Hello. I have two GTX 1070, and I would like to know if its worth upgrading to one 2080. Unfortunatelly there isn’t enough information about performance on Cycles with RTX cards, but it appears that one single 2080 is more powerful than two 1070 on CUDA, and even more powerful using OptiX.
Someone with a RTX card can share some thoughts?
Thanks!
Hi, check the tread about Optix support in Cycles.
Optix is close to two times faster than Cuda but there are some features missing at this point.
Cheers, mib
You can go for a 2070 super. It’s almost on par with the 2080 and costs almost 20% less. Much better vfm.
Interesting. Unfortunatelly, it seems really hard to get one here in Brazil.
Nice. I will search there.
I tried Optix when 2.81 was still in beta (2080 Ti). I remember that cycles rendering was slightly faster than CUDA+CPU (Ryzen 7 2700X) combined. Personally I was not that impressed.
What blew my mind though was how quickly Ambient Occlusion baking became with RTX in Substance Designer. What took easily 15-20 mins before now happens in just a few seconds.
I’m afraid of something like this. Investing a huge amount of money to gain 10, 20% more… don’t think its worth it if this would be the case. However, I think that Optix seems like a great engine with more to be developed in the future.
For a comprehensive comparison click here.
Hey, thanks for the link.
I also want to renew my hardware.
I thought of a rtx 2060 graphics card.
There are different versions.
Are 8Gb Ram necessary?
Do I need more than 6 GB of RAM for files like the barbershop?
benchmarks/cycles/barbershop
Best regards
Wuerzie
Barbershop will fit into 6GB of GPURam but who knows what the future will bring.
Hi Markus,
thanks for your answer. Most Times I render anmations.
I think a blender file witch use more than 6 GB ram, renders to long for an animation.
I just upgraded to a 2070 from having a Radeon R9 280x, and my render times were greatly decreased. I set OptiX on recently as well. Can’t tell you much about increase there…from CUDA to OptiX, but a 2080 is a beast. If you have the money, it’s not going to be a bad investment.
We recently put together a dual 2080 rig for our ‘render machine’ and it’s a monster! All of our workstations have 2080s in them…I assume the tech dude did his research. I think you could get away with a 2070…look into the ‘super’ versions of the RTX cards as well.
On a single home system that holds very true.
Anyways, if you’re trying to achive a certain realism that gets into such numbers of memory consumption, it’s time to think about growing into a render farm.
Hi I have buy new hardware.
The Barbershop rendered
with CPU in 15 min. Samplings 32 + Denoising, Tiles 16x16
with GPU CUDA in 10 min. Samplings 32 + Denoising, Tiles 128 x128
with GPU experimental in 5 min. (is this OptiX ?) Samplings 128 + Denoising, Tiles 128 x128
with GPU experimental in 7:28 min. (is this OptiX ?) Samplings 200 + Denoising, Tiles 128 x128
The result is very bad when I use If I use samplings 32 for experimental too.
I need many more samplings when I use GPU experimental.
It takes a lot of time to transfer the data to the GPU.
Cuda is faster for small scenes.
Cuda: The RTX use only 7% of its performants.
GPU: experimental use only 3% of its performants.
Is that normal?
Here is the result.
GPU experimental in 7:38 min. ( OptiX ?) Samplings 200 + Denoising, Tiles 128 x128
Which blender version?
Which file exactly?
Are you running Optix or CUDA?
Can you be a bit more clear?
Windows 10 (1903)
Blender 2.81
Asus Prime B450-Plus Mainboard
Corsair Vengeance LPX 16GB (2x8GB) DDR4 3200MHz
AMD Ryzen 7 2700X Prozessor 3.7GHz
MSI 8GB D6 RTX 2060 Super Ventus
OptiX
Cycles GPU experimental
Samplings Render 200
Denoising
Rendertime 7:38
Tiles 128 x128
Sounds about right, on my RTX 2070 I’m getting 07:20.57, the only thing I changed was the tile size (512X512). I think with Optics you need to have large tile sizes again?
