2016 Skylake/TB3 rMBP w/ eGPU vs. Mac Pro?

[alexdc]

Can someone smarter than me tell me the negatives of what this setup would be versus a nMP?

2016 Skylake-based Retina MacBook Pro w/ external GPU(s) vs. New Mac Pro

So it's generally agreed upon based on Intel announcements re: 1st Quarter 2016 that the 2016 Retina MacBook Pro will have a Skylake CPU and natively-supported Thunderbolt 3 using USB-C connectors.
Intel is officially stating that they are making a big push for external GPUs with TB3, and with its insane bandwidth it would definitely be sufficient for high end GPUs in SLI/Crossfire configurations.

If you could have the mobility of an already powerful rMBP, and an eGPU enclosure that could hotplug into it to beef it up in a desktop setting... why wouldn't you??? Wouldn't it theoretically perform nearly or equally well in video editing, visual effects, AutoCAD, etc. as a Mac Pro? I understand of course that at this point it's still all conjecture... but still!

(I put this in the Mac Pro thread because I was mainly aiming for a Mac Pro at next year's refresh until I found out about Intel's ambitious plans for TB3-enabled eGPUs. Arghhh!)

 

[MacVidCards]

Some interesting numbers coming

 

[koyoot]

http://www.techpowerup.com/reviews/AMD/R9_Fury_X_PCI-Express_Scaling/11.html You can check what affect will have reduced bandwidth of PCIex on performance of a GPU in certain circumstances. Not big a deal at least in gaming.

As I said before. For me, it all looks like we are approaching era when you have a computer, and a computational cluster of GPUs that you connect via Thunderbolt or any other solution that can be in future. Things like NVlink will help communicate between GPUs in that rack, releasing the bandwidth of the connection for feeding the data to the GPUs.

 

[alexdc]

http://www.techpowerup.com/reviews/AMD/R9_Fury_X_PCI-Express_Scaling/11.html You can check what affect will have reduced bandwidth of PCIex on performance of a GPU in certain circumstances. Not big a deal at least in gaming.

As I said before. For me, it all looks like we are approaching era when you have a computer, and a computational cluster of GPUs that you connect via Thunderbolt or any other solution that can be in future. Things like NVlink will help communicate between GPUs in that rack, releasing the bandwidth of the connection for feeding the data to the GPUs.

Exactly. Thanks for the link, very interesting and thorough test. I'm looking forward to the multi-GPU test he says is coming up next.

And honestly I think that era you describe is now upon us, with TB3's release in the next couple months.

 

[yfile]

Keep in mind that one fast GPU needs one fast CPU core (approximately). So if you want more than four GPUs you need powerfull Mac Pro. Four GPUs connected to Macbook Pro would be nice of course.