TB vs M.2 eGPUs in terms of scalability on Mac
Hi, I'm right before investing in external GPU on Mac since Metal render engines are showing up (Octane X & Redshift). My two biggest concerns (which I take in mind for future possibilities of expansion) are:
- Should I use 'proper' way of setting up eGPU via TB or should I make a modification inside my Mac to use faster M.2 slot?
- In terms of scalability, what's the best:
- multiple eGPU cases plugged into TB ports?
- multiple eGPUs + M.2 eGPU?
- multiple GPUs connected to PCIe splitter to M.2 (mining rigs?)
I take also into account hardware limitations (bandwidth, CPU) since my late 2014 iMac has only two TB2 ports. But I'm really interested after seeing @itsage builds with M.2 eGPUs which are kind of last resort for non TB3 users that want maxed out performance. As I said before I'll mainly use external GPUs for 3D rendering so compute/time factor is hugely important. Why not building a PC workstation? Simply because I want to utilize as much as I can from current setup + Metal apps are just arriving on Apple systems. : - )
If anybody has experience with multiple eGPUs on Mac, please share your thoughts.
Some posts that are around this topic, very useful and informative:
PCIe splitter used in Mac Pro 5.1 connecting 4 GPUs:
M.2 vs TB eGPU benchmarks:
Pls see the build in my sig for the most comprehensive analysis on here of M.2 versus TB3 performance.
From my experience with eGPU rendering, this limited bandwidth is not as crucial for raytracing rendering as it would be for realtime applications and I'm not sure you would notice a difference in performance. The rendering process happening on the GPU can take place largely independently from the CPU and it only needs to update the image buffer with a certain (non-realtime) interval. The speed at which this image buffer update happens shouldn't matter much either, as a well-written rendering engine normally doesn't pause its raytracing for this process.
@nando4 @oscar-j @itsage - thank you for your replies. I agree with you, considering that this setup will be utilized only for 3D rendering, the bottlenecks are not so much important here + the 'proper' way of utilizing TB eGPU seems to be okay : - )
One more question about choosing the right GPU - Does utilizing newest card such as Big Navi or Navi for TB2 Mac (late 2014 iMac, i5 4690) affect lower render speed/time? Will I get lower Benchmark results from OctaneBench, Luxmark or any other OpenCL/Metal Benchmark app if I'd connect those newest card to TB2 instead of TB3 Macs?