2>Done building project “Theta.csproj”.
1>D:\Henry\Mining\Grin\GrinGoldMiner-master\GrinGoldMiner-master\src\Cudacka\Cudacka.vcxproj(32,3): warning MSB4011: “D:\Henry\Mining\Grin\GrinGoldMiner-master\GrinGoldMiner-master\src\Microsoft.VCToolsVersion.14.11.props” cannot be imported again. It was already imported at “D:\Henry\Mining\Grin\GrinGoldMiner-master\GrinGoldMiner-master\src\Cudacka\Cudacka.vcxproj (17,3)”. This is most likely a build authoring error. This subsequent import will be ignored.
1>D:\Henry\Mining\Grin\GrinGoldMiner-master\GrinGoldMiner-master\src\Cudacka\Cudacka.vcxproj(96,3): warning MSB4011: “D:\Henry\Mining\Grin\GrinGoldMiner-master\GrinGoldMiner-master\src\Microsoft.VCToolsVersion.14.11.props” cannot be imported again. It was already imported at “D:\Henry\Mining\Grin\GrinGoldMiner-master\GrinGoldMiner-master\src\Cudacka\Cudacka.vcxproj (17,3)”. This is most likely a build authoring error. This subsequent import will be ignored.
1>C:\Program Files (x86)\Microsoft Visual Studio\2017\Community\Common7\IDE\VC\VCTargets\BuildCustomizations\CUDA 9.1.targets(428,9): error MSB6003: The specified task executable “cl.exe” could not be run. The system cannot find the file specified 1>Done building project “Cudacka.vcxproj” – FAILED.
I gained 25ms on plain 1070, overclocked it goes down to 400ms.
Need new power supply for the donated 1080 Ti - got some reboots, but gained 30ms with optimizations and 20ms with overclocking.
John reported 270ms base so that could push it down to 230ms overclocked. This is not likely to change for a while until connection to grin node and OpenCL is finished.
Edit: Ok my 1080 Ti seems to be running memory at 10 Gbps instead of 11 Gbps, that explains why I can’t match John’s 1080 Ti.
Running ./Theta -n 0 -r 100
I see times of 263 ms and up, averaging around 266 ms.
Btw, this is not my 1080 Ti, but Yeastplume’s machine, the “Yeastmonster”.
Is this with the new experimental cu file in the new folder or the usual one? I made a new file for 1080 Ti. All other things being equal I gain flat 15-20ms with the new file.
Oh, I didn’t realize I had to move the kernel.cu from the GPUs_8GB_+ subdir.
After doing that, I see the improvements: down to 240 ms, averaging around 243 ms.
When I do longer runs like -r 1000, they invariably abort with
CUDA trimmer not responding
at iteration 65. If you can fix that, I’ll do some more testing of cycle coverage, and then we can move on to bounty payment…
Damn, works on windows few thousands in and going, so it is something in the C# control code at around when 42 cycle is found, I changed few numbers and pushing to repo. If it won’t work I’ll have to move the 1080 to my linux machine later.
Pretty sure it is fixed now, the C# program was not actually waiting for solution to be recovered by GPU and then it looked like trimmer is frozen while it was actually still recovering solution.
