- Moderator
- #21
Jeff Kilgroe
Well-known member
Are you working at 120fps? That's a big deal if you are, that's the same as four of those 4KHD streams playing back at 30fps. If you shot it at 120fps, that's irrelevant, and we would then want to know what frame rate you're actually editing, playing back at. If this is 4KHD at 24fps or 30fps, then yes, you have a bottleneck. If it's playing back at 120fps, then I'd say it's doing well...
SLI is neither here nor there as it's not applicable in this situation. SLI is the multiplexing of video outputs between GPUs to create a single frame. In short, it's a gaming thing and doesn't apply to GPU multi-processing via CUDA, OpenCL, OpenGL, etc.. SLI is not used by pro video apps and such. Multiple GPU processing via CUDA/ OpenCL is supported on OS X as it is on Windows. Applications like Resolve, when not using the Lite version, will use multiple GPUs. Premiere only uses a single GPU at this time for playback and editing, but will use multiple for render output. This will hopefully change in the future, but that's still the current state of things. RC-X utilizes multiple GPUs, however, it seems that one decent GPU is usually more than sufficient as the GPU is rarely the bottleneck these days for the limited scope of working with a single clip at a time.
Provided everything is configured properly, you shouldn't see a huge difference between OS X and Windows performance. Windows will have the edge for GPU functionality as the drivers are iterated faster and better optimized to some extent, although nVidia does do a good job at keeping the Mac codebase up to date when they do make updates available. I do suspect your storage to be a contributing factor here. Although, there could be other bottlenecks. Shiblon brings up a good point about PCIe bandwidth. Restricting the Titan X to 8 lanes is probably not going to hamper its performance in this case. However, the PCIe situation could be more dire than that. Often times PCIe slot bandwidth can be shared between slots. If the system is loaded up, your GPU could be sharing 8 lanes with another PCIe card. That can definitely make an impact.
Definitely. You will gain quite a bit of additional memory bandwidth and of course the additional CPU cores and bandwidth with the newer CPU. More PCIe lanes too. The 4770K only has 16 PCIe 3.0 lanes to offer up. So if you have *ANY* PCIe card in your system in addition to the Titan X, you have already compromised PCIe bandwidth to your GPU. And as I said above about more dire situations, you can be spread very thin in a hurry. Additionally, PCIe v2.0 cards can become problematic in this situation as they can often force linked slots into 2.0 mode and if that happens to be your GPU, you've just cut its per-lane throughput by almost half.
I think this is a good idea -- to test with the other Windows installation. If it performs much better, than that would be a good indication of a problem with your OS X installation. If it's similar, then you have hardware bottlenecks to chase down.
I think a second Titan X in this system would be a waste of time, given the available PCIe bandwidth. You don't need the SLI bridge installed unless you want to use SLI in games that support it. Once again, that has no bearing on GPU acceleration or computation in Windows or elsewhere.
