The scene showed peak memory of 863.68 , I did a f 12 render (even restarted my computer, prior to GPU test)
Using 2.68 official. Using full GI default settings in cycles.
cpu time - 5:05 720 p resolution
gpu tiime -7:12 720p resolution
It’s just strange because I have never had a time (if I was able to render with my GPU) that the CPU was faster. You can see its a very close camera angle…I don’t know if that is making a difference.
I can’t post the blend file because of size, but here is the render. Any thoughts would be appreciated.
2.68 has really improved cpu rendering times. Dont know if thats teh problem orr not.
Did you do any tests in the past?
On the bmw benchmark scene my rendertime dropped about 2.5 minutes.
I haven’t done any tests on this exact scene, but numerous tests on similar scenes… close up of the slot machine. The tests I was doing usually was to determine if I could even render with GPU (it depends on how many objects are present at that particular camera angle) But if I was able to render with GPU, it was always faster than CPU. The tests I did previously were with 2.68RC1 and I sure did notice the speedup in CPU on those renders that required me to use it. But this still has me stumped.
I may go back and do another test on a different, but similar scene.
Hi, do you use the same tile size with CPU and GPU?
GPU perform a lot better with bigger tiles, at least 256x256.
I have a GTX 550Ti too and it is ~ double as fast than my i5 3570K on any scene.
The 2.68 is 30-50% faster on CPU on windows, 10 % on Linux.
I can´t imagine so much difference on different camera position.
If your file is smaller than 30 MB you can use http://www.pasteall.org/blend/
I can test it on my system.
Mib… I appreciate the offer, but this file is 165 mb unpacked right now… but you touched on something that I should get clarification on regarding the tiles sizes…
-When rendering on my GPU my default size tile is 256x256
-But often times, depending on the scene, I am over my memory limit with GPU and have to render w/ cpu… usually I just quickly change the setting to CPU and then hit F12
-But here is my question … If I have to go to CPU, and the size I am rendering is 720p… what should I set the tile size for when switching and using CPU?
OK now for an update…
I moved a poly heavy object that was on the layer I was rendering (but it was out of camera view) to a different layer.
when I went to rerender…all of sudden things got WAY better and more normal… the memory usage on scene dropped to 402 and my times dropped to the following:
CPU 3:15
GPU 1:15
I want to say this change made all the difference. But I can also say that wierd stuff like this has happened to me countless times… sometimes just shutting blender down and reopening it, or even restarting my computer, can drastically change my render times with either CPU or GPU… why this is, I have no idea.
Let me end this by asking one final question … if I have 3 objects in a scene on the same render layer… and based on the camera angle I set up… only object 1 and 2 are in the camera view… does object 3 suck up resources and add to the GPU memory…even though it is not in the camera view scene? I tend to think it does, but I would just like to get confirmation.
After the restart, was Blender the only app running? I find web browsers chew up quite a bit of GPU memory, especially with multiple tabs. I imagine all GPU accelerated apps (eg Photoshop) would be the same.
I have had Blender slow to a crawl rendering on GPU a couple of times, haven’t worked out why, but restarting Blender brings back blisteringly fast speeds for me.
Yep, only blender was running… I’m always pretty carefull about checking for other apps running because I’ve experienced exactly what you are referring to when I accidentally have left PS on.
only object 1 and 2 are in the camera view… does object 3 suck up resources and add to the GPU memory…even though it is not in the camera view scene?
Yes, unfortunately all raytracers work like this, all objects in a scene are calculated.
3. If you can reproduce such situation with a file I call it a bug e.g. memory leek, post to bug tracker.
Such errors are hard to catch.
I decided to test this on my laptop (no GPU.) I opened up 2.67b and opened a scene. It rendered in 6:48. I closed the program and opened 2.68 and the same scene rendered in 7:43. I closed the app and opened 2.67b again and the same scene rendered in 6:40. I went back to 2.68 and it rendered in 7:41. What would cause the new version to run so much slower? Everything else was the same. I didn’t use the computer while it was rendering.
Mib… so do you recommend, if I have to use the CPU for rendering due too scene size, I should switch the tile setting to 64x64 and then use 256x256 for GPU rendereing?
I should switch the tile setting to 64x64 and then use 256x256 for GPU rendereing?
Yes, CPU work much better on smaller tiles.
You can try setup tiles for GPU match exact render dimension /4, 960x540 for HD, may you get a few % faster render.
