Gpu vs cpu

[glorybound]

What should I expect to be faster for stills? 36-Core Dual E5-2696 v3 Max 3.8GHz 128GB RAM or EVGA GeForce GTX 1080Ti 11GB

I ask because I ran RT, using the GPU and it was slower than rendering using progressive in the CPU. I also ran RT in hybrid mode and that was also slower, which didn't make much sense.

[savage309]

Are you sure you are rendering for the same noise levels?

Best,
Blago.

[glorybound]

Are you sure you are rendering for the same noise levels?

Best,
Blago.

I am just using the defaults, so I'll have to check. It sounds like RT should be faster and RT with GPU/CPU should be fastest, correct?

[francomanko]

Would definitely be interested in the results because that's a massive price difference between the two

[savage309]

The image sampler behaves differently in RT/Adv. The easiest way is to compare the renders after the same amount of time. Also, make sure to use V-Ray GPU as a production render, since it's ray bundle size / rays per pixel / etc settings there are optimized for final frame rendering.

Best,
Blago.

[FLP]

I believe (correct me if I'm wrong) that GPU's typical excel at calculating direct lighting. So your interior scenes which rely on indirect lighting (GI) would require more than a single GPU to match the performance of your CPU. Plus not having to think about memory constraints on your GPU vs CPU is something to consider.

[savage309]

I believe (correct me if I'm wrong) that GPU's typical excel at calculating direct lighting. So your interior scenes which rely on indirect lighting (GI) would require more than a single GPU to match the performance of your CPU. Plus not having to think about memory constraints on your GPU vs CPU is something to consider.

This is not quite the case. This is more what like the results should look like.

Best,
Blago.

[glorybound]

I believe (correct me if I'm wrong) that GPU's typical excel at calculating direct lighting. So your interior scenes which rely on indirect lighting (GI) would require more than a single GPU to match the performance of your CPU. Plus not having to think about memory constraints on your GPU vs CPU is something to consider.

Okay, good to know, Thank you!

[glorybound]

This is not quite the case. This is more what like the results should look like.

Best,
Blago.

I'll read earlier. Are there any setting I should tweak to improve GPU render times?

[savage309]

I'll read earlier. Are there any setting I should tweak to improve GPU render times?

Ideally it should work out of the box.

Best,
Blago.

[Deflaminis]

I'm curious how this ends up. I've got the some type of results but my machine isn't pimped out like Bobby's is. The CPU's win every time in my tests, but every article I read tells me this is impossible. I'm doing something wrong but I don't know what...

[glorybound]

Hybrid has to be faster if it is using both the CPUs and the GPU's. My current interior is using 19MB or RAM and I have a 11MB vcard, so I am assuming everything can't fit on the card. Unless it's not apple and apples when it comes to CPU and GPU RAM. Regardless, in hybrid mode shouldn't it fill the vcard and rest renders on the RAM?

[glorybound]

After reading the below article, there should be no reason why GPU+CPU wouldn't be faster than anything else. In my case, it isn't the case, so I am confused.

https://labs.chaosgroup.com/index.ph...rid-rendering/

[savage309]

After reading the below article, there should be no reason why GPU+CPU wouldn't be faster than anything else. In my case, it isn't the case, so I am confused.

https://labs.chaosgroup.com/index.ph...rid-rendering/

As I said, if you are not testing rendering for the same time + comparing noise after that (and the same settings, for example BF+BF), your test might not be accurate.

Best,
Blago.

[BBB3]

I've run into an issue before where RT was rendering very slowly (in production mode). Usually, this goes away when stopping and restarting the render.