Some useful links l found while exploring Deep/machine learning, AI etc:
I get the below error message snippet a lot while trying to compile Nvidia Cuda 10 samples for testing Cuda GPU development [ c++ code]:
The Windows SDK version 10.0.15063.0 was not found. Install the required version of Windows SDK or change the SDK version in the project property pages or by right-clicking the solution and selecting “Retarget solution”.
145>Done building project “topologyQuery_vs2017.vcxproj” — FAILED.
Solution l have found is to double check the Visual Studio 2017 installer and check installing below. It seems even though l have a full VS 2017 installation with all the default sections installed, a more recent SDK is installed on my workstation.
After doing above, l was able to successfully compile most of the Cuda Code samples as shown below, without the previous error. Yes!!
Invisible Box – Flip Fluids Add-on Fluid Simulation – Blender
Music: Ardi Rida – Hope DanceEDM
This is still baking, so far 60Gb bake files and climbing at 334 out of 501 frames. Domain resolution used – 500 – ie High Quality.
Bake files grew from 55Mb [frame 1] to currently 277Mb [frame 333] !
Will update on the progress and final render.
and finally… after almost 2 days baking non stop…..
Below shows me trying to render one single frame and yet getting out of CUDA memory error.
I’m using a GTX1080 with 8Gb VRAM, on a i7 Quad core PC with 32Gb RAM
Switching to a CPU render and the results are below.
Time taken 1 hour for frame 348
Until l can find a quicker way to render each frame, by cuda which is usually faster especially since l have access to about 3000+ cuda cores collectively on my core workstation, l will give this project a rest for now.
I tried command line blender rendering of frames 16, 348 with my 8gb gtx1080 / l get the cuda error as the gpu ram used reached about 7gb out of the 8gb,
Currently rendering the first 200 frames as shown below.
Discovered this free but very good Benchmark software – Catzilla – Click here for the download website.
Here are my various PC results.
A17 laptop – GTX860M [4 years old]
PC with 2xGTX560Ti [recent build]
PC with GTX1080 8G + GTX970 4G [ 5 year old PC , i7 2600K at 3.4Ghz 32Gb ram with new Nvidia cards]
Quick summary/my findings:
- less render time running Linux [LinumMint] than under Windows 10 [all things turned off], same hardware
- SSD’s help
Linux run: Workstation with GTX1080 8Gb, GTX970 4Gb, 32Gb RAM, Quad core Intel processor. No SSD
Same hardware as above but run under Windows 10
Ran on same hardware, only using the GTX1080 8Gb , under Linuxmint – impressive resuly showing in my view and from cuda/nvidia research l did that the 4Gb of the 2nd graphics card limits the performance.
Home PC #2 Results below, newer PC, no SSD, 2 x GTX560Ti
Below results from Alienware 17 [4 years old] with Samsung Evo 500Gb SSD, GTX860M
Updating to the latest cuda development toolkit on Linuxmint 18.1 [with a GTX 1080+970]
First, update to R390 Driver
sudo dpkg -i cuda-repo-ubuntu1704-9-1-local_9.1.85-1_amd64.deb
sudo apt-key add /var/cuda-repo-9-1-local/7fa2af80.pub
sudo apt-get update
sudo apt-get install cuda
GTX1080 at about 80 degrees C – and doing 474 Sol/s
GTX970 at aboyt 56 degrees C – and doing 275 Sol/s
OS – Linuxmint
Notes, screenshots and results using, installing, testing Nvidia Cuda on Linuxmint 18.1
“CUDA is a parallel computing platform and programming model invented by NVIDIA. It enables dramatic increases in computing performance by harnessing the power of the graphics processing unit (GPU).“
- Puget systems
- nVidia devtalk
- NVidia CUDA
- SimpleScreenRecorder – awesome! used to create the video below of nbody simulation
sudo add-apt-repository ppa:maarten-baert/simplescreenrecorder
sudo apt-get update
sudo apt-get install simplescreenrecorder
Tests done on my custom built Workstation PC as well as Alienware 17 laptop – all GPU specs below.
Quick comparisons [not the focus of this post] :):
- Puget systems laptop, with a GTX 980M 8Gb – 91 billion int./sec, 1819 Gflops
- My Alienware 17 with a GTX 860M 2Gb – 43.3 b.i.p.s, 866 Gflops
- My workstation with a GTX 970 4Gb -> 146 b.i.p.s, 2919 GFlops.
Screenshots in no particular order.
Below result from Puget Systems test on their laptop, for comparison
Below my results