About graphic cards

[eugeniodeckard]

hello.
I have a titan x pascal, want to upgrade and buy other or other graphics cards for reduce render times.
Do you have experience mixing different graphics cards architectures?, any problem?
i think about two options:

+ 1 x 2080 ti ( + my titan x pascal)
or
+ 2 1080 ti ( + my titan x pascal)

what is your recommendation?

[dougWilkinson]

i've only mixed the 9XX series with the 10XX series and its fine, no problems.
at work i have a box with two RTX 2080 cards and ive used the nightly builds with RTX support. i'm not sure if a non-rtx card will fall back to using straight CUDA or just be ignored. but in plain vanilla CUDA mode i think you can easily mix many architectures.

if i was in your situation i'd prob look for one or two cheap 1080 ti cards to tide you over untill next gen nvidia cards... my RTX cards at work have been nice but still have issues here and there.

good luck

[Micha]

I use 2x2080ti and 1x1080ti at Rhino and the RTX cards works stable. I wouldn't buy 1080ti anymore. One 2080ti is approx. 1,5 times faster than a 1080ti. The RTX support could make the difference much higher. So, finally a 2080ti could be faster than two 1080ti. I would be careful. Also if you later would upgrade to a second 2080ti than you could get approx. 22GB VRAM per NVlink. This isn't possible per 1080ti.

Also keep in mind how much cards you want to add to your machine in the next time. Since the card count is quite limited I decided me to add only most powerful cards to use the exist space.

[francomanko]

so I know this is a slightly spurious question, but ignoring the ram issue and render differences, are the gpus out performing cpu $ for $?

[dougWilkinson]

I use 2x2080ti and 1x1080ti at Rhino and the RTX cards works stable. I wouldn't buy 1080ti anymore. One 2080ti is approx. 1,5 times faster than a 1080ti. The RTX support could make the difference much higher. So, finally a 2080ti could be faster than two 1080ti. I would be careful. Also if you later would upgrade to a second 2080ti than you could get approx. 22GB VRAM per NVlink. This isn't possible per 1080ti.

Also keep in mind how much cards you want to add to your machine in the next time. Since the card count is quite limited I decided me to add only most powerful cards to use the exist space.

this prob comes down to a difference of opinion, but the 2080 ti are hugely overpriced for what you get in my opinion. RTX is only just hitting the stable nighties in the last week. On top of this i have the NV Link working on my 2x 2080 and have had mixed results with it. We expect to see next gen RTX cards early next year and rumors say that they will be a big leap over the current gen in both VRAM amount and RT cores. 2080ti just aren't a good investment right now in my opinion. Having said that they are VERY NICE if you have limited space like you mentioned, and money isn't the top concern

[Micha]

so I know this is a slightly spurious question, but ignoring the ram issue and render differences, are the gpus out performing cpu $ for $?

It's an interesting question. It's a pity that the V-Ray benchmark doesn't allow to render the same scene for GPU and CPU. The kernel of the question is for me - we need to know the speed of a GPU compared to a CPU for the same quality for the same scene. For example I have the impression a 2950x is half so fast like a 2080ti. If this raw impression would be true than we could get an answer, where to find more power for money.