Topic: hashkill windows version needed (Read 2638 times)
Ahh ok, thank for the explanation.
Well yes - you see you run your python miner and it uses a single device only. If you have e.g 4 GPUs, you would run 4 instances. That's no problem because those are 4 separate processes each of them having its own context and queue, operating on a single device only.
Hashkill works on all GPUs and creates one (or more) threads per device, each of them having its own context and own queue, all of them working withing a single process (single address space). On linux, this works if you have SDK 2.3 or higher. On Windows, this would still work, however performance would not be optimal and CPU usage would be always ~100%.
On single-GPU systems, that would not matter much. On multi-gpu systems on Windows (or Linux with older SDK) this sucks. And I can do nothing about it.
Hashkill works on all GPUs and creates one (or more) threads per device, each of them having its own context and own queue, all of them working withing a single process (single address space). On linux, this works if you have SDK 2.3 or higher. On Windows, this would still work, however performance would not be optimal and CPU usage would be always ~100%.
On single-GPU systems, that would not matter much. On multi-gpu systems on Windows (or Linux with older SDK) this sucks. And I can do nothing about it.
....is useless as kernel invocations are to some extent serialized and CPU usage is high due to spinlock usage in the ocl runtime.
... wat.
Heh, well keep up all the good work, we who cannot code thankye for all the work you do.
It's a conspiracy to drive miners away from proprietary platforms. I also get a percentage from FSF's donations in BTC for doing this 
Now seriously I would not mind someone porting that to Windows as I have no plans to do that soon (have other more important stuff to do - it is not about bitcoin mining only, but also password cracking and adding new algos is a priority over porting to windows. I also happen to have my own life, a full time job and a kid to take care of). However, porting this to Windows is not gonna be easy as the code itself is messy and depends on linux-specific stuff a lot. It was never intended to be ported to win.
Also, no, the performance advantage would be no more than 3% as compared to other miners. With the latest version, I had to focus on correctness and feature support over performance so the hashing speed kinda decreased using the "fast" command-line options. So no way you will be getting 5% more than phatk for example, except on 69xx hardware probably.
Also, it would not run nicely on multi-gpu systems on Windows and the reason for that is that the windows opencl runtime does not honor GPU_USE_SYNC_OBJECTS. Until AMD fixes that on Windows, supporting multi-gpu configurations from inside a single-process application (even multithreaded) is useless as kernel invocations are to some extent serialized and CPU usage is high due to spinlock usage in the ocl runtime.
Now seriously I would not mind someone porting that to Windows as I have no plans to do that soon (have other more important stuff to do - it is not about bitcoin mining only, but also password cracking and adding new algos is a priority over porting to windows. I also happen to have my own life, a full time job and a kid to take care of). However, porting this to Windows is not gonna be easy as the code itself is messy and depends on linux-specific stuff a lot. It was never intended to be ported to win.
Also, no, the performance advantage would be no more than 3% as compared to other miners. With the latest version, I had to focus on correctness and feature support over performance so the hashing speed kinda decreased using the "fast" command-line options. So no way you will be getting 5% more than phatk for example, except on 69xx hardware probably.
Also, it would not run nicely on multi-gpu systems on Windows and the reason for that is that the windows opencl runtime does not honor GPU_USE_SYNC_OBJECTS. Until AMD fixes that on Windows, supporting multi-gpu configurations from inside a single-process application (even multithreaded) is useless as kernel invocations are to some extent serialized and CPU usage is high due to spinlock usage in the ocl runtime.
I support this initiative.
If hashkill is indeed faster (5% or more) a donation will certainly be sent.
If hashkill is indeed faster (5% or more) a donation will certainly be sent.
hashkill, says it is faster than other miners.
But it is only LOCKED to Linux as developer "gat3way" is an opensource Linux fan or may be don't know Windows programming.
Some one with windows programming can develop windows version of hashkill in command prompt or better in GUI itself.
For this i hope a bounty can be set, so some programmers write windows version.
But it is only LOCKED to Linux as developer "gat3way" is an opensource Linux fan or may be don't know Windows programming.
Some one with windows programming can develop windows version of hashkill in command prompt or better in GUI itself.
For this i hope a bounty can be set, so some programmers write windows version.
Jump to: