for 7xxx cards are also higher because of larger thread_concurrency values, for instance a thread_concurrency of 24000 means 1.5GB system RAM per card is required.
You shouldn't need a lot of RAM for cgminer, it offloads it to the GPU unlike reaper which for some odd reason keeps it in the system RAM. reaper is coded really sloppily while cgminer is pretty immaculate. So I am trying to figure out how much memory will I need for two rigs, mining only with cgminer: First rig: 4 x 5870 and 1 x 6990 with 2GB RAM already on the motherboard Second rig: 4 x 7970 ......................with 2GB RAM already on the motherboard Can you give me an estimation? Thanks. 2GB for the first rig is fine as long as you are only using cgminer. With the 7970s, you're going to want to use reaper to achieve the highest hash rates at the lowest possible power consumption. You will need ~1.5 GB * 4 = 6 GB + 1 GB system = 7 GB RAM minimum. Although, you can use 2 GB and cgminer if you want, I just overall much prefer reaper for 7xxx series. 8 GB of RAM is only $40 now, so.
|
|
|
tacotime, thanks for this guide!!
I tried mining with my 7970, but I can't make more than 550kH/s (with core:1050, mem:1600). I've tried reaper and cgminer with tons of different configurations.
I use Windows7 x64 and Catalyst 13.1 - might this be a problem? Should I use Linux? A different AMD driver?
As it says in the guide, use a ratio of 0.57 1050 MHz core / 0.57 = ~1840 MHz RAM clocks; you are only using memory clocks of 1600 MHz. The only way you can hit 1840 MHz RAM is by overvolting the RAM in afterburner. Try instead 950 MHz core / 0.57 = ~1670 MHz RAM clocks, you should get over 550 KH/s with less heat
|
|
|
I run my 6970 at 915 MHz with a ratio of 0.8
saphirre trixx sucks balls whenever i set manual fan control it changes for a few seconds to my trixx setting then it reverts back to the cards default setting what would that make your memory clock? i took core*0.8 * 2 (ddr effective rate?) and got 1464 if thats your mem clock then using that ratio doesn't get me any more khash than the stock ratio of memory to gpu which is 800/1250 = 0.64 915 MHz / 0.8 = approx 1150 MHz I think I get like 20-30 more kh/s at 1350 MHz, at the expense of more heat
|
|
|
I run my 6970 at 915 MHz with a ratio of 0.8
|
|
|
My solution for the Satoshi Dice spam is here: 1) Pools or big solo miners remove all transactions from blocks they mine that don't have at least a 0.01 BTC fee attached to them or 2) which don't have at least 0.01 BTC per each output.
You might set these values arbitrarily to something else, it doesn't matter, this is the easy solution in my opinion.
People keep arguing it's in the best interest of the miners to take all transaction fees, no matter if they are miniscule or not. This is not true. Miners are not submissive nodes in the network that follow the client userbase blindly, they are the network. When clients start to see that it takes days for their transactions to go through, they will start paying greater fees or stop making these spam transactions.
Note that this seldom hurts the vast majority of users in the network, who will almost always make transactions with outputs of more than 0.01 BTC. They don't need to pay any greater fees and their commerce is unimpeded, while spammers like Satoshi Dice are made to pay you more.
|
|
|
Just get BTCGuild, 50BTC, DeepBit, and Slush to remove all transactions that don't have at least a 0.01 BTC fee attached to them or which don't have at least 0.01 BTC per each output. The solution is really, really easy and makes miners more money. Miners have room to be greedy.
Clients will learn real quick to incorporate fees, and SD will learn real quick to stop making these spam transactions. There's no reason that SatoshiDice just couldn't present the user with a page for each transaction showing if they won or lost, and use up their own server space instead of that of the block chain.
It's not even remotely a problem, the pools just need to act now to preserve the integrity of the block chain. And I'm guessing we'll see this within the next month or two, and SD will be forced to change its ways.
|
|
|
Yeah, I switched about 3 months ago. I know burnside works really hard to keep the place running, but he needs to implement stratum and vardiff at this point because he can't compete with the larger pools. It will be more beneficial to him in the end to use stratum anyhow, more miners, more blocks, more fees for him (if he had vardiff and stratum at 1.5% fees PPLNS I'd be back in an instant).
|
|
|
I'm sorry but I had to move to litecoinpool.org until you can figure out what is wrong. I don't like switching teams in the middle of the game but the 40% average stale rate is killing me. I get <1% there. Come to ltcmine.ru and get 0.27% stales, that's what I'm getting right now.
|
|
|
Same here. I'm trying to mine for pool-x too. And what should I write in "Additional parameters" to boost my hashrate? Because actually I'm sitting at ~20 khash with a Radeon HD5700, 0% usage. This doesn't use gpu, you want my guiminer: https://bitcointalk.org/index.php?topic=150331.0
|
|
|
Works really great.
Question, the default setting, is it mining with reaper or cgminer?
Thanks
Think I have this bug figured out too; external_path_ref = (data.get('external_path') or _('CGMINER')) ... self.txt_external.SetValue(external_path_ref) This sets the configurable external path self.txt_external (hidden in my implementation) but not the actual value that determines which miner it is, the self.external_path variable to CGMINER when no other value is given (eg when starting for the first time). I need to change self.external_path here too, so that the miner knows what miner it is before you hit the start button. When you restart the miner, it should show up as cgminer (probably why I never caught this bug, haven't deleted my settings .ini file in a while).
|
|
|
Okay, I see the problem now. # Command line arguments for cgminer here: # -u <username> # -p <password> # -o <http://server.ip:port> # -d <device appear in pyopencl> # -l <log message period in second> # -T <disable curses interface and output to console (stdout)> cmd = "%s --scrypt -u %s -p %s -o %s%s:%s -d %s --thread-concurrency %s -w %s -v %s -I %s -g %s -l 5 -T %s" % ( path, self.txt_username.GetValue(), self.txt_pass.GetValue(), http_header, self.host_without_http_prefix, self.txt_port.GetValue(), self.device_index, self.txt_thrcon.GetValue(), self.txt_worksize.GetValue(), self.txt_vectors.GetValue(), self.txt_intensity.GetValue(), self.txt_gputhreads.GetValue(), self.txt_flags.GetValue()) return cmd, os.path.dirname(path) If you look in the above code, it selects for -d (device) # but not --gpu-platform #. To fix this, I just need to add the --gpu-platform platform flag, since the GUIminer code already gets this information as self.platform_index. I will update this in version 0.03 so you don't need to put the flag in any more. This is apparently a bug carried over from the original GUIminer.
|
|
|
No
I'll check after and make sure that the platform is in the flags, I'd thought it was but maybe I'd made a mistake
|
|
|
GUIMINER SCRYPT GOOD JOB! more donation for you Thank you! If anyone spots anymore bugs please let me know sooner than later and I'll try to fix them all by the end of this week
|
|
|
|