Great news for Mac users!!! And a ? for Davinci experts.

So I already have one of the latest 27" iMacs, (Non PRo), but fully spec'd, including the 580.

Davinci already chews through everything I give it with this machine, but now with full eGPU support from Apple you can use multiple GPUs (Studio) version required, and tackle all of your cutting and coloring demands on a low-end Mac laptop.

Watch the video all the way through... we are working in exciting times!!
Apple's eGPU support is still in it's infancy, so I can only see this getting better and better.


Question for the Davinci pros here:
When using multiple GPUs it's favored to match cards or you will only get the benefits from the lowest rated card. Am I correct on that assumption?
So should you try to match eGPU cards for the very same reason?

Thanks, Tim.

Hi.

I am sorry Tim.

You use 10 min to prove that a Graphics Cards in an external exposure a eGPU, is faster than the built in Intel IRIS Plus Graphics 640.

Nobody expect you, will try to run Resolve on a Intel IRIS Plus Graphics 640.

And then you choose to use the Old Resolve 14. I wonder if it can be an old YouTube you use?

Regards Carsten.

you will only get the benefits from the lowest rated card. Am I correct on that assumption?
So should you try to match eGPU cards for the very same reason?

That is the case the way the memory mapping works for threading media onto multiple GPUs.

Memory is not cumulative -- IOW you cannot build a skyscraper -- but you can build a mini-mall. If you gang 4 6GB Titans, you do not achieve a 24 GB VRAM resource... but you can thread more fps of processing through that many more "lanes." No, you don't get more lanes than the slot offers, but in the case of nVidia, it's that many more CUDA cores that the application sees; but it is vital that the GPUs themselves are more or less identical.

jPo, CSI

Yeah, what Joe says. eGPU is not a total solution yet. I don't think you're going to get real-time playback from 4K files on a 4K monitor on a laptop. I think external GPUs are a real thing and will work eventually, and there are already legit PCIe expanders for GPUs that are used everyday in very heavy-duty Resolve systems. But this is impossible to do as a small, inexpensive, portable system.

Technically you can use a internal PCI, when you remove the WiFi card, but you have to open the housing of you laptop an cut a hole in it, to get the cable out - not for the faint of heart

Technically you can use a internal PCI, when you remove the WiFi card, ...

sure, this kind of homegrown solutions are always possible. but you wouldn't get very satisfaying support for video work in this case, because WLAN Mini PCI-Express modules are usually only connected like weak ordinary PCIe x1 cards.

sure, you can connect even quite expensive nvidia GPUs via raiser adapters to this kind of inferiour interface. that's more a less the common practice in blockchain mining farms. but that's a very specialized usage scenario. at blockchain mining you don't have much I/O traffic. you just upload a block of data to the card and then let it do the much more time consuming cryptographic number crunching in an independent manner. but utilizing GPUs for video processing is bound to quite different requirements. in this case you always have to transfer a lot of data between the card and the main system. it's not as I/O demanding as other kinds of high performance computation, which really may use and benefit from special GPU connection and switching technicologies, like NVLink Fabric. but this kind of GPU interconnection possibilities isn't available in GTX cards anyway. that's one of the reasons, why resolve doesn't scale exceptionally well on adding additional GPUs. one powerful single card in an adequate connected slot is often the much more sufficient and economic solution in this particular case.

FWIW so far in my experiments with eGPU and Resolve (and R3D footage), I have not had much luck getting a big performance boost with dual AMD GPUs (10.13.4, Resolve 14.3, both assigned and using OpenCL in prefs, 2016-2017 15" MBP, 8100MB RAM allocated, one eGPU per controller). Curious to know if others have had a different experience on a similar setup.

It is possible only certain features support dual eGPU (for now), but so far one AMD (OpenCL) eGPU seems to drive a similar amount of performance as two. Hopefully BMD will provide some detailed guidance and/or documentation on:

• Confirmation that a connected eGPU does not have to be driving a screen in order to provide acceleration.

• Which features benefit from two or more AMD eGPU.

• Which of those same features also benefit from OpenCL to the same degree as CUDA
I'm using CL to start because Metal is not an option for debayering, although CL only provides a modest boost over Quad core CPU debayering speeds from what I've seen. Possibly it's a RAM bottleneck but no matter what I do the app will not assign more than 8100MB out of a theoretical total of 14GB or so (if you assume the OS will always take a couple for itself).

• Which features (if any) are CUDA-only... I don't believe any are but can't hurt to ask.