sia did that for both - and nvidia cards came out far worse than implementing cuda ...
but there you have it ...
#crysx
When did he do that?
I can only hope that NVidia's implementation of OpenCL has improved by now.
It's their unwillingness to participate in an open standard that makes their OpenCL implementation so "average".
the one miner ...
it is opencl and runs for both nvidia and amd based gpus ...
nvidia came out poorly in the trade off ...
https://github.com/NebulousLabs/Sia-GPU-Miner ...
the algo is blake2b from what i understand ...
whether its their unwillingness or not - like microsoft and their ridiculous 'standards' as opposed to the 'real' standards - nvidia are the largest gpu manufacturer and developer in the world ...
so they build their own standard - which could be the crappiest coding on the planet ( jump in microsoft again ) but still maintain the number one position for the hardware and software working together ...
#crysx
It's Nvidia's fault. They wanna pretend they're Apple and not give a shit about interoperability, then it should be their problem.
EDIT: I dumped the signature2 kernel from my miner - it was triggering a bug in my other kernels and didn't really make sense to seperate anyways.
I see Wolf beat me too it but it was only about 3 maybe 6 months ago Nvidia started really supporting OpenCl but not the full version of it from my understanding. And Windows 10 with Nvidia is giving the main support. I am no where near a dev so you can look into it further if you like.
Their cards have supported OpenCL pretty much always - but only 1.0 or 1.1 - AMD has supported 1.2 for AGES, and 2.0 is out - they support that, as well.
Saved a metric shitton of GPU memory - how much depends on what you fed to the (fairly stupid) thread concurrency parameter. Now, to dump that param entirely (along with lookup gap, which does nothing for the algo and serves simply to determine how much memory you waste on useless allocation,) and then take a look at parallelism possibilities.
ok- but thats NOT what i was getting at ...
i dont like non-standardization as much as the next guy - BUT - the point of the matter is that we are STUCK with what we are given by nvidia ... just like STUCK with microsoft garbage for SO long ...
so irrelevant whose fault it is - its a matter of working with what we have ...
i totally agree with you both - i think nvidia should stop playing apple and microsoft and start playing oss properly ...
its such a pity amd drivers are a massive crap to install on systems as well ... the drivers and support for oss implementations are great - but the installation routines and support on that side suck the proverbials ... nvidia drivers install seamless ...
when there is a standar we can ALL work from ( yes i know - opencl IS out and available - before you both hammer me for saying this ) - AND companies 'conform' to the standards AND support them - then there will always be a rift in what software we can use for one or the other ...
so - irrespective of who is at fault - we are stuck with what we have with amd and nvidia and microsoft and linux ...
i just cant wait for the time when we can use ONE miner to do ALL the mining on ALL the os systems ... IF that ever happens of course
...
wolf - the code for spreadx11 has been hammered for ages and i know you have made some improvements / optimizations to it ... but how will it fare insofaras mining goes? ... is it better than the private miner sp distributes? ... which as you know is based on tsivs miner ( ccminer-spreadminer ) ...
im online now anyway - so available for a chat ...
#crysx
