I have an opportunity to buy a 2080 Ti at a reasonable price and I’m thinking whether I should make the jump or not.
I’m currently running 2x1080 Ti and overall happy with performance but if I would upgrade what would be the best choice?
One 2080 Ti or two 2070 Super with Link? I’m planning to use E-Cycles so I’m expecting there’ll be a significant jump in performance from my current GPUs.
Can I also use hybrid setup? Like 2080 Ti and 1080 Ti or 2080 Ti and 2070 Super together? Can I use 2080 Ti from different manufacturers if I buy first one from one manufacturer and later the second one from another? Won’t there be any conflicts?
I can tell you that when you want to get a new GPU… powerful like 20 series. You’ll also need to have a powerful CPU that can keep up with powerful GPU.
balance is the key for computers… it take a skill to be able to do that. balancing betwteen GPU and CPU will make a good gaming computer or high performance. But if you have a CPU that can keep up with 1080Ti, then I can tell you that it cannot keep up with 20 series RTX… at best, it can use 50% or less of the new GPU if you are lucky… That’s why some people would overclocked their CPU to do this for the balance.
of course, the CPU and 1080 will bring each other out at 99% more or less.
I would recommend that you use 2 of exactly same series and brand from manufacturer. I looked in this long time ago. If you use, let’s say 2080 and 2070. it’s not possible because you know the difference 2070 and 2080. Of course, 2080 is slightly more powerful than 2070. well I cannot find a good example to say it in metaphor way
I hope this help you understand more about computer better than before.
I don’t know what you’re smoking because this is blatantly wrong. Cycles uploads the entire render kernel onto the GPU and renders from there, there’s next to no interaction between the CPU and GPU after Cycles starts rendering. This is why people can make PCs with a lightweight CPU and 8 or more GPUs using PCIE splitters and the whole thing works fine.
Again, completely wrong. You can use any combination of GPU’s for rendering as long as they all use the same API (OpenCL/CUDA/Optix). That is, all AMD using OpenCL, all Nvidia using Cuda, or all Nivida 2000 series using Optix. You can also use a non-compatible GPU to drive the screen while the others do the rendering (Ie. using an AMD gpu to drive the monitor while using 2000 series Nvidia RTX to render with Optix).
Anyways, to answer the OP’s questions…
1080 Ti are still really good GPUs. 11GB of memory is nothing to sneeze at. One downside of replacing or combining them with 2070s is that the 2070 has only 8GB of RAM, meaning you’ll be limited sooner on scene size/complexity. I don’t think Blender has support for NV-Link so you’re going to be limited to 8GB no matter how many 2070s you have.
An upside of using 2000 series GPUs is you’ll be able to take advantage of the experimental Optix API introduced with 2.81 or 2.82 which has some serious performance gains over CUDA, with some caveats. For example, you can’t use simultaneous GPU/CPU rendering, and you can’t use out-of-system memory for scenes bigger than your VRAM.
As I mentioned above, you can run almost any combination of GPU. Every Nvidia desktop GPU and most laptop GPUs since the 600 series meet the CUDA requirement, while 2000 series RTX cards can run both Optix and Cuda. As long as your case has the space and your motherboard/CPU has enough PCI lanes you can throw as many GPUs as you want at Blender.
If I was in your shoes, I’d get the 2080 Ti and try and squish it in with both 1080 Tis to keep that 11GB memory advantage, and then maybe keep an eye out for other good deals on 2080 Tis to replace the 1080s so I could go for Optix.
E-Cycles supports rendering with your 1080Ti and the RTX 2070 super or 2080Ti together, all with OptiX If you manage to fit your 2x 2070 super + your 2x 1080Ti in one computer, it’s certainly a very powerful combination! Please send the bench results if you do it
I don’t know what you’re smoking because this is blatantly wrong. Cycles uploads the entire render kernel onto the GPU and renders from there, there’s next to no interaction between the CPU and GPU after Cycles starts rendering. This is why people can make PCs with a lightweight CPU and 8 or more GPUs using PCIE splitters and the whole thing works fine.
I agree. This statement confused me as well. Anyway I’m confident my rig can work with 2 two Ti easily if necessary.
Again, completely wrong. You can use any combination of GPU’s for rendering as long as they all use the same API (OpenCL/CUDA/Optix). That is, all AMD using OpenCL, all Nvidia using Cuda, or all Nivida 2000 series using Optix. You can also use a non-compatible GPU to drive the screen while the others do the rendering (Ie. using an AMD gpu to drive the monitor while using 2000 series Nvidia RTX to render with Optix).
Yes, it’s what I was thinking. Leave one 1080 Ti for the screen while rendering on 2080 Ti.
1080 Ti are still really good GPUs. 11GB of memory is nothing to sneeze at. One downside of replacing or combining them with 2070s is that the 2070 has only 8GB of RAM, meaning you’ll be limited sooner on scene size/complexity. I don’t think Blender has support for NV-Link so you’re going to be limited to 8GB no matter how many 2070s you have.
Thank you for clarifying it for me. I wasn’t aware you need software support to benefit from NV-Link bridge. It changes things.
An upside of using 2000 series GPUs is you’ll be able to take advantage of the experimental Optix API introduced with 2.81 or 2.82 which has some serious performance gains over CUDA, with some caveats. For example, you can’t use simultaneous GPU/CPU rendering, and you can’t use out-of-system memory for scenes bigger than your VRAM.
It’s fine. I don’t plan to render on CPU anyway. Even though I have a good CPU (i7-8086K) I prefer GPU rendering and don’t use hybrid rendering.
I’d like to take advantage of E-Cycles. I like Cycles render. It gives me almost all I need. But if I can get decent boost with RTX on E-Cycles than the upgrade to 2080 Ti makes more sense to me.
As I mentioned above, you can run almost any combination of GPU. Every Nvidia desktop GPU and most laptop GPUs since the 600 series meet the CUDA requirement, while 2000 series RTX cards can run both Optix and Cuda. As long as your case has the space and your motherboard/CPU has enough PCI lanes you can throw as many GPUs as you want at Blender.
Unfortunately leaving 2 addition 1080 Ti isn’t an option. I have on purpose smaller case so 2 cards max in PCI x8 mode I think. It might be 16x but not sure right now. Anyway the performance difference is neglectable here.
Currently the cost worries me. If I can get new Asus ROG 11GB OC 2080 Ti for 30% less of market price. If it’s a good deal compared to 2070 Super?
Absolutely two 2070 supers. Much better price/performance ratio. Even without the NVLink module.
2070 Super is almost a regular 2080.
Yes you can use a hybrid setup, however you only get to use the lowest amount of GPU memory of all cards at all the cards as far as I know. So lets say you didn’t get the NVLink, 8 gigs of memory in 2070 Super will make it so 2080ti and 1080ti will only use 8 gigs of memory even though they have 11. If you get the NVLink for the 2070 Supers, they would add up to a single dual GPU card with 16 gigs of memory, so now you would be capped at 11 gigs in each card.
You can use cards from different manufacturers as long as they are all NVidia cards. It should be totally fine, on paper.
How many 16x slots do you have that you can install a 2 slot card into? There are some triple slot RTX cards too, gotta be careful about that. Also you gotta make sure you can cool all of them, the moment you hit 83 (afaik) degrees on a card, it will start throttling and become substantially slower.
There is some but little merit to the powerful CPU argument. Blender needs to translate the scene to Cycles and probably compile some stuff and copy the data to the GPU memory before anything happens. That is all CPU, and usually single threaded stuff. On Intel side anything over a 9700k, or on AMD side anything over a 3700x will do the trick nicely.
edit: just remembered that Optix support doesn’t yet support shader raytracing. Any ambient occlusion type nodes you have in your shaders will cause the Optix framework to fail. In those cases you need to revert back to CUDA. CPU/GPU combination was already mentioned I see.
Great to hear. But I couldn’t. 2 cards max so I need to decide where I should make the upgrade or not. But latest viewport denoising tech is really impressive in E-Cycles so it makes me think . If not for the latest advancement in the rendering and support for RTX I wouldn’t bother.
dang it. I forgot this is about 3D modeling… not gaming. well even it’s for 3D modeling. I am trying to make it very simple and basic that one person can have a good understanding of computers and their relationship with 3D model and games.
It should be possible to use CPU and GPU at same time using Blender 2.79 or latest version if i remember correctly. having CPU and GPU can cut the workload in half for them to do. But When it come to gaming, a old CPU cannot use the full potential of a newer GPU like 2080. That’s the basic I am trying to say.
now It look like I really forgot it’s about 3D modeling. so I’ll leave the decision to AlexeyAdamitsky
so good luck
Absolutely two 2070 supers. Much better price/performance ratio. Even without the NVLink module.
2070 Super is almost a regular 2080.
What if I can get top 2080 Ti brands 30% cheaper from market price? Does it make better sense to go for 2080 Ti?
How many 16x slots do you have that you can install a 2 slot card into? There are some triple slot RTX cards too, gotta be careful about that. Also you gotta make sure you can cool all of them, the moment you hit 83 (afaik) degrees on a card, it will start throttling and become substantially slower.
It’s tight but I have good air flow planned. 2 3 slot cards fit fine.
Yes you can use a hybrid setup, however you only get to use the lowest amount of GPU memory of all cards at all the cards as far as I know.
Yes, I’m aware of this limitation. But @TheTuxedo mention above that Blander doesn’t recognize NV-Link so you’ll still have 8GB even with dual 2070 Super and NV-Link bridge. Which changes things.
Optix support doesn’t yet support shader raytracing. Any ambient occlusion type nodes you have in your shaders will cause the Optix framework to fail. In those cases you need to revert back to CUDA. CPU/GPU combination was already mentioned I see.
Thanks for warning me. I wasn’t aware of this limitation.
Well, a 2080ti is twice the price of a 2070 Super. But the 2080ti should be roughly 40 percent faster than a 2070 Super in simple rendering tasks. It is your call really, but if I could get a good price on them I would definitely get the 2080tis
From what I’ve read. cycles uses 32 bit floating point calculations. Theoretical FP32 performance of 2080ti is 13448 Gflops, while 2070 Super is 9062 Gflops. There is a 48 percent increase from 2070 Super to 2080ti. Of course operative word being ‘theoretical’.
Reference-
Didn’t know about the NVLink support. Dang, that sux. Hopefully they will add it later on. I was relying on that for budgeting my next office workstation…
But the lack of link support in Blender brings something important to the table. How big are your textures? Usually it is not the geometry that fills the GPU memory but the textures.
If you think you would hit 7 gigs or more in GPU memory you should absolutely get the 2080tis.
From what I’ve read. cycles uses 32 bit floating point calculations. Theoretical FP32 performance of 2080ti is 13448 Gflops, while 2070 Super is 9062 Gflops. There is a 48 percent increase from 2070 Super to 2080ti. Of course operative word being ‘theoretical’.
If you think you would hit 7 gigs or more in GPU memory you should absolutely get the 2080tis.
I rarely go over 4K for textures. Trying to use procedural materials as much as possible but I would like to have 11 GB as a safety net.
I’m leaning more towards 2080 Ti. I don’t need 2 now so maybe later when Blander supports NV-Link I could add another one or just jump on the new generation with bigger VRAM per card.
balancing betwteen GPU and CPU will make a good gaming computer or high performance. But if you have a CPU that can keep up with 1080Ti, then I can tell you that it cannot keep up with 20 series RTX… at best, it can use 50% or less of the new GPU if you are lucky… That’s why some people would overclocked their CPU to do this for the balance.
If you manage to fit your 2x 2070 super + your 2x 1080Ti in one computer, it’s certainly a very powerful combination! Please send the bench results if you do it 
. If not for the latest advancement in the rendering and support for RTX I wouldn’t bother.
dang it. I forgot this is about 3D modeling… not gaming. well even it’s for 3D modeling. I am trying to make it very simple and basic that one person can have a good understanding of computers and their relationship with 3D model and games.
