Does anyone think/thought about using sse/avx technology in ccminer ? sp_ ?

I know that ccminer spends a little time @CPU comparing to GPU but still there is a room for improvment.

thanks. but where it comes from ? why it is not 1024x1024x64 ?
is it e.g. maximum x-, y-, or z-dimension of a grid of thread blocks or something like that ? hardware related ?

what is
in the ccminer code
and why it is what it is ?
I mean 256x256x8 ?

thanks.

look at my instructions too
https://bitcointalk.org/index.php?topic=167229.msg8195127#msg8195127

bigjme: wish you 1/3 month wages ;-)

Questions for developers:
1. why memory for hash is not allocated in GPU shared memory but in global ?
2. can OpenMP be used to parallelize tasks on CPU ? e.g. validation.
3. did anyone thought of splitting communication host<->GPU into 2 separate threads ? basically one thread constantly and asynchronously reads data from stratum/pool and copies it to an array located in GPU, GPU kernel does all calculations and save it to an array @GPU, second thread constantly and asynchronously reads calculated data from an array @GPU to an array @host, validates them @CPU and sends it over the network to the pool. the idea is to avoid GPU kernel from waiting for host.

what more experienced developers think about those ideas ? I am lousy programmer so do not expect proof of concept from my side soon :-(

Experts and experienced programmers,

I want to add some effort to ccminer development. Can you help me and point where to start ? I have some C/C++ experience but have never used CUDA.

Thanks in advance.

 
just added link to OC howto: http://www.overclock.net/t/504832/how-to-nvidia-overclocking-in-linux

A start for 'Compile under Linux FAQ':

Based on Linux Mint (Ubuntu clone) x86-64:
- install >340 driver from nvidia or your distro repository
- install nvidia_uvm driver from your distro repository
- install CUDA toolkit 6.0 from nvidia
- downloaded current ccminer source from github (e.g. cbuchner1, djm34, tsiv)
- autoconfig.sh
- configure --with-cuda=/usr/local/cuda-6.0
- make -j <number of your CPU cores> (e.g. make -j8)
- export LD_LIBRARY_PATH=/usr/local/cuda-6.0/lib64
- export PATH=$PATH:/usr/local/cuda-6.0/bin
- ccminer -a x11 -benchmark

In case of runtime errors make sure:
- nvidia and nvidia_uvm modules loaded into kernel you can add 'nvidia_uvm' string to /etc/modules to have nvidia_uvm loaded automatically during start
- /dev/nvidia, /dev/nvidia0 exist. Should be created by kernel modules
- /dev/nvidia_uvm exists. Can be created with:
sudo mknod -m 660 /dev/nvidia-uvm c 250 0
sudo chmod o+rw /dev/nvidia-uvm

If you want quick compilation (for compute_50) modify Makefile.am: change all compute_30/sm_30 into compute50/sm_50 and get rid of compute_35/sm35 strings. It has to be done before running autoconfig.sh.
Thanks to that compilation takes ca 15 minutes but ccminer will work only on maxwell cards with compute 5.0.

Link to good OC @Linux howto:
http://www.overclock.net/t/504832/how-to-nvidia-overclocking-in-linux

redirect stdout+stdin to a file instead /dev/null. you can truncate the file periodically to avoid excesive growth.

Linux is always slower than windows but more responsive with running ccminer though. I have both OSs on the same hardware and linux do ca 5% less hashes. I use Linux Mint x86-64, CUDA 6.0 and 340.2x (do not remember last digit) driver.

OK, I figured it out myself. Here is what is needed in order to have ccminer working @linux once you have compiled the ccminer.
#1 except nvidia module nvidia_uvm module needs to be loaded in the kernel. you can do it adding 'nvidia_uvm' string to /etc/modules (@linux mint) and nvidia_uvm will be loaded automatically. nvidia_uvm is a separate package @linux mint.
#2 /dev/nvidia_uvm device needs to be created and accessible to 'others'. you can do it this way:
and you can run ccminer as a user - not root.

if you want quick compilation modify Makefile.am: change all compute_30/sm_30 into compute50/sm_50 and get rid of compute_35/sm35 strings. thanks to that compilation takes ca 15 minutes but ccminer will work only on maxwell cards with compute 5.0.

in my case I get ca 2300kh/s instead of 2600 @win8 for x11 algo. but @win I add 100MHz to GPU.

I am stuck with ccminer compilation @linux (linux mint actually). Please help me.

- installed 340.24 driver from xedgers repo
- installed CUDA toolkit 6.0 from nvidia
- downloaded current ccminer source from github (once cbuchner1, once djm34)
- autoconfig.sh
- configure --with-cuda=/usr/local/cuda-6.0
- make -j8
- export LD_LIBRARY_PATH=/usr/local/cuda-6.0/lib64
- export PATH=$PATH:/usr/local/cuda-6.0/bin
- ccminer -a x11 -benchmark with or without exiting X

"Unable to query CUDA driver version! Is an nVidia driver installed?"
:-(

/dev/nvidia* exists

any ideas what I did wrong ?

I use ccminer compiled by crypto mining blog. it's hashrate depends a little on cpu/motherboard (pc in general) speed not only the graphics card. e.g.
@intel2core duo x11 hashrate is ca 100kh/s lower than on intel i7.
so
@intel2core it is 2500-2600 when @i7 it is >2600.

I bought 5 MSI cards because were the cheapest. I have been using them for mining for ca 4 months 24/7 with no problems at all. hashrates:
x11 ca 2500-2600
x13 ca 1900-2000
@win7

heh, dead simple for who did it once ;-)

I know I need CUDA Dev Kit. I want to use 6.5 to have faster compilation and newest nVidia driver: 340.52. What repository should I use ? djm34 or cbuchner1 ? djm34 implemeted many of recent improvments so I guess I should use his repo. Am I right ?

Nice to read your posts pros! I envy your programming skills 

I got enough of Windows-nothing-having-OS and want to switch to Linux. I plan to use Linux Mint what is your opinion ? Maybe other distro is better for mining ? I cannot use KopiemTu because I am selling my hashing power to NiceHash with my own multialgo script.

I am looking for how-to regarding ccMiner/nvMiner compilation @Linux but cannot find any. Can someone share his experience ?

Thanks in advance.

hello!

can anyone tell me the hashrate of gtx 760 ? e.g. for x11 and x13.

thanks.

BC - it's gonna be huge

well maybe this is an opportunity for next ebay ? or ebay replacement ?

hey google ! wake up ! do something !