I recently switched over from an iMac to a new PC. Everything has drastically improved, except going into the ‘rendered’ viewport. In particular, it’s when I’m dealing with highly subdivided or displaced meshes (3’x3’ cube subdivided 100 times for a realistic displacement). Blender completely freezes, becomes “unresponsive” for about 30-45 seconds and then continues without a hitch.
Old iMac specs:
El Capitan OS
Quad core i7 3.4 GHz processor
GTX 675MX
24GB ram
1TB Fusion drive
New PC specs:
Windows 10 OS
Xeon E5-2630 v3 8-core 2.4 GHz
ASUS x99-DELUXE Motherboard
Nvidia GTX 960 (display), 2x 980 ti (dedicated to Blender) GPUs
16GB ram
256 SSD
It boggles my mind how my old iMac can provide BETTER viewport performance than the PC. Unless it’s a Windows or hardware issue, I feel like I’ve tried everything I can think of in terms of Blender settings. I’ve scoured the internet but nothing seems to even slightly improve this problem. Hoping someone can provide some insight because I’ve exhausted searching the web.
I’m a PC noob, so perhaps it’s a BIOS setting or something else I’m missing entirely…any help would be much appreciated. I’m using Blender 2.75.
I recently made a similiar switch from my old iMac (2009) to a new screaming fast PC…
so, while I’m by no means an ‘expert,’ hopefully I can offer a few suggestions.
It looks like you have only about 75% of the RAM that you had on your old iMac. And while a PC with the x99 chipset should be running the new DDR4 RAM as opposed to your old I’m which is most likely running the DDR3, having that extra 8 GBs of RAM probably goes a long way.
That may only be part of the issue. I looked up your processor an that looks pretty solid as well, though it has fairly low clock speeds compared to whats out there.
Considering how much calculation blender needs to do for high poly objects like the ones you’re describing (which are done on CPU), that may also be part of the issue.
Rarely is dragging performance related to one thing in particular. Typically, its about optomising your system in away that works best for your needs.
Remember, even major studios have to wait forever for their computers to calculate final scenes and renders. According to the making of ‘Gravity’, if they had tried to process and render all of the effects in that movie on one (super) computers, they would have had to start rendering it over 5000 years ago to be finished today
In the meantime, the old ways of improving viewport performance are; working with a few windows open as possible, making good use of wireframe and bounding box views (when possible of course), and you can always use proxy objects (low poly) as placeholders in your scenes until you are ready for the big time-consuming objects.
Thanks for the reply. Initially I had though RAM could be an issue, but when using resource monitor my current memory is never used beyond 50-something percent. Is it possible for RAM to reach its limit without showing so on the resource monitor? I also wonder if it’s a Windows 10 issue since I’ve heard it struggles a bit more with performance compared to other OS. I’ve also started running into the “exceeded CUDA time limit” issue…but at least that seems easily fixable.
Thanks for the tips, I’ll probably have to work on re-adjusting my workflow and rely less on the rendered view. I don’t mind long renders, I’m just trying to avoid having Blender completely freeze up whenever I have a complex scene/object…but in terms of the rendering speed itself I haven’t had issues. Maybe I overestimated the power of my new PC with my project. Still hoping that I’m just overlooking something.
Hello,
I have been nearly having the same problem, arising just today. Whenever I render, I receive a message: CUDA error at cuCtxCreate: Launch exceeded timeout. I originally thought that it was a GPU problem, because I could render just fine with CPU. However, when I go into a different blender file, with nearly triple the amount of vertices, I can render just fine with the GPU. Now, I’m not so sure that this is a hardware problem. I’m not sure if this is similar to the problem you are having, but that’s just my little bit. I might make a separate thread for it.
(3’x3’ cube subdivided 100 times for a realistic displacement)
Your GPU might not be able to support that level in rendered view port. It maybe takes a lot of time to compute everything, that is why the hitch.
Also check the number of samples in render settings, the more the samples the more time it takes for rendering mode to be visible in view port.
Try reducing the subdivision levels to check whether the hitch is still there or not.
Edit: Need to mention: Your 980 ti is being used for final rendering. You viewport render is all running from your display graphic card that is GTX 960, try changing your display device to 980 ti and then check whether you have the same problem?
Initially I increased the timeout from 2 seconds to 60, and while this helped, I found that raising it to 120 seconds did the trick. Haven’t had the issue since. It’s fairly easy to follow so hopefully it helps you.
I will definitely give that a try. I actually have 3 GPUs: 2x 908 ti and gtx 960. Initially it was suggested to designate the 960 for the display, and have Blender only use the 980 ti’s. Do you recommend plugging the display into one of the 980 ti’s and let Blender utilize all 3?
I recently did some research into displacement, and I think I’m going overboard on the subdivisions. Going to try using higher resolution textures with lower subdivides and see how that looks.
Man I am using a subsurf with 6 and another subsurf with 2 on an asteroid and rendering with cpu is taking 1 hr per frame when it is very near to camera.
Subdividing 100 times, you are lucky your GPU is not crashing. Mine Nvidia driver is crashing even though I am using cpu for rendering because of these subdivision levels.
Yes try using 980ti as display device, it has more RAM and shaders.
I have ordered an R9 390 instead of a 980 ti. It is half the price of 980 ti and has 8 GB vram.
Try doing the same, sell your 960 and get an AMD one. First you will get 8 GB vram just for display and 4 monitor support and excellent view port rendering.
Second is when rendering with blender you can use two 980 ti and that will increase the rendering speed because Blender can use either one gpu of the same brand are all three. So if you use three Nvidia, it will use any one or all three.
Being an AMD for display, your display memory will not get clubbed with the 980 ti, and you can continue working with another blender file while the 980 ti keeps rendering your animation in background.
and because you are not rendering with AMD, so you have no problems with the features it doesn’t support.
You must have already done some cooling requirements for running three cards in one cpu.
I’ve been doing some playing around, and I definitely was using too many subdivides. Of course I love how exact and photo realistic I can get the geometry displacement with high subdivides, but I realized I just needed to put more work into my normal maps…this saves my viewport a lot of lag. I’d be curious to know just how much of CGI is actual displacement vs really good normal maps.
Thanks for all your suggestions, I may go that route. I also overclocked my GPUs which seemed to help boost performance a tad. I splurged for the CPU cooling and large high quality corsair fans (it’s a Slade from Digital Storms), so I’ve been able to keep the temperatures pretty low…though I haven’t really pushed it too hard yet.
