Showing results for 
Search instead for 
Did you mean: 

Blender Discussions


macOS MBP dGPU 560 - Spike in VRAM Use = crashing. (RPR 2.0.2 - Blender 2.8)

My MBP had an uptime of 7 days yesterday (no other crashes other than RPR crashing Blender). I installed iStat Menu 6 (iStat Menus) so that I could monitor VRAM use on the 560.

Yesterday I noticed that my system was using 50% of the VRAM with my normal apps open. When opening Blender 2.8 and telling RPR to do a Preview Render the VRAM use would spike to 97% and Blender would immediately crash.

So I restarted my Mac and did the same tests. VRAM use without any apps running is around 23%.

Test one was with EEVEE and Glassy Cubes Blend attached. The VRAM max goes to 47%. See 01 EEVEE.png.

Then I switch the renderer to RPR and the VRAM spikes to the high 85% and crashes - see file 02 RPR Trying.

I then load yesterdays Hope Cube file (attached) and the RPR Preview Render spikes again to 97% and crashes.

I load the Hope Cube again and set RPR as renderer then this time press F12 set at 720P. It renders fully on the GPU with a considerably lower VRAM use.

I don't know if this kind of testing is useful but the spikes in VRAM use do coincide with RPR crashing on my Mac.

I would obviously expect VRAM use to increase during rendering. However memory management within the 560 dGPU when RPR is running doesn't seem to be working correctly. Renders begin but crash within a few seconds.

MBP 2018 i7 3.1GHz with 560 4GB dGPU.

Update: I've added a WX 9100 test at the end. It will render the F12 and will also render the Preview (but as soon as you pan around the cube it also crashes - obviously memory usage on a 16GB GPU is a lot lower).

I firmly believe that if AMD solve the issues with the 560 and test development builds with it the bugs it reveals will make all the other cards work reliably as well. It takes seconds to crash the 560 where it can take minutes to do the same crash on a WX 9100. 

Further very important update:

use_gl_interop = False

Was set for all of the above tests. I need it to be false to try using an eGPU.

I have since set it back to True and all of the above tests do now work again with the 560 dGPU.

1 Reply

This is fixed in the 2.0.10-hotfix build.

Please mark as closed.