NVIDIA AND AMD GatelessGate

This is a test build for newest gatelessgate

there are 2 separate builds depending on your hardware.

NEW AMD/NVIDIA GATELESSGATE BUILD

This build includes the dropin replacement "gatelessgate" that was coded by zawawa. Currently my AMD R9290 does 150 sols/s, up from 70 sols/s from previous build.

CUDA version to follow in v0.5...

CUDA Compute 30 and Less Build, CUDA SA is default now

I created a separate dll for the OLD Compute 30 & 20 CUDA cards since it was by far the most performant for those platforms. You will have to delete cuda_silentarmy.dll and rename cuda_silentarmy_sm30.dll to cuda_silentarmy.dll

If you have one these cards please let me know the performance. On my GTX 650 I get 8 to 10 sols/s.

Silentarmy V5 CUDA enhancements

I have updated the cuda kernel with the enhancements to use 10xx specific ASM and more efficient band. usage.

Post some benchmarks to let me know if i made the changes successfully.

v0.4i "Works"

For any older than maxwell cards, it will run a mixture of v4 and v5 silentarmy, which is at least as good.
For any maxwell or newer, hopefully the fixes I put in will make it work at its full potential.

TODO: Work on the atomic code some more and see what else I can do for it

Im also including a handy debug tool.

New Debug Test Tool

I have wrapped the different methods into 1 debug tool that will run on all devices and show the sols/s per device per method.

You can adjust the run count with the -r command. eg. kakminer.exe -r 100

Let me know the results.

v0.4h CUDA silent army v5

Adding a test program that just performs test on the first nvidia gpu available (maztheman_t1.7z)
PLEASE RUN LIKE THIS:

kakminer.exe 1>test.log

Post your log files. This will greatly help me in figuring out what hardware can do what.

v0.4g STABLE NVidia Build

With this release I have removed some cuda code that I think was limiting the amount of block/thread combos.
I also fixed the -cs switch so it can be used with any combo of cpu flags or ocl flags.

FOR MULTI GPU TESTERS:
to select a different block size and thread size for each card try this:
If you had three cards:
-t 0 -cb 64 64 128 -ct 32 32 64 -cs -cd 0 1 2

#Added
Gtx 500 series compatibility with v0.4g_sm20

v0.4f More Silentarmy V4 Optimizations!!

Optimization: significantly reduce CPU usage and PCIe bandwidth (before:
~100 MB/s/GPU, after: 0.5 MB/s/GPU), accomplished by filtering invalid
solutions on-device

Try the "v0.4f_SM50_62.7z" for GTX 900 series and GTX 1000 series

v0.4e Quick Release for update to silent army (CUDA and OCL)

Just a quick update for some more optimizations