Mixing Pascal and Maxwell Architecture.
KindredArts
Posts: 1,335
I'm really hoping some of you super savvy brain-boxes can lend me a hand on this one, because there seems to be a lot of conflicting information. I currently run a (Maxwell) GTX Titan x as my main render card, and with christmas around the corner, i wanted to add another one. The only problem is, they are like gold dust at the moment, and i'm very wary of buying second hand cards. Since i can't locate another Titan x, would the sky fall in if i ran a pascal 1080 alongside it? I know iray throttles to your slowest card, which is fine since they are both rather quick. But will the mixing of architecture produce odd results? I really don't want to ditch my titan, it's like a brother to me, help me forum yoda's!
Ps. I know pascal is still in beta, i'll wait for a stable release before picking anything up.

Comments
I have no idea, how good or bad, it will be, but 1080 has only 8 GB of VRAM, so your iray rendering, will use only that amount of VRAM available, if you decide to use both cards.
Are you saying that the 1080 will limit my titan x to 8gb for parity? I thought the limitation was cuda, so both cards would run at the same clock speeds. I dont really mind the cards reaching parity to be honest, its more the limitations of mixing architecture. This could either be a nightmare scenario for iray, or no problem at all, i honestly have no clue.
^to clarify - Iray tries to load the whole scene onto both cards - if it goes over your 8GB for the one card, that card won't be used at all. (And if it goes over the 12GB or whatever the TitanX has, you'll be in CPU mode regardless).
No idea if you can mix the two cards or not, sorry.
I am running a GTX1080 and a GTX980 side by side without issues. They work fine together in the Beta. They even work together in Vue when using the new pathtracer. As others mentioned if your scene uses less than 8Gb both cards will run, if it uses 8-12 Gb only your Titan will render and at more than 12 it's going to the CPU.
Ciao
TD
Thanks for the clarification SW.
Thanks Thd. Might be a silly question, but have you verified that both cards are actually in use. And have your render times reduced as a result?
Yes, I have. Using the Test scene from the bennchmark thread, the DS release version (GTX980 only) renders it in 179.6s, The DS Beta (uses both cards) does it in 94.4s. Both cards show as active in my GPU monitor app when using the BETA.
Ciao
TD
Awesome, well that answers a lot of questions. Cheers TD!
The cards run completely independent of each other - I have a 980ti and a 1080 and they both run at their maximum clock speeds.
@ original question
I only tried mixing an even older Kepler first generation Titan with two GTX 1080 and saw a speed improvment over just using two GTX 1080 as expected.
-> Based on very limited tests mixing different generation graphic cards in Iray seems to work with basic features.
Thanks to everyone posting above, it's good to see real world results.
If you're not in a hurry you might want to wait for CES2017 in January as rumors say the upcoming 1080 Ti might be released, which should be placed between the 1080 and Titan X, with 12GB and probably 3328 Cuda cores. Prices are expected to be less than $1000 and it wouldn't limit your Titan X to 8GB render speed using only the Titan X when rendering larger scenes >8GB then. But maybe it's only rumors .. ;-)
Oh what a good tip, cheers bud. I want to upgrade to a skylake chip and DDR4 too, so i'll probably do that first and see what the GPU landscape looks like in a few months. I'm still waiting on a stable pascal iray SDK to see decent numbers, fingers crossed it's not too far away.
I'm not quite clear about Iray and VRAM. I have a GTX970 with 4GB VRAM. If I were to buy a 1070 with 8GB and slot it in next to the 970, would Iray be limited to the smaller VRAM capacity of the 970 or would it disable the 970 but continue to use the 1070 if the scene was greater than 4GB?
In other words, would a scene <4GB use both cards but a scene >4GB use only the 1070 or would it default to CPU?
This said, the upcoming Kaby Lake desktop processors (K-series) are expected to officially debut on CES2017 too.. as the notebook versions are already available for some time and there have already i7-7700K samples been benchmarked and overclocked, these should be available to the market very soon then. They use same socket as Skylake and so run on older motherboards too, as long as manufacturer offers a bios update for new CPUs, though there will be new series of chipsets too (e.g. Z270). The fastest one (i7-7700K) will be clocked at 4.2GHz (4.5GHz turbo) compared to Skylake (i7-6700K) which is 4.0Ghz (4.2GHz turbo).
But then.. you can always wait for a next generation. Decisions, decisions.. ;-)
Good question, i second this.
Oh boy, my buying decisions have been heavily compromised in the last few hours. the 6700 was exactly my pick, which is a step up from the fx 8350 im using now. I imagine no matter what i buy, there will be something better right around the corner. Such is technology :(
See thd777's post, I think he's correct. I corrected my first post, as it might have added to the confusion. iray/DS uses all GPUs for rendering that have enough memory to load the full scene.
In your case <4GB both cards, 4-8GB only 1070, >8GB CPU only (though I don't know is if CPU adds to the GPU rendering on scenes <8Gb too ..?)
Correct on the way the cards work. As for adding in the cpu - probably not worth it if you have higher-end gpus. I have a stress-test scene that renders in 57 minutes on a factory-overclocked 980ti; if I add in my cpu (I7 3590, 3.5 GHz 6-core) the render time drops to 52 minutes and the cpu contributes something under 500 iterations to the 6,000+ total. And the cpu runs all threads at 100% for the entire 52 minutes.
It could just be me, but that sentence mashed my brain a bit. Are you saying that if the scene is over 4gb it would render only on the 1070. Whereas if the scene is below 4gb, it would use both of his cards? If that's the case, does that mean that iray is loading the exact same scene into the VRAM of both cards, and pairing their clock speed to converge the render?
Edit: I've just read the post you were answering to, and it makes more sense now, i'm just a silly moose. My question still stands though, for clarity.
Thanks. That clears up some of my confusion. I thought that if the scene exceeds the VRAM on the card with the least (in my example the 4GB 970) then Iray would default to CPU regardless of the 8GB 1070.
Yes, iray is loading the exact same scene into VRAM of both cards, that's why only those cards can be used that have enough memory. Not pairing the clock speed (at least I think so), but the number of Cuda cores it uses for rendering. Don't know about the technical details though, just from what I read myself.
Thanks for the information, learning something new every day ;-)
That clears a lot up, thanks Bad.
Hello
In regards to running maxwell and pascal cards please don't think this a silly question because i am new to this.
I am currently running a "980ti" maxwell i purchased a "titan x" pascal the week it was released i am using the GR
not the beta and was just wondering how are you running the cards with a SLI bridge or what or should i just wait
for a new release??