Tweet
Hello,
I spent some time today benchmarking our new high end system. It is running dual Intel X5550 2.67 cpus and it has one GTX285 and twoTesla C1060 GPU cards and 32GB of ram.
Th system showed some very impressive times with the 3dspeedmachine mental ray scene rendering it in 5 mins 44 secs compared to 7 mins 46 secs on a dual X5482 system (http://www.3dspeedmachine.com/?page=3&scene=33).
I ran the test with V-Ray and the render time didn't seem as fast as it should. Upon further inspection I saw that V-Ray was not taking advantage of hyperthreading on the cpu. This is important because the calculation phase on this system is about 90% of the render time. When the rendering phase begins it really flies.
See the image below, which was taken during the calc phase. You can see the cpus at ~50% (this implies that 100% of the physical cores are being used but 0% of the virtual cores) . As soon as the calculation phase is over the the the cpus back off for about a second and then ramp up to 100% as seen in the image below.
Can someone confirm this for me? I think V-Ray should render this scene 50-60% faster than it is. I should point out that my dongle was at a different office when I wanted to run these tests so I had to use the demo version of V-Ray. I'm certian the demo version could be better optimized for HT. I would like to see if I would get the same result in SP3 (demo version is sp2) but maybe some here may know? I ran each scene 3 times.
thanks,
-joe
Here is the grab taken during the calculation phase:
Here is a grab taken during the render phase (note 100% usage of the virtual cores)
And here is a grab from the Mental ray scene (100% during calc phase and render phase):
I spent some time today benchmarking our new high end system. It is running dual Intel X5550 2.67 cpus and it has one GTX285 and twoTesla C1060 GPU cards and 32GB of ram.
Th system showed some very impressive times with the 3dspeedmachine mental ray scene rendering it in 5 mins 44 secs compared to 7 mins 46 secs on a dual X5482 system (http://www.3dspeedmachine.com/?page=3&scene=33).
I ran the test with V-Ray and the render time didn't seem as fast as it should. Upon further inspection I saw that V-Ray was not taking advantage of hyperthreading on the cpu. This is important because the calculation phase on this system is about 90% of the render time. When the rendering phase begins it really flies.
See the image below, which was taken during the calc phase. You can see the cpus at ~50% (this implies that 100% of the physical cores are being used but 0% of the virtual cores) . As soon as the calculation phase is over the the the cpus back off for about a second and then ramp up to 100% as seen in the image below.
Can someone confirm this for me? I think V-Ray should render this scene 50-60% faster than it is. I should point out that my dongle was at a different office when I wanted to run these tests so I had to use the demo version of V-Ray. I'm certian the demo version could be better optimized for HT. I would like to see if I would get the same result in SP3 (demo version is sp2) but maybe some here may know? I ran each scene 3 times.
thanks,
-joe
Here is the grab taken during the calculation phase:
Here is a grab taken during the render phase (note 100% usage of the virtual cores)
And here is a grab from the Mental ray scene (100% during calc phase and render phase):
Comment