tsiv
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 07:30:51 PM |
|
http://pastebin.com/sV7C2TWRTry that for Hamsi. It sure is not pretty but it's a whole lot faster than the one I originally left X13 with. My notes say +59% on the 750 Ti ![Tongue](https://bitcointalk.org/Smileys/default/tongue.gif)
|
|
|
|
SavellM
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 07:33:26 PM |
|
Hey sp_ quick question.
I've just downloaded your Release10 (my first download of yours). What is the difference between ccminer and ccminer12?
Also it says that its missing msvcr120.dll
I managed to copy this from tpruvot.
Is this correct? Or do you have a better one to use?
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 07:40:43 PM |
|
ccminer12 is for compute 5.2 cards. 970/980. Since this is just a build from the github it doesn't contain the dependency files. You can download the files in the latest tvpovot release.
|
|
|
|
SS2006
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 07:59:31 PM |
|
what is the range of intensity parameter we can set?
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 08:28:01 PM |
|
http://pastebin.com/sV7C2TWRTry that for Hamsi. It sure is not pretty but it's a whole lot faster than the one I originally left X13 with. My notes say +59% on the 750 Ti ![Tongue](https://bitcointalk.org/Smileys/default/tongue.gif) Yes, faster indeed! Thanks, I will add it. Tested it to be 14.34% faster than my last version. (750ti)
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 08:29:16 PM |
|
what is the range of intensity parameter we can set?
try. max is 30, but the miner will probobly crash and spit out wild hashes. on the 970 20 or 21 @x11? Default is 19
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 08:30:23 PM |
|
Release 10 works perfectly. No problems whatsoever. Hashrates (750Ti O/C) X11: 3000 kH/s X13: 2320 kH/s X15: 2000 kH/s
Congratulations, sp_. You have beaten the code I did for CUDA... I think a couple months ago now, on X11. I think all three. You should help me push it up to 3,5 MHASH. Still some easy pickings.
|
|
|
|
ryen123
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 08:45:03 PM |
|
Release 10 works perfectly. No problems whatsoever.
Hashrates (750Ti O/C) X11: 3000 kH/s X13: 2320 kH/s X15: 2000 kH/s
What are your clocks with the 750Ti? I can get only 2950-2970KH/s with 1246MHz Core / 1580MHz Memory 1210MHz Core / 1600MHz Memory @jpouza I set -i 20 and managed to get the figures above.
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 08:45:52 PM |
|
Yes, faster indeed! Thanks, I will add it. Tested it to be 14.34% faster than my last version. (750ti)
Actually it was even faster with 64 registers But I guess your default already was 64 regs.. My default is 80, so I had to add launchbounds to force 64 registers, and running only with 32 threads. Then I get it to be 20% faster.
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 09:07:26 PM |
|
Commited tsivs hamsi, and removed memoryleaks in x11 and x13. Also merged in tvprovot's stratum bugfix.
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 09:09:28 PM |
|
@jpouza I set -i 20 and managed to get the figures above.
The default is set to -i=19 because -i=20 will cause out of memory on the 1GB cards (750,750ti) On 970/980 -i should be set to 20 or higher.
|
|
|
|
Epsylon3
Legendary
Offline
Activity: 1484
Merit: 1082
ccminer/cpuminer developer
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 09:24:41 PM |
|
http://pastebin.com/sV7C2TWRTry that for Hamsi. It sure is not pretty but it's a whole lot faster than the one I originally left X13 with. My notes say +59% on the 750 Ti ![Tongue](https://bitcointalk.org/Smileys/default/tongue.gif) Yep, i get much more with your code ![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif) i keep it tx. 2250KHs on a 750 Ti, vs 2150 before changes, and slower with sp commit (tried differents maxregcount on linux) edit: i used 72 registers, 64 is too low
|
|
|
|
jpouza
Legendary
Offline
Activity: 2730
Merit: 1116
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 09:37:53 PM Last edit: November 24, 2014, 09:55:30 PM by jpouza |
|
@jpouza I set -i 20 and managed to get the figures above.
The default is set to -i=19 because -i=20 will cause out of memory on the 1GB cards (750,750ti) On 970/980 -i should be set to 20 or higher. -i 21 on 980 boosted almost 200KH/s! -i 20 on 750Ti 2GB nothing changed, will try to increase memory clock. EDIT: nothing, still can't reach 3MH/s with 750 Ti, even with 1600MHz mem clock and -i 20. 980 just amazing, 9,8 MH/s achieved with -i 21 WOW subefotos
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 09:48:48 PM |
|
The best setting for the TI is betweeen 19 and 20, so I changed the default. Preparing a new build now. #11. We should add xintensity to ccminer for optimal performance.
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 10:02:33 PM |
|
http://pastebin.com/sV7C2TWRTry that for Hamsi. It sure is not pretty but it's a whole lot faster than the one I originally left X13 with. My notes say +59% on the 750 Ti ![Tongue](https://bitcointalk.org/Smileys/default/tongue.gif) Yep, i get much more with your code ![Smiley](https://bitcointalk.org/Smileys/default/smiley.gif) i keep it tx. 2250KHs on a 750 Ti, vs 2150 before changes, and slower with sp commit (tried differents maxregcount on linux) edit: i used 72 registers, 64 is too low I use launchbound 256,4 and only 32 threads. Seems to be the fastest on windows. I get no memory spills with 64 registers for compute 5.0, but for compute 3.0 there are spills (because of the missing funnelshift)
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 10:04:42 PM |
|
Eplyson, the API seems to fail when you run two instances of ccminer. Probobly because the default port is used.
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 11:01:10 PM |
|
I reset the default throughput to be the same as in build 8, it increases the x11 hash with around 30KHASH (750ti). Now trying out 72 registers and no launchbound as suggested @x13. But so slow to build...
|
|
|
|
Epsylon3
Legendary
Offline
Activity: 1484
Merit: 1082
ccminer/cpuminer developer
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 11:25:50 PM |
|
I reset the default throughput to be the same as in build 8, it increases the x11 hash with around 30KHASH (750ti). Now trying out 72 registers and no launchbound as suggested @x13. But so slow to build...
I why i prefer to do tests on linux when i can... I work over nfs on windows and build in putty ![Wink](https://bitcointalk.org/Smileys/default/wink.gif) Increasing too much intensity will make Windows very slow, its why i dont use the max possible as default. But i understand your ".5" intensity, could be done like that rounded to the nearest multiple of 256 for the api its normal... "-b 0" to fully disable it, or set a custom port if required... There is only 4 tries to bind the first minute, then stops. Nice also to know there is another instance running ![Wink](https://bitcointalk.org/Smileys/default/wink.gif)
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 24, 2014, 11:32:17 PM |
|
yes, this is how the X-intensity is implemented in sgminer. Just Multiply the factor by 256.
-xi 128 = gives throughput of 128*256
or perhaps the constant is more than 256. Don't remember. But the idea is to multiply the constant with the inputparameter.
|
|
|
|
sp_ (OP)
Legendary
Offline
Activity: 2912
Merit: 1087
Team Black developer
|
![](https://bitcointalk.org/Themes/custom1/images/post/xx.gif) |
November 25, 2014, 12:19:55 AM |
|
Buildt a new version with tsivs x13 improvements. Some bugfixes, and bether default intensity for 750ti (I get +30 KHASH) version 10 had a memory leak. If the miner was running for a long time, the performance went down. The 11th public beta(windows executable) is available here: (11-25-2014) http://www.filedropper.com/release11The sourcecode is available here: https://github.com/sp-hash/ccminer
|
|
|
|
|