Shopping Cart
Priority Question
February 24, 2011 - 7:04pm
- 1189 reads
Hello,
I'm trying to convert height map to cone map. And it's a bit big resolution file. So when I do that, my pc freezing, even I can't move my mouse , I don't see even any progress update in screen... so you know these already. The question is, there is a way to reduce xnormal's speed?? I tried put low priority in task manager but it looks like it uses calculation over GPU and I don't have control over it...
Maybe you know some tricks, like disabling or deleting CUDA plugins? or some command line parameters? or some tricks on windows that disables CUDA?? I need my computer alive you know :/
If you are using the Cuda Optix renderer then these isnt much you can do about the freezing.
If you are using the CPU rendering and you have a multi core CPU, then you change the CPU affinity in Windows to restrict xNormal to a defined number of cores.
To do this you need to open Task Manager> Right click the xNormal process> Set Affinity and then check/uncheck the cores of your choosing.
Hope that helps!
good suggestion thanks.
However even it's able to use 4 of my cores, it's using only one core and using %25 of my cpu... I get freeze even with that...
The problem appears that because of the image resolution is high. If I drop the image size by 8/1 I can see progress bar even I can move my mouse with 4 seconds delay shuttering...
And is it possible to split my high res image to lower ones and put them together after converting?
Btw, I have 8gb ram and 1gb video card ram.
How can I set xnormal to use CPU or GPU? Do I have to install CUDA toolkit? Because in plugings -> graphics driver , there are 4 options avalible. dx9, raytracing cpu, openGL, dx10.
You can change the renderer by going to the baking options and then changing the renderer to Default Bucket renderer (CPU).
Also, what size image are you using?
I get a freeze and mouse lag only when using the GPU renderer which is due to the GPU handing over resources to xNormal rather then using its power purely for the display. With CPU you should be fine 
4096x4096
yep that was default bucket. Changed to cuda and I get a bit more performance on converting I think. And lesser delay on mouse lag.
And I will try the big resolution one now, I will push the reset button I think or will I able to press cancel button...
ahh.
The Cuda renderer is extremely fast, so it may just be worth waiting for it to render as it should take seconds to finish.
You will be able to cancel but it may take a few seconds to respond
yeah abort button worked after 5mins... And I don't think it will be that fast 
Anyway is that possible to split image thing because I'm sure if I split it, it will be more faster...
I guess spliting will not work, because height source option depends on maxRGB so in every splited image, maxRGB is different I believe... or it uses the 255,255,255 globally for maxRGB...
Btw, that CUDA option is not working with height map to cone map... What a trouble. It's using %25 of my cpu and freezing like -55c.
What GPU do you have?
I average around 30-60 seconds or so for a 4k map with 600 rays with my 460gtx.
Unfortunately, I dont think its possible to automatically split the render 
If you had identical high and low points on each mesh, then in theory you could do it manually though.
Btw, that CUDA option is not working with height map to cone map... What a trouble. It's using %25 of my cpu and freezing like -55c.
The Cuda renderer only works with AO and Normal iirc.
I have 9600 GT. Cuda cores 64.
Mate, can you try converting a height map to cone map in tools panel. Because I'm not baking a mesh. Try under 1024 px res grayed image.
edit : tried spliting and merging them. It's failed, there are seems generated over the edges of the splits. and not matching over all of the image.
Ahh sorry, i didnt realise you were converting a map to a cone map.
I have the same issue with it being very slow (max 22% cpu usage), though im not getting the lag that you talked about (though xNormal stops responding a few times)
I think its prob. down to the age of the plugin. Maybe Jogshy can update or improve it a little so that it uses all the CPU?
yeah.. I guess, there's a limitation over buffers which is being a problem when you try to convert high res images... If that can't be solved, then there should be something storing image's max/min rgb values or what max/min values depended and split and generate... then merge.
Edit: I just left my pc for 8 hours while I sleep. %0 progress... Don't know if progress bar was not updating.
In macbook pro, there's no lag like the freeze on my pc. I can use mouse to select icons etc... I guess difference is windows settings, driver settings etc...
Yea, hopefully Jogshy will get some free time to update the plugin
The Relaxed Cone Mapping is very computationally-expensive. Really, very.
I had to use the GPU because using the CPU could take ** days ** of computation. A 1k x 1k requires near 1 Petaflop. A 9600GT gonna have serious problems computing that because the g80 architecture is a bit old. A GT200 will be better and Fermi will take much less time.
The problem is that I'm using OpenGL to compute it because OpenCL was not available when I wrote the tool. That means xN will eat all the GPU resources without having the chance to control the priority. Adding an extra GPU won't solve it neither because the tool is not prepared to use SLI.
I plan to solve it in xn4 using OpenCL and progressive rendering but I'm afraid for xn3 the only solution is to use a Fermi card and go make something until the render finishes. A workaround is to use simple Cone Mapping instead of the Relaxed/Quad one.
CUA/Optix can be only used to render AO/NM maps, it's not used for the tools. And nope, you don't need to install de CUDA toolkit. The xNormal's installer installs all that is required.
Good information!
Thanks Jogshy
I just installed rivaTurner to see if I can get benefits from it's openGL settings.
Only settings that can help xnormal's computing is, Amount of system memory for PCI textures = "5mb"
And I increased it to 500mb, And woot, actually I got response from my pc, I can press abort button when I start generating high resolution image... Thanks for good info Jogshy, too bad at these days, we can't find something that actually beats the new computer's calculation power, but that's the one of them... I did my calculation, the height map will take minimum 4-5 days to generate. even with full speed GPU fan working, it will probably damage my card
ah.. well I just read "OS Frozen while running xNormal" thread... and I feel hopeless now... I will just experiment with plugin settings then.. with lower resolution image. and see if I can make it work with reasonable computing speed... Have no idea how x slower may be.