Quote from: Thirtybird on March 11, 2014, 07:47:23 PM

Quote from: Rabbit80 on March 11, 2014, 06:21:26 PM

Fair enough.. However the R7 240 only has 320 stream processors, the 7750 only 512. This allows much lower TC without errors for some reason. The 290 has 2560 stream processors. With a TC of 28000, a lookup gap of 2 and intensity down at 12 it only manages 75KH/s. Any increase in intensity then causes hw errors.

Using my earlier config of TC 42000, lookup gap 3 and intensity 19, i can get nearly 150KH/s. Increasing system RAM to 8GB and lowering lookup gap back to 2 gives me around 190KH/s.. (getting a new board in the next couple of days with more RAM on it - YAY!!)

You obviously don't realize that thread-concurrency actually has nothing to do with shader count when it comes to scrypt-chacha. Thread-concurrency only sets the scratchpad buffer size (TC*128 / (1024/LG) = #MB in the buffer (with some rouding on Lookup gaps not evenly divisible into 1024). I don't use thread-concurrency actually - my dev version (The 3_4_3-Testing branch on my github) finally supports setting just a buffer size - internally, it does set the TC value which is what gets passed to the scrypt-chacha openCL kernel program though. So my command line looks really strange compared to yours I'm sure

yacminer -d 0,1 --remove-disabled --scrypt-chacha --worksize 128 -g 2 --lookup-gap 2 --raw-intensity 640 --buffer-size 1520 --lots-of-other-stuff

You are right about one thing - with 4 GB of system RAM, I cannot allocate enough system memory to allocate 3.6 GB per card when initializing them. I can, however, allocate 1.5GB 2 times on each card just fine without any loss in performance (from what a colleague has tested on the same card allocating the whole buffer for 1 thread - he also gets 2.9 KH/sec).

With that said, the one thing from above that allows you to do that is :
-g 2
This runs two separate threads within the miner. You cut your TC in half and then subtract a bit for headroom to ensure it isn't putting one of your threads in dynamic GPU memory. If you aren't using YACMiner, you're missing out on being able to fine tune your intensity using --raw-intensity. Using a miner with just -I, I would have to choose -I 9 (which correlates to --raw-intensity 512), I would be losing performance as I am able to launch an additional 128 shader threads per thread per card(128*2*2 = 512 shader threads, or roughly 25% performance). Now, this card, even at N=14 it is actually shader limited. On cards with higher shader counts, you do need to be able to set a higher buffer-size (meaning more system ram), or increase the LG (as you are doing), or run more threads (-g 2).

Experiment - tuning cards for high N factor takes experience, patience, and a willingness to try lots of different configurations. In my case, I found a better way to control the GPU and made the correlation to how it works at higher N factors. I made the software updates, and those are available to anyone.

O mighty Thirtybird GOD of Hash speed !!

Still trying to sort out maximum hash speeds from my R9 290's

(TC*128 / (1024/LG) = #MB in the bufferYACMiner

what is the 128 ? is that work size ?

128*2*2 = 512 shader threads
whats the 2x2

YACMiner would that work with https://phpminer.com/
or is their any web tools available for YACMiner rigs ?

you might be able to explain this to me

R9 290's HW error TC 32765 any intensity above I 12 hash speed around 80 kh/s
yet
no HW error when TC 42000 I 20 ?? at around 160 kh/s

now your saying if I add GPU threads I must halve TC ? ( in the past I only dropped I 1 level )
another thing thats got me stumped I have multiple identical rigs and I found with TC 42000 I 20
if I used 2x threads and dropped I to 19 my reject rate dropped kh/s same and if I used 4x threads dropping I 2 levels reject rate dropped even more same hash speed

also with YACMiner

./autogen.sh
CFLAGS="-O2 -Wall -march=native" ./configure

what are the options ??

I am assuming same as cgminer

CFLAGS="-O2 -Wall -march=native -I/opt/AMDAPP/include" LDFLAGS="-L/usr/lib" ./configure --enable-opencl --enable-gpu --enable-scrypt-chacha --enable-scrypt

Huh

configure: WARNING: unrecognized options: --enable-gpu, --enable-scrypt-chacha

Configuration Options Summary:

curses.TUI...........: FOUND: -lncurses
OpenCL...............: NOT FOUND. GPU mining support DISABLED
configure: error: No mining configured in

also tried

CFLAGS="-march=native -O2 -pipe" ./configure --enable-scrypt

I had the same issue with cgminer not seeing GPU's but the --enable-gpu option fixed that for me ..

sudo apt-get install ocl-icd-opencl-dev