# Wish: light cache calculation based on GPU

**URL:** <https://forums.chaos.com/t/wish-light-cache-calculation-based-on-gpu/93220>\
**Category:** Requests\
**Created:** [March 18, 2019, 8:15am UTC](https://forums.chaos.com/t/wish-light-cache-calculation-based-on-gpu/93220 "2019-03-18T08:15:56Z")\
**Posts on this page:** 10\
**Page:** 1

<div class="post-metadata">

**Author:** ![Micha](https://dub1.discourse-cdn.com/chaosprd/user_avatar/forums.chaos.com/micha/32/59168_2.png) [@Micha](https://forums.chaos.com/u/Micha)\
**Post date:** [March 18, 2019, 8:15am UTC](https://forums.chaos.com/t/wish-light-cache-calculation-based-on-gpu/93220/1 "2019-03-18T08:15:56Z")

</div>

Hi,

could it be possible to use the GPU for the light cache calculation too? The CPU based LC calculations looks so slow and I suppose so in the future users will invest in GPU power more than in CPU power. What are the plans?

-Micha

---

<div class="post-metadata">

**Author:** ![Muhammed\_Hamed](https://dub1.discourse-cdn.com/chaosprd/user_avatar/forums.chaos.com/muhammed_hamed/32/73779_2.png) [@Muhammed\_Hamed](https://forums.chaos.com/u/Muhammed_Hamed)\
**Post date:** [March 18, 2019, 11:00am UTC](https://forums.chaos.com/t/wish-light-cache-calculation-based-on-gpu/93220/2 "2019-03-18T11:00:49Z")

</div>

What version of Vray are you using?  
in Next, LC renders very fast compared to 3.6.. takes few seconds here for any scene  
It would be nice if it is calculated on GPU, but I don’t mind … It is already very fast on CPU

---

<div class="post-metadata">

**Author:** ![Muhammed\_Hamed](https://dub1.discourse-cdn.com/chaosprd/user_avatar/forums.chaos.com/muhammed_hamed/32/73779_2.png) [@Muhammed\_Hamed](https://forums.chaos.com/u/Muhammed_Hamed)\
**Post date:** [March 18, 2019, 6:47pm UTC](https://forums.chaos.com/t/wish-light-cache-calculation-based-on-gpu/93220/3 "2019-03-18T18:47:15Z")

</div>

What resolution are you rendering at?  
40 seconds is quite fast btw for interior scenes and such, but if your resolution is low or it is a basic scene you might want to upload it to support, so they have a look  
I do think that you need at least 2000 subdivs for light cache in Next, it will make rendering much faster for GI intensive scenes.  
Did you try a different scene?

---

<div class="post-metadata">

**Author:** ![Muhammed\_Hamed](https://dub1.discourse-cdn.com/chaosprd/user_avatar/forums.chaos.com/muhammed_hamed/32/73779_2.png) [@Muhammed\_Hamed](https://forums.chaos.com/u/Muhammed_Hamed)\
**Post date:** [March 18, 2019, 6:48pm UTC](https://forums.chaos.com/t/wish-light-cache-calculation-based-on-gpu/93220/4 "2019-03-18T18:48:32Z")

</div>

Don’t render on your 780, the 1080ti should be fast enough and you will have plenty of VRAM  
Usually you would want to avoid similar old cards, or use it for viewport to avoid issues

---

<div class="post-metadata">

**Author:** ![Micha](https://dub1.discourse-cdn.com/chaosprd/user_avatar/forums.chaos.com/micha/32/59168_2.png) [@Micha](https://forums.chaos.com/u/Micha)\
**Post date:** [March 19, 2019, 6:22am UTC](https://forums.chaos.com/t/wish-light-cache-calculation-based-on-gpu/93220/5 "2019-03-19T06:22:44Z")

</div>

At other scenes it works better. I wonder that the CPU LC looks so noisy and and the GPU BF is so clean and so fast. In the future I plan to upgrade to two GXT2080ti (per nvlink) to get max speed and memory. So, the scene could be rendered in 5 minutes and I suppose, if the LC pass could use the GPU power than I could save same time.

I’m curious what the Chaosgroup team thinks, is the LC pass on the to-to-list for GPU support?

---

<div class="post-metadata">

**Author:** ![Muhammed\_Hamed](https://dub1.discourse-cdn.com/chaosprd/user_avatar/forums.chaos.com/muhammed_hamed/32/73779_2.png) [@Muhammed\_Hamed](https://forums.chaos.com/u/Muhammed_Hamed)\
**Post date:** [March 19, 2019, 10:24am UTC](https://forums.chaos.com/t/wish-light-cache-calculation-based-on-gpu/93220/6 "2019-03-19T10:24:07Z")

</div>

> [@Micha](#):
>
> I wonder that the CPU LC looks so noisy and and the GPU BF is so clean and so fast.

LC is calculated beforehand, which is not the case with BF.  
BF is not faster to calculate, and it is not big deal if LC has noise as it is just an estimation for secondary bounces  
LC is way faster than BF for secondary GI and the bounces are set to 100 by default  
You would want to increase you BF bounces to match the same look, making your render even slower  
@vlado@a7az0th thoughts on LC on GPU?

---

<div class="post-metadata">

**Author:** ![a7az0th](https://dub1.discourse-cdn.com/chaosprd/user_avatar/forums.chaos.com/a7az0th/32/37410_2.png) [@a7az0th](https://forums.chaos.com/u/a7az0th)\
**Post date:** [March 19, 2019, 10:37pm UTC](https://forums.chaos.com/t/wish-light-cache-calculation-based-on-gpu/93220/7 "2019-03-19T22:37:55Z")

</div>

Hey,

we do plan to move LightCache calculation to GPU, however at this point I cannot give you an estimate when that feature might be available . We are currently doing RND on different approaches for offloading that task to the GPU , but is really too early to say more.

---

<div class="post-metadata">

**Author:** ![Muhammed\_Hamed](https://dub1.discourse-cdn.com/chaosprd/user_avatar/forums.chaos.com/muhammed_hamed/32/73779_2.png) [@Muhammed\_Hamed](https://forums.chaos.com/u/Muhammed_Hamed)\
**Post date:** [March 20, 2019, 12:06am UTC](https://forums.chaos.com/t/wish-light-cache-calculation-based-on-gpu/93220/8 "2019-03-20T00:06:54Z")

</div>

Awesome!  
Thanks so much Alex

---

<div class="post-metadata">

**Author:** ![AbeTJH](https://avatars.discourse-cdn.com/v4/letter/a/f14d63/32.png) [@AbeTJH](https://forums.chaos.com/u/AbeTJH)\
**Post date:** [March 22, 2019, 7:51pm UTC](https://forums.chaos.com/t/wish-light-cache-calculation-based-on-gpu/93220/9 "2019-03-22T19:51:31Z")

</div>

In the meantime would it be possible to enable CPU light cache calculation when using VRay GPU IPR?

Because the GPU IPR uses BF and BF, I’ve noticed that it occasionally gives very different results than a final render using BF and LC (at least with the default settings).

---

<div class="post-metadata">

**Author:** ![vlado](https://avatars.discourse-cdn.com/v4/letter/v/779978/32.png) [@vlado](https://forums.chaos.com/u/vlado)\
**Post date:** [March 26, 2019, 6:57pm UTC](https://forums.chaos.com/t/wish-light-cache-calculation-based-on-gpu/93220/10 "2019-03-26T18:57:06Z")

</div>

Yes, we have this working in some experimental builds, so hopefully it can be enabled for users as well after we fix the issues.

Best regards,  
Vlado
