cbuchner1 (OP)
|
|
February 28, 2014, 01:29:40 PM |
|
oh look what I just got from nVidia: Attaching the modified keccak.cu. I just got 144 MH/s on a 750 Ti with it (compiled natively for sm_50 with 80 registers). The results should be correct, but I’m not completely sure – can you verify them on a pool?
I received this update after exchanging some discussion about the nvcc compiler's possible use of the funnel shifter on 64 bit rotate/shift instructions. I will try this out after work - 144 MHash/s on a 750 Ti sounds like a dream come true. And maybe the gain carries over to Compute 3.5, too. so far I've seen only 80 MHash/s...
|
|
|
|
patoberli
Member
Offline
Activity: 106
Merit: 10
|
|
February 28, 2014, 01:31:24 PM |
|
Would say the GT-640. Some 72 kh/s with scrypt, but still some 1.45 kh/s with keccak. While a 780 Ti (this is now only a guess from reading past posts!) manages around 700 kh/s scrypt and around 4-5 kh/s keccak. [Edit] Or do I currently mix up keccak with an other algorithm?
|
YAC: YA86YiWSvWEGSSSerPTMy4kwndabRUNftf BTC: 16NqvkYbKMnonVEf7jHbuWURFsLeuTRidX LTC: LTKCoiDwqEjaRCoNXfFhDm9EeWbGWouZjE
|
|
|
cbuchner1 (OP)
|
|
February 28, 2014, 01:33:10 PM |
|
Would say the GT-640. Some 72 kh/s with scrypt, but still some 1.45 kh/s with keccak. While a 780 Ti (this is now only a guess from reading past posts!) manages around 700 kh/s scrypt and around 4-5 kh/s keccak.
confusing Keccak and Yacoin's scrypt-jane:14, maybe?
|
|
|
|
ManIkWeet
|
|
February 28, 2014, 01:44:45 PM |
|
you can run autotune with the -D flag and abort as soon as you see that there's not much to be gained from waiting longer. It's also a good idea to log the output to a file for later review.
the 2>&1 syntax merges stderr and stdout into a single output stream.
cudaminer.exe -D -d 0 -l T 2>&1 >autotune_logfile.txt
Only partly logs output to the file on Windows 7... I noticed the best autotune from -L 4 so far.
|
BTC donations: 18fw6ZjYkN7xNxfVWbsRmBvD6jBAChRQVn (thanks!)
|
|
|
cbuchner1 (OP)
|
|
February 28, 2014, 01:50:15 PM |
|
I noticed the best autotune from -L 4 so far.
I think you haven't mentioned yet which hardware you are tuning for. I was seeing about one kHash/s gain on my GTX 780, for example. My formerly 13 kHash/s 3x 780Ti rig is now pulling 15.6 kHash/s. Christian
|
|
|
|
ManIkWeet
|
|
February 28, 2014, 01:51:10 PM |
|
I noticed the best autotune from -L 4 so far.
I think you haven't mentioned yet which hardware you are tuning for. I was seeing about one kHash/s gain on my GTX 780, for example. My formerly 13 kHash/s 3x 780Ti rig is now pulling 15.6 kHash/s. Christian I have just one card, Asus GTX 780
|
BTC donations: 18fw6ZjYkN7xNxfVWbsRmBvD6jBAChRQVn (thanks!)
|
|
|
myagui
Legendary
Offline
Activity: 1154
Merit: 1001
|
|
February 28, 2014, 01:52:53 PM |
|
oh look what I just got from nVidia: Attaching the modified keccak.cu. I just got 144 MH/s on a 750 Ti with it (compiled natively for sm_50 with 80 registers). The results should be correct, but I’m not completely sure – can you verify them on a pool?
I received this update after exchanging some discussion about the nvcc compiler's possible use of the funnel shifter on 64 bit rotate/shift instructions. I will try this out after work - 144 MHash/s on a 750 Ti sounds like a dream come true. And maybe the gain carries over to Compute 3.5, too. so far I've seen only 80 MHash/s... Can't wait to try that out on my 750 TI, would be awesome to have this for the Helixcoin launch tonight!!!
|
|
|
|
cbuchner1 (OP)
|
|
February 28, 2014, 01:53:05 PM |
|
I have just one card, Asus GTX 780 weird. I am running mine on -L 6.
|
|
|
|
cbuchner1 (OP)
|
|
February 28, 2014, 01:53:32 PM |
|
Can't wait to try that out on my 750 TI, would be awesome to have this for the Helixcoin launch tonight!!!
if the nVidia code is free of errors, you will have it for the Helixcoin launch tonight.
|
|
|
|
ManIkWeet
|
|
February 28, 2014, 01:57:18 PM |
|
My autotune results for Asus GTX 780 OC: -L 4 = -l t128x2 = 5-5.5khash/s pretty much 50/50 -L 5 = -l t79x4 = ~5khash/s -L 6 = -l t191x2 = 4.3-5khash/s -L 7 = -l t118x4 = 4-5khash/s more often closer to 4 than 5 Maybe some memory limitation by windows, or maybe the fact that my PCI runs on 2.0 x 16 Edit: Correct output to file syntax appears to be cudaminer.exe > autotune.txt 2>&1
|
BTC donations: 18fw6ZjYkN7xNxfVWbsRmBvD6jBAChRQVn (thanks!)
|
|
|
cbuchner1 (OP)
|
|
February 28, 2014, 02:06:20 PM |
|
Correct output to file syntax appears to be cudaminer.exe > autotune.txt 2>&1
oh I did not know the order mattered that much should cudaminer maybe automatically save autotune logs? maybe with autotune date and GPU model in the file name and the given command line arguments in a comment line in the log file ? Christian
|
|
|
|
ManIkWeet
|
|
February 28, 2014, 02:11:01 PM |
|
should cudaminer maybe automatically save autotune logs? maybe with autotune date and GPU model in the file name and the given command line arguments in a comment line in the log file ?
Christian
This could be very useful, both for troubleshooting and general launch configs, perhaps it's possible to obtain the hardware id aswell? We all say GTX 780 but there's 4 different ones. Or perhaps the clock speeds?
|
BTC donations: 18fw6ZjYkN7xNxfVWbsRmBvD6jBAChRQVn (thanks!)
|
|
|
|
bigjme
|
|
February 28, 2014, 02:24:29 PM |
|
Correct output to file syntax appears to be cudaminer.exe > autotune.txt 2>&1
oh I did not know the order mattered that much should cudaminer maybe automatically save autotune logs? maybe with autotune date and GPU model in the file name and the given command line arguments in a comment line in the log file ? Christian I attempted to get autotune to save its results to a file. It worked for the first 13 runs of any scrypt and then cudaminer would crash. This was on a launch about a month ago
|
Owner of: cudamining.co.uk
|
|
|
lordaccess
Member
Offline
Activity: 69
Merit: 10
|
|
February 28, 2014, 02:50:26 PM |
|
Why are everyone's 780s or Titans so slow? :p Mine can do 700kh/s without breaking a sweat...and more. Teach me master!
|
|
|
|
patoberli
Member
Offline
Activity: 106
Merit: 10
|
|
February 28, 2014, 03:33:01 PM |
|
Even better, save it in a csv format to make it maybe comparable.
And you're right, I mixed keccak with scrypt-jane, silly me.
|
YAC: YA86YiWSvWEGSSSerPTMy4kwndabRUNftf BTC: 16NqvkYbKMnonVEf7jHbuWURFsLeuTRidX LTC: LTKCoiDwqEjaRCoNXfFhDm9EeWbGWouZjE
|
|
|
trancelord
Newbie
Offline
Activity: 19
Merit: 0
|
|
February 28, 2014, 04:03:10 PM |
|
Hey guys, I have a question.
how do you set cudaminer to switch to pools automatically?
like say a pool doesnt reconnect, you can autoswitch to another pool after a while.
i know cg miner does it.. but how can u do it on cuda? what is the command?
tryin to set it up and run it automatically cuz i gotta work tonight and im mining something.
thanks
|
|
|
|
ManIkWeet
|
|
February 28, 2014, 04:07:14 PM |
|
Hey guys, I have a question.
how do you set cudaminer to switch to pools automatically?
like say a pool doesnt reconnect, you can autoswitch to another pool after a while.
i know cg miner does it.. but how can u do it on cuda? what is the command?
tryin to set it up and run it automatically cuz i gotta work tonight and im mining something.
thanks
You can set a specific amount of retries, and then make cudaminer terminate automatically when those retries have ran out. Use -r # for the retry count and -R # for retry seconds Examples: :loop cudaminer.exe -R 5 -r 10 -o pool1.com -O Acc.Worker:Pass cudaminer.exe -R 10 -r 5 -o pool2.com -O Acc.Worker:Pass cudaminer.exe -R 15 -r 2 -o pool3.com -O Acc.Worker:Pass goto loop
|
BTC donations: 18fw6ZjYkN7xNxfVWbsRmBvD6jBAChRQVn (thanks!)
|
|
|
Silverwolf_Ru
Full Member
Offline
Activity: 120
Merit: 100
Astrophotographer and Ham Radioist!
|
|
February 28, 2014, 04:10:21 PM |
|
Good evening everyone. Nice to see you all again. Ready for the Helixcoin launch? I might only have a 560 Ti but I will mine mine mine. Won't settle until I have at least one hundred thousand coins. All thanks to Mister Christian here. Should I try and compile the current branch Christian, or will you do it? Five hours until launch.
PS! Any chance of further Fermi optimizations? Even GTX 750 Ti are bloody expensive over here in Russia.
|
Bitcoin: 17kz4pWKoMoVupGUYgj8kGomxXUkDHNtVe Shadowcoin: Seta8CFwP6yvbeCkgfjxXjpkokrQMQovGF ~Coin of the Future!
|
|
|
ManIkWeet
|
|
February 28, 2014, 04:12:19 PM |
|
Ready for the Helixcoin launch?
Yep, though they haven't released any nodes to use for "addnode=IP" configs
|
BTC donations: 18fw6ZjYkN7xNxfVWbsRmBvD6jBAChRQVn (thanks!)
|
|
|
|