Experimented more. If I have -C 0 it assigns 1,75 GB Ram, if it's -C 1 it assigns 1,13 GB and -C 2 also 1,13 GB for auto tune.
Fastest I found was
cudaminer.exe -a scrypt-jane -i 0 -l K27x2 -H 2 -d 1
Which provides a stable 1.45 kh/s. Still experimenting a little bit more.
Topic: [ANN] cudaMiner & ccMiner CUDA based mining applications [Windows/Linux/MacOSX] - page 975. (Read 3426921 times)
January 29, 2014, 07:13:08 AM
January 29, 2014, 07:03:25 AM
I received my Geforce GT 640 with 2 GB of Ram today. Somehow I don't get on good speeds with scrypt-jane (Yacoin) with it. Best so far was 0.96 kh/s with
cudaminer.exe -a scrypt-jane -i 0 -l X23x3 -o http://yac.coinmine.pl:8882 -O pato.2:password -C 1 -L 2 -H 2 -d 1
Also autotune only assigns 1.135 GB Ram instead of the full 2 GB (this is with a build from yesterday, so around 24hrs old, without the recent autotune fixes).
cudaminer.exe -a scrypt-jane -i 0 -l X23x3 -o http://yac.coinmine.pl:8882 -O pato.2:password -C 1 -L 2 -H 2 -d 1
Also autotune only assigns 1.135 GB Ram instead of the full 2 GB (this is with a build from yesterday, so around 24hrs old, without the recent autotune fixes).
Yacoin on GT640: K kernel or T kernel (T for compute 3.5 only). Lookup gap 1 or 2. Don't go higher, the card doesn't have enough compute reserves. My GT 640 with 1GB RAM requires a lookup gap of 2, getting around 1.5-1.6 kHash/s. Your 2GB card might not need it. Best to try.
X is for Fermi.
Christian
January 29, 2014, 06:49:08 AM
I received my Geforce GT 640 with 2 GB of Ram today. Somehow I don't get on good speeds with scrypt-jane (Yacoin) with it. Best so far was 0.96 kh/s with
cudaminer.exe -a scrypt-jane -i 0 -l X23x3 -o http://yac.coinmine.pl:8882 -O pato.2:password -C 1 -L 2 -H 2 -d 1
Also autotune only assigns 1.135 GB Ram instead of the full 2 GB (this is with a build from yesterday, so around 24hrs old, without the recent autotune fixes).
cudaminer.exe -a scrypt-jane -i 0 -l X23x3 -o http://yac.coinmine.pl:8882 -O pato.2:password -C 1 -L 2 -H 2 -d 1
Also autotune only assigns 1.135 GB Ram instead of the full 2 GB (this is with a build from yesterday, so around 24hrs old, without the recent autotune fixes).
January 29, 2014, 06:14:53 AM
Anyone having proper startup arguments for GTX 660 with the latest version from github ? I can't seem to hit the same performance as older versions and autotune can't seem to find a sweet spot with the latest Y kernels either.
On older versions I was using :
cudaminer.exe --algo=scrypt-jane -d 0 -H 2 -C 0 -m 0 -b 32768 -L 3 -l K10x18 -s 120 -o stratum+tcp://yac.coinmine.pl:9088 -O user:pwd
Also, weird thing, I was running same script on my main video card and it was hovering at ~3kh/s . After stopping it and restarting it , I got 2kh/s. No change in frequency or temperatures, what the ...
On older versions I was using :
cudaminer.exe --algo=scrypt-jane -d 0 -H 2 -C 0 -m 0 -b 32768 -L 3 -l K10x18 -s 120 -o stratum+tcp://yac.coinmine.pl:9088 -O user:pwd
Also, weird thing, I was running same script on my main video card and it was hovering at ~3kh/s . After stopping it and restarting it , I got 2kh/s. No change in frequency or temperatures, what the ...
January 29, 2014, 05:35:10 AM
cbuchner1, did you note my earlier post about autotune problems and K kernel performance regression?
Do you need any additional information to diagnose those problems?
Do you need any additional information to diagnose those problems?
Part of the problem is that previously the loop trip count N=1024 was hardcoded and the kernel always assumed to operate in a single, linear memory block. The process of making it more flexibe to work with any N value, and to operate also on chunked memory cost a bit of performance.
Because there is now a faster replacement kernel for scrypt called "Y" I will not be addressing the performance drop of the K kernel now (later, maybe...). "K" is still kicking butt in scrypt-jane with high N factors and with lookup gap. That's what you will want to use it for.
About autotune being wonky: Part of this can be attributed to the "boost" feature of the GPUs. these decide pretty randomly when to clock down and when to clock up. So the measured values can be jumping up and down pretty badly, making an accurate assesment very hard.
What I fixed yesterday was measurements showing "infinite" hashing speeds for very fast kernels on Windows, such as N-factor 7 or 8 scrypt-jane coins. These would always win over any correct measurements because apparently infinite is always bigger
Christian
January 29, 2014, 04:52:05 AM
By comparison with the recent code from github I am only getting 190kh/s when I use autotune. Crashes otherwise.
somehow running two or more GPUs in one cudaminer instance is no longer working.
workaround: run two cudaminers, one with -d 0, the other with -d 1
getting less kHash/s with the github version? consider using the Y kernel instead.
Christian
Thanks for that tip, Christian! My 770 SC went from 375 to 418. Good gain!
January 29, 2014, 03:25:10 AM
cbuchner1, did you note my earlier post about autotune problems and K kernel performance regression?
Do you need any additional information to diagnose those problems?
Do you need any additional information to diagnose those problems?
January 29, 2014, 02:06:15 AM
I forget now... and I can't easily search this thread for what I'm looking for... What exactly does -H 2 and -m do?
January 28, 2014, 10:26:38 PM
Sooo Christian, I don't want to be pushy, but any progress on that next version? Specifically I'm interested in a failover option.
Specifically I'm interested in a failover option.In my opinion failover at this stage is kind of a luxury, but the simplest way to do it was to set cudaMiner to exit if it crashes instead of getting stuck in a loop and spamming the screen (counting the number of total errors and if that number reaches X in the last Y seconds, exit) so you could use a batch file where you could just list all the failovers you want.
I was thinking about this - specifically for Linux - at the moment I am trying to cobble something together using swatch to monitor the output of cudaminer - configure swatch to take an action on seeing n amount of say "disconnect" or "timeout" and killing the current cudaminer before restarting on the next pool configured. In theory it should be fairly straightforward but I'm not having any joy with swatch in its "--read-pipe" mode, it doesn't seem to spot even the most simple of things. If I get it figured I'll post it up here.
January 28, 2014, 06:46:25 PM
Hmm I went back to YACoin after I got my 1LTC, found 1 block today...
Ever since that wallet changed it feels like less...
Edit:
Fine I will stop complaining
Ever since that wallet changed it feels like less...
Edit:
Fine I will stop complaining
lucky devil, since the wallet update my luck has been aweful
January 28, 2014, 05:42:11 PM
Just wanted to come and say thanks! Signed up just to do so. I'll make it a point to donate to the address given on the first page.
I got it running with a fresh clone and compile from github, running on Debian Wheezy 32-bit. I had to grab and install the 5.5 CUDA dev package from NVIDIA's site because the CUDA package available in the repos is archaic, because. well... Debian.. but it works like a charm. In case anyone comes across this and plans on using the same distro.
I got it running with a fresh clone and compile from github, running on Debian Wheezy 32-bit. I had to grab and install the 5.5 CUDA dev package from NVIDIA's site because the CUDA package available in the repos is archaic, because. well... Debian.. but it works like a charm. In case anyone comes across this and plans on using the same distro.
January 28, 2014, 05:25:50 PM
Hmm I went back to YACoin after I got my 1LTC, found 1 block today...
Ever since that wallet changed it feels like less...
Edit:
Fine I will stop complaining
Ever since that wallet changed it feels like less...
Edit:
Fine I will stop complaining
January 28, 2014, 05:22:08 PM
By comparison with the recent code from github I am only getting 190kh/s when I use autotune. Crashes otherwise.
somehow running two or more GPUs in one cudaminer instance is no longer working.
workaround: run two cudaminers, one with -d 0, the other with -d 1
getting less kHash/s with the github version? consider using the Y kernel instead.
Christian
Better with Y kernel:
Code:
[2014-01-28 16:17:18] GPU #1: GeForce GTX 660 Ti, 286.84 khash/s
[2014-01-28 16:17:28] Stratum detected new block
[2014-01-28 16:17:28] GPU #1: GeForce GTX 660 Ti, 286.53 khash/s
[2014-01-28 16:17:39] Stratum detected new block
[2014-01-28 16:17:39] GPU #1: GeForce GTX 660 Ti, 282.79 khash/s
[2014-01-28 16:17:46] Stratum detected new block
[2014-01-28 16:17:46] GPU #1: GeForce GTX 660 Ti, 280.32 khash/s
[2014-01-28 16:18:01] Stratum detected new block
[2014-01-28 16:18:01] GPU #1: GeForce GTX 660 Ti, 284.01 khash/s
[2014-01-28 16:18:04] Stratum detected new block
[2014-01-28 16:18:04] GPU #1: GeForce GTX 660 Ti, 280.76 khash/s
[2014-01-28 16:18:13] Stratum detected new block
[2014-01-28 16:18:13] GPU #1: GeForce GTX 660 Ti, 283.66 khash/s
[2014-01-28 16:18:15] Stratum detected new block
[2014-01-28 16:18:15] GPU #1: GeForce GTX 660 Ti, 279.79 khash/s
-H 1 -d 1 -i 0 -l Y7x32
January 28, 2014, 05:13:33 PM
I only found 2 blocks for yacoin in 16 hours today. The new wallet seems much slower for mining. I was getting 7 blocks a day and the new one is a lot slower. No idea why
January 28, 2014, 05:12:12 PM
I am giving up on Microcoin mining. Despite running 3.5 MHash/s I only found 2 blocks in 24 hours.
difficulty of Yacoin is going down, so I am back to Yacoin mining for now.
Christian
difficulty of Yacoin is going down, so I am back to Yacoin mining for now.
Christian
Hehh, YAC solomining variance screwed me out of a couple hundred coins and I just started solomining microcoin and found a block 5 hours ago and now 9 minutes ago (just checked for the sake of this comment and voila). Found a stale block too earlier.
Only have 369 kH/s average hashrate. The difficulty of microcoin is going down too (3.59 -> 2.42 in the last 22 hours or so), but that's obvious.
January 28, 2014, 05:07:52 PM
I am giving up on Microcoin mining. Despite running 3.5 MHash/s I only found 2 blocks in 24 hours.
difficulty of Yacoin is going down, so I am back to Yacoin mining for now.
Christian
difficulty of Yacoin is going down, so I am back to Yacoin mining for now.
Christian
January 28, 2014, 05:06:29 PM
By comparison with the recent code from github I am only getting 190kh/s when I use autotune. Crashes otherwise.
somehow running two or more GPUs in one cudaminer instance is no longer working.
workaround: run two cudaminers, one with -d 0, the other with -d 1
getting less kHash/s with the github version? consider using the Y kernel instead.
Christian
January 28, 2014, 05:05:30 PM
Getting some weirdness with my EVGA 660 GTX Ti's :
You're better off running two cudaminer instances, one for each card (-d 0 and -d 1 in another) and probably with using -H 2.
I saw a few pools which tracked shares/sec (not just hashes/sec) which I think is way more useful.
January 28, 2014, 04:50:39 PM
Getting some weirdness with my EVGA 660 GTX Ti's :
64-bit 12/18/2013 build
-a scrypt -H 1,1 -d 0,1 -i 0,0 -l K7x32,K7x32
EVGA Model: 02G-P4-4069-KB
It seems to accept the shares and it doesn't crash. But why?
By comparison with the recent code from github I am only getting 190kh/s when I use autotune. Crashes otherwise.
64-bit 12/18/2013 build
-a scrypt -H 1,1 -d 0,1 -i 0,0 -l K7x32,K7x32
EVGA Model: 02G-P4-4069-KB
It seems to accept the shares and it doesn't crash. But why?
By comparison with the recent code from github I am only getting 190kh/s when I use autotune. Crashes otherwise.
January 28, 2014, 04:28:25 PM
Sooo Christian, I don't want to be pushy, but any progress on that next version? Specifically I'm interested in a failover option.
Specifically I'm interested in a failover option.In my opinion failover at this stage is kind of a luxury, but the simplest way to do it was to set cudaMiner to exit if it crashes instead of getting stuck in a loop and spamming the screen (counting the number of total errors and if that number reaches X in the last Y seconds, exit) so you could use a batch file where you could just list all the failovers you want.
Jump to: