Topic: [ANN] Kryptohash | Brand new PoW algo | 320bit hash | ed25519 | PID algo for dif - page 30. (Read 149437 times)

I meant to say max group work size.  Sorry.

Delete the .bin and try --worksize 1024 and see what happens.


Also, if you want your 750s with a different worksize value, you can use for example: --worksize 1024,256,256,256

I see Preferred vector but not worksize, I do see max work size reported though:

[2015-01-03 17:55:10] CL Platform vendor: NVIDIA Corporation
 [2015-01-03 17:55:10] CL Platform name: NVIDIA CUDA
 [2015-01-03 17:55:10] CL Platform version: OpenCL 1.1 CUDA 6.5.30
 [2015-01-03 17:55:10] List of devices:
 [2015-01-03 17:55:10]    0   GeForce GTX 970
 [2015-01-03 17:55:10]    1   GeForce GTX 750 Ti
 [2015-01-03 17:55:10]    2   GeForce GTX 750 Ti
 [2015-01-03 17:55:10]    3   GeForce GTX 750 Ti
 [2015-01-03 17:55:10] Selected 0: GeForce GTX 970
 [2015-01-03 17:55:18] Preferred vector width reported 1
 [2015-01-03 17:55:18] Max work group size reported 1024
 [2015-01-03 17:55:18] Max mem alloc size is 1073741824
 [2015-01-03 17:55:18] GPU 0: No Dynamic Interval was set. Disabling Dynamic Intensity
 [2015-01-03 17:55:18] GPU 0: No Intensity value was set. Disabling Intensity
 [2015-01-03 17:55:18] No binary found, generating from source
 [2015-01-03 17:55:18] Setting worksize to 256
 [2015-01-03 17:55:18] cl_amd_media_ops not found, will not set BITALIGN
 [2015-01-03 17:55:18] BFI_INT patch requiring device not found, will not BFI_INT patch
 [2015-01-03 17:55:18] CompilerOptions: -D WORKSIZE=256
 [2015-01-03 17:55:18] Binary size for gpu 0 found in binary slot 0: 423533
 [2015-01-03 17:55:18] Initialising kernel kshake320-546-uint2.cl without bitalign, 1 vectors and worksize 256
 [2015-01-03 17:55:18] Creating kryptohash buffer size 872415232
 [2015-01-03 17:55:18] initCl() finished. Found GeForce GTX 970
 [2015-01-03 17:55:18] Pushing sem post to thread 0
 [2015-01-03 17:55:18] Init GPU thread 1 GPU 1 virtual GPU 1
 [2015-01-03 17:55:18] CL Platform vendor: NVIDIA Corporation
 [2015-01-03 17:55:18] CL Platform name: NVIDIA CUDA
 [2015-01-03 17:55:18] CL Platform version: OpenCL 1.1 CUDA 6.5.30
 [2015-01-03 17:55:18] List of devices:
 [2015-01-03 17:55:18] opencl_thread_init: thread id=0 using kshake320v2
 [2015-01-03 17:55:18]    0   GeForce GTX 970
 [2015-01-03 17:55:18] Waiting on sem in miner thread
 [2015-01-03 17:55:18] Popping work from get queue to get work
 [2015-01-03 17:55:18] Got work from get queue to get work for thread 0

Add these two paramters to the command line:   --debug  2>log.txt

And look inside log.txt  what is the Preferred Work Size for the 970.

Tried changing the --worksize to 128 and 512 same issue with HW error. I am not sure if there are other multiples to try with it.

I have also been deleting the .bin files everytime I run the bat.

There is no need to go back to beta2. The Kernel is the same.

You can remove the '--cl-opt-disable' option, delete the 'kshake320-546-uint2{GPUNAME}v1w256i4.bin' file and cgminer beta3 should run just like the beta2.  
 
Or, you can take advantage of the Intensity setting.  For your R270, you can use intensity 3 or 4 on each card.  Try specifying:  -I 3  or -I 4



As I said, the Kernel hasn't changed.  If the '--cl-opt-disable' option isn't helping on the 970, just remove it and cgminer should pretty much work as before.  Don't forget to delete 'kshake320-546-uint2{GPUNAME}v1w256i4.bin' every time you make a change so, cgminer is forced to re-compile the Kernel next time it is executed.

Also, try changing the Work Size using the '--worksize' option on your 970.  The default is now 256 and your nVidia GPU might not like that value.

 

cgminer is dead on my R280x after mining for half an hour.

@wr104

I am also getting the HW errors again with the Nvidia cards.  Was there something besides the .bin files that needed to be deleted?

EDIT: It works fine for my 750ti's. Getting around 80 kh/s. Up from about 50 kh/s. The 970 gets HW errors though. Not sure what is going on with tat yet.

I've switched to beta2,I am worried about my power supply.

Power of my 6*R270 mining rig increase from 620watt to 850 watt,hashrate from 90kh/s to 133kh/s.

About the Intensity feature:

The implementation is different from the traditional one found in cgminer with scrypt support.  

In the cgminer-khc case, the Minimum Intensity is always 2 and the Maximum Intensity depends on what GPU you have.  The Max is calculated as the number of Stream Processors divided by the Work Size (default: 256).

Examples of Max Intensity using 256 for work size:


R9 290x:  2816 / 256 = 11
R9 290 :  2560 / 256 = 10
R9 280x:  2048 / 256 = 8
HD7970 :  2048 / 256 = 8
HD6970 :  1536 / 256 = 6


Using a number for Intensity higher than the values shown above will have no effect on these GPUs.

Finally, the Dynamic Intensity feature is disabled by default.  It doesn't work as expected.

catalyst 14.12, underclocked/undervolted R9 280x ~234Kh/s

@ wr104

Quick test on GTX970 with the --cl-opt-disable crashed the driver.

Will play with it more later when I get back in.

EDIT: Huge speed increase in the 280X without the  --cl-opt-disable.

Your 5970 is VLIW5. I don't think disabling CL optimization will help on this GPU.   I know that on my 6850 with VLIW4, the option doesn't help.

The switch only helps if you have an AMD GPU with GCN architecture, like the R9 200 series.


My R9 290x went from 205Kh/s to 348Kh/s.

5970,AMD 13.12 = 20kh/s decrease 12x from Beta2 .
What is your hashrate wr104 ?

New cgminer-khc v3.7.6 Beta3 available.

https://github.com/kryptohash/cgminer-khc/releases/tag/v3.7.6-Beta3


Changes since 3.7.6 Beta2

- Added the '--cl-opt-disable' option that disables OpenCL optimization for kernels. Disabling it could significantly increase performance of the kshake320 algorithm on some high end AMD GPUs with GCN architecture, like the HD7970 and the R9 200 series. (Tested using AMD 14.12 Omega Drivers).

   Example:
      cgminer --kryptohash --kernel kshake320v2 --cl-opt-disable -o http://127.0.0.1:38912 -u user -p password --shaders {xxx} --shaders-mul {yyy}

- Added back the GPU Intensity parameter.


Important Notes:

1) Disabling OpenCL optimization doesn't guarantee better performance on every AMD GPU.  nVidia GPUs were not tested.

2) Every time you add/remove the '--cl-opt-disable' switch, ensure you also delete the old 'kshake320-546-uint2{GPU_NAME}v1w256i4.bin' file in order to allow cgminer to recompile the OpenCL kernel.

I can't login in lazycoins

Has anyone been able to withdraw from LazyCoins since the holidays?  I've had one "Pending" for a day now.

I've added the two exchanges to the OP.

I will release the new cgminer-khc tomorrow.  I'm still debugging a couple of issues.

Development, please add trading platform OP, thank you! No transaction volume trading platform

Thus is true, but that only means someone could get more blocks and does not affect the security of the blockchain, presently the only pool in existence is running at -8.2MH/s where as solo miners account for about 1/3 of the total network hash rate.  View the blocks solved on the pool and note that there are gaps in there that solo miners are getting.  Also take into account that anyone has the ability to mine on the pool.  It is the miners choice in the matter, same can be said for any coin out there.