3x Titan Xs vs 3x 980s: How much will I miss?

[Scott C]

In practical terms for someone primarily cutting in Premiere CC, some light After Effects work, and Resolve, shooting on Dragon, how much will I lose by going with a tri-980 build vs tri-Titan Xs? For company/vendor reasons, let's just assume those are the only two choices here. In a set budget I'm questioning the value of the faster GPU build vs buying 1-2 larger/better monitors. I'm not working in super-quick turnaround/broadcast type space here, so if it's the difference between a 20 minute render and a 25 minute render I'm really not sweating it.

Thoughts?

 

[Eric Bowen]

Well Premiere doesnt scale out over more than 1 GPU for rendering as much as Davinci or Octane. So a single Titan X card would be better for both realtime playback and rendering than 3 980 GTX cards. Realtime is still limited to 1 GPU as far as I know. Davinci would have better performance with 2 Titan X cards versus 3 980 GTX cards. I would suggest less Titan X cards or 980Ti cards than 3 980 GTX.

 

[Scott C]

Well Premiere doesnt scale out over more than 1 GPU for rendering as much as Davinci or Octane. So a single Titan X card would be better for both realtime playback and rendering than 3 980 GTX cards. Realtime is still limited to 1 GPU as far as I know. Davinci would have better performance with 2 Titan X cards versus 3 980 GTX cards. I would suggest less Titan X cards or 980Ti cards than 3 980 GTX.

Thanks Eric. I guess then my question is, how big is the gap between them? If I'm working in a 6K/23.976 timeline, is that the difference between 1/2 res and full res playback? Would for instance a motion tracker in After Effects take 2x longer? Is it the difference between realtime and choppy Resolve performance? If it helps, the processor is a 5960X, and the system has 32GB of ram (assuming again here these are not changeable. Although I'm hoping to add ram after the fact).

 

[Eric Bowen]

AE doesnt have GPU acceleration so there wont be any impact on AE. The Titan X card could absolutely be the difference between 6K playback at full or half resolution depending on the complexity of the timeline. The greater the FX load or GPU acceleration elements on the timeline the higher load put on the GPU during realtime playback. Rendering would just slow down by a percentage which probably wont effect you much. GPU accelerated Debayering is really what you have to consider since that puts a massive load on the GPU especially Vram. Any R3D 4K Res plus will put enough load on the GPU to justify a Titan X over a 980 GTX due to the limitation of 1 GPU for realtime playback. Davinci will scale for that but that add's latency which means less ability to stack FX or other complexity to the timeline and play it back at full res or even half res with R3D so yes it does mean the difference between choppy or realtime playback depending on the complexity of the timeline. Consider this to be a performance load ceiling question. The 980 GTX will will have a far lower point of realtime capability for playback with R3D than the Titan X. If your editing already pushes the system 80% + then I would not go with the lower end card for 6K media. Also keep in mind 32GB of ram is not really enough to make the most efficient use of 3x GPU's since buffers have to be duplicated so much with GPU acceleration. 1 GPU even with 12GB vram would perform better with that amount of system ram.

 

[Scott C]

AE doesnt have GPU acceleration so there wont be any impact on AE. The Titan X card could absolutely be the difference between 6K playback at full or half resolution depending on the complexity of the timeline. The greater the FX load or GPU acceleration elements on the timeline the higher load put on the GPU during realtime playback. Rendering would just slow down by a percentage which probably wont effect you much. GPU accelerated Debayering is really what you have to consider since that puts a massive load on the GPU especially Vram. Any R3D 4K Res plus will put enough load on the GPU to justify a Titan X over a 980 GTX due to the limitation of 1 GPU for realtime playback. Davinci will scale for that but that add's latency which means less ability to stack FX or other complexity to the timeline and play it back at full res or even half res with R3D so yes it does mean the difference between choppy or realtime playback depending on the complexity of the timeline. Consider this to be a performance load ceiling question. The 980 GTX will will have a far lower point of realtime capability for playback with R3D than the Titan X. If your editing already pushes the system 80% + then I would not go with the lower end card for 6K media. Also keep in mind 32GB of ram is not really enough to make the most efficient use of 3x GPU's since buffers have to be duplicated so much with GPU acceleration. 1 GPU even with 12GB vram would perform better with that amount of system ram.

Thanks Eric. I've found another option where I may be able to do a single or dual Titan X instead of 3 980s. My next question would be how big of a gap would you anticipate in Resolve between a single Titan vs 2 of them?

 

[Eric Bowen]

Resolve has a significant performance difference between 1 and 2 GPU's for realtime playback. I suspect it has to do with what memory management profiles utilized depending on 1 or 2 GPU's and what vram is buffered for drawing out on the screen. A single Titan X would be the 1 card that could handle realtime playback if a single card could with Resolve for the frame rate of media you want to use. However you have a much better chance for realtime playback at the frame rate desired with 2 cards even 2x Titan X cards with 6K media. Keep in mind the higher the resolution of media with GPU accelerated debayering the exponential increase in vram utilized per frame. That means higher frame rates will require more buffering to maintain playback realtime without dropping frames.

 

[David Moffitt]

As someone who has dual Titan Xs, I can hardly imagine a situation that needs 3. You have to have ONE HECK of a CPU and RAM setup to not have that be the bottleneck...